Search results for: parallel algorithms

Authors: Ani Demetrashvili

Abstract:

As online learning continues to reshape the landscape of education, questions arise regarding its efficacy for diverse learning styles, particularly for visual learners. This abstract delves into the impact of online learning on visual learners, exploring how digital mediums influence their educational experience and how educational platforms can be optimized to cater to their needs. Visual learners comprise a significant portion of the student population, characterized by their preference for visual aids such as diagrams, charts, and videos to comprehend and retain information. Traditional classroom settings often struggle to accommodate these learners adequately, relying heavily on auditory and written forms of instruction. The advent of online learning presents both opportunities and challenges in addressing the needs of visual learners. Online learning platforms offer a plethora of multimedia resources, including interactive simulations, virtual labs, and video lectures, which align closely with the preferences of visual learners. These platforms have the potential to enhance engagement, comprehension, and retention by presenting information in visually stimulating formats. However, the effectiveness of online learning for visual learners hinges on various factors, including the design of learning materials, user interface, and instructional strategies. Research into the impact of online learning on visual learners encompasses a multidisciplinary approach, drawing from fields such as cognitive psychology, education, and human-computer interaction. Studies employ qualitative and quantitative methods to assess visual learners' preferences, cognitive processes, and learning outcomes in online environments. Surveys, interviews, and observational studies provide insights into learners' preferences for specific types of multimedia content and interactive features. Cognitive tasks, such as memory recall and concept mapping, shed light on the cognitive mechanisms underlying learning in digital settings. Eye-tracking studies offer valuable data on attentional patterns and information processing during online learning activities. The findings from research on the impact of online learning on visual learners have significant implications for educational practice and technology design. Educators and instructional designers can use insights from this research to create more engaging and effective learning materials for visual learners. Strategies such as incorporating visual cues, providing interactive activities, and scaffolding complex concepts with multimedia resources can enhance the learning experience for visual learners in online environments. Moreover, online learning platforms can leverage the findings to improve their user interface and features, making them more accessible and inclusive for visual learners. Customization options, adaptive learning algorithms, and personalized recommendations based on learners' preferences and performance can enhance the usability and effectiveness of online platforms for visual learners.

Keywords: online learning, visual learners, digital education, technology in learning

Procedia PDF Downloads 33

Authors: Alexandru-Ion Marinescu

Abstract:

There exist a plethora of methods in the scientific literature which tackle the well-established task of credit score evaluation. In its most abstract form, a credit scoring algorithm takes as input several credit applicant properties, such as age, marital status, employment status, loan duration, etc. and must output a binary response variable (i.e. “GOOD” or “BAD”) stating whether the client is susceptible to payment return delays. Data imbalance is a common occurrence among financial institution databases, with the majority being classified as “GOOD” clients (clients that respect the loan return calendar) alongside a small percentage of “BAD” clients. But it is the “BAD” clients we are interested in since accurately predicting their behavior is crucial in preventing unwanted loss for loan providers. We add to this whole context the constraint that the algorithm must yield an actual, tractable mathematical formula, which is friendlier towards financial analysts. To this end, we have turned to genetic algorithms and genetic programming, aiming to evolve actual mathematical expressions using specially tailored mutation and crossover operators. As far as data representation is concerned, we employ a very flexible mechanism – LINQ expression trees, readily available in the C# programming language, enabling us to construct executable pieces of code at runtime. As the title implies, they model trees, with intermediate nodes being operators (addition, subtraction, multiplication, division) or mathematical functions (sin, cos, abs, round, etc.) and leaf nodes storing either constants or variables. There is a one-to-one correspondence between the client properties and the formula variables. The mutation and crossover operators work on a flattened version of the tree, obtained via a pre-order traversal. A consequence of our chosen technique is that we can identify and discard client properties which do not take part in the final score evaluation, effectively acting as a dimensionality reduction scheme. We compare ourselves with state of the art approaches, such as support vector machines, Bayesian networks, and extreme learning machines, to name a few. The data sets we benchmark against amount to a total of 8, of which we mention the well-known Australian credit and German credit data sets, and the performance indicators are the following: percentage correctly classified, area under curve, partial Gini index, H-measure, Brier score and Kolmogorov-Smirnov statistic, respectively. Finally, we obtain encouraging results, which, although placing us in the lower half of the hierarchy, drive us to further refine the algorithm.

Keywords: expression trees, financial credit scoring, genetic algorithm, genetic programming, symbolic evolution

Procedia PDF Downloads 113

Authors: Juan S. Mantilla, Peter Thomson

Abstract:

Seismic events may result in discomfort on occupants of the buildings, structural damage or even buildings collapse. Traditional design aims to reduce dynamic response of structures by increasing stiffness, thus increasing the construction costs and the design forces. Structural control systems arise as an alternative to reduce these dynamic responses. A commonly used control systems in buildings are the passive friction dampers, which adds energy dissipation through damping mechanisms induced by sliding friction between their surfaces. Passive friction dampers are usually implemented on the diagonal of braced buildings, but such devices have the disadvantage that are optimal for a range of sliding force and out of that range its efficiency decreases. The above implies that each passive friction damper is designed, built and commercialized for a specific sliding/clamping force, in which the damper shift from a locked state to a slip state, where dissipates energy through friction. The risk of having a variation in the efficiency of the device according to the sliding force is that the dynamic properties of the building can change as result of many factor, even damage caused by a seismic event. In this case the expected forces in the building can change and thus considerably reduce the efficiency of the damper (that is designed for a specific sliding force). It is also evident than when a seismic event occurs the forces in each floor varies in the time what means that the damper's efficiency is not the best at all times. Semi-Active Friction devices adapt its sliding force trying to maintain its motion in the slipping phase as much as possible, because of this, the effectiveness of the device depends on the control strategy used. This paper deals with the development and performance evaluation of a low cost Semiactive Variable Friction Damper (SAVFD) in reduced scale to reduce vibrations of structures subject to earthquakes. The SAVFD consist in a (1) hydraulic brake adapted to (2) a servomotor which is controlled with an (3) Arduino board and acquires accelerations or displacement from (4) sensors in the immediately upper and lower floors and a (5) power supply that can be a pair of common batteries. A test structure, based on a Benchmark structure for structural control, was design and constructed. The SAVFD and the structure are experimentally characterized. A numerical model of the structure and the SAVFD is developed based on the dynamic characterization. Decentralized control algorithms were modeled and later tested experimentally using shaking table test using earthquake and frequency chirp signals. The controlled structure with the SAVFD achieved reductions greater than 80% in relative displacements and accelerations in comparison to the uncontrolled structure.

Keywords: earthquake response, friction damper, semiactive control, shaking table

Procedia PDF Downloads 377

Authors: Monique M. Keller, Roline Barnes, Corlia Brandt

Abstract:

Introduction Scientific evidence on the hand forces and the types of grasps measurement during daily tasks are lacking, leaving a gap in the field of hand rehabilitation and robotics. Measuring the grasp forces and types produced by the individual fingers during daily functional tasks is valuable to inform and grade rehabilitation practices for second to fifth metacarpal fractures with robust scientific evidence. Feix et al, 2016 identified the most extensive and complete grasp study that resulted in the GRASP taxonomy. Covid-19 virus changed data collection across the globe and safety precautions in research are essential to ensure the health of participants and researchers. Methodology A cross-sectional study investigated six healthy adults aged 20 to 59 years, pilot participants’ hand forces during 105 tasks. The tasks were categorized into five sections namely, personal care, transport and moving around, home environment and inside, gardening and outside, and office. The predominant grasp of each task was identified guided by the GRASP Taxonomy. Grasp forces were measured with 13mm force-sensing resistors glued onto a glove attached to each of the dominant and non-dominant hand’s individual fingers. Testing equipment included Flexiforce 13millimetres FSR .5" circle, calibrated prior to testing, 10k 1/4w resistors, Arduino pro mini 5.0v – compatible, Esp-01-kit, Arduino uno r3 – compatible board, USB ab cable - 1m, Ftdi ft232 mini USB to serial, Sil 40 inline connectors, ribbon cable combo male header pins, female to female, male to female, two gloves, glue to attach the FSR to glove, Arduino software programme downloaded on a laptop. Grip strength measurements with Jamar dynamometer prior to testing and after every 25 daily tasks were taken to will avoid fatigue and ensure reliability in testing. Covid-19 precautions included wearing face masks at all times, screening questionnaires, temperatures taken, wearing surgical gloves before putting on the testing gloves 1.5 metres long wires attaching the FSR to the Arduino to maintain social distance. Findings Predominant grasps observed during 105 tasks included, adducted thumb (17), lateral tripod (10), prismatic three fingers (12), small diameter (9), prismatic two fingers (9), medium wrap (7), fixed hook (5), sphere four fingers (4), palmar (4), parallel extension (4), index finger extension (3), distal (3), power sphere (2), tripod (2), quadpod (2), prismatic four fingers (2), lateral (2), large-diameter (2), ventral (2), precision sphere (1), palmar pinch (1), light tool (1), inferior pincher (1), and writing tripod (1). Range of forces applied per category, personal care (1-25N), transport and moving around (1-9 N), home environment and inside (1-41N), gardening and outside (1-26.5N), and office (1-20N). Conclusion Scientifically measurements of finger forces with careful consideration to types of grasps used in daily tasks should guide rehabilitation practices and robotic design to ensure a return to the full participation of the individual into the community.

Keywords: activities of daily living (ADL), Covid-19, force-sensing resistors, grasps, hand forces

Procedia PDF Downloads 186

Authors: K. H. Reeta, Bhavana Prasher, Mitali Mukerji, Dhwani Dholakia, Sangeeta Khanna, Archana Vats, Shivam Pandey, Sandeep Seth, Subir Kumar Maulik

Abstract:

Introduction Research has demonstrated a connection between coronary artery disease (CAD) and genetics. We did a deep literature mining using both bioinformatics and manual efforts to identify the susceptible polymorphisms in coronary artery disease. Further, the study sought to validate these findings in an Asian population. Methodology In first phase, we used an automated pipeline which organizes and presents structured information on SNPs, Population and Diseases. The information was obtained by applying Natural Language Processing (NLP) techniques to approximately 28 million PubMed abstracts. To accomplish this, we utilized Python scripts to extract and curate disease-related data, filter out false positives, and categorize them into 24 hierarchical groups using named Entity Recognition (NER) algorithms. From the extensive research conducted, a total of 466 unique PubMed Identifiers (PMIDs) and 694 Single Nucleotide Polymorphisms (SNPs) related to coronary artery disease (CAD) were identified. To refine the selection process, a thorough manual examination of all the studies was carried out. Specifically, SNPs that demonstrated susceptibility to CAD and exhibited a positive Odds Ratio (OR) were selected, and a final pool of 324 SNPs was compiled. The next phase involved validating the identified SNPs in DNA samples of 96 CAD patients and 37 healthy controls from Indian population using Global Screening Array. ResultsThe results exhibited out of 324, only 108 SNPs were expressed, further 4 SNPs showed significant difference of minor allele frequency in cases and controls. These were rs187238 of IL-18 gene, rs731236 of VDR gene, rs11556218 of IL16 gene and rs5882 of CETP gene. Prior researches have reported association of these SNPs with various pathways like endothelial damage, susceptibility of vitamin D receptor (VDR) polymorphisms, and reduction of HDL-cholesterol levels, ultimately leading to the development of CAD. Among these, only rs731236 had been studied in Indian population and that too in diabetes and vitamin D deficiency. For the first time, these SNPs were reported to be associated with CAD in Indian population. Conclusion: This pool of 324 SNP s is a unique kind of resource that can help to uncover risk associations in CAD. Here, we validated in Indian population. Further, validation in different populations may offer valuable insights and contribute to the development of a screening tool and may help in enabling the implementation of primary prevention strategies targeted at the vulnerable population.

Keywords: coronary artery disease, single nucleotide polymorphism, susceptible SNP, bioinformatics

Procedia PDF Downloads 73

Authors: Klaus Peter Scherer, Daniel Knöll, Constantin Rieder

Abstract:

Problem and Purpose: Intelligent systems are available and helpful to support the human being decision process, especially when complex surgical eye interventions are necessary and must be performed. Normally, such a decision support system consists of a knowledge-based module, which is responsible for the real assistance power, given by an explanation and logical reasoning processes. The interview based acquisition and generation of the complex knowledge itself is very crucial, because there are different correlations between the complex parameters. So, in this project (semi)automated self-learning methods are researched and developed for an enhancement of the quality of such a decision support system. Methods: For ophthalmic data sets of real patients in a hospital, advanced data mining procedures seem to be very helpful. Especially subgroup analysis methods are developed, extended and used to analyze and find out the correlations and conditional dependencies between the structured patient data. After finding causal dependencies, a ranking must be performed for the generation of rule-based representations. For this, anonymous patient data are transformed into a special machine language format. The imported data are used as input for algorithms of conditioned probability methods to calculate the parameter distributions concerning a special given goal parameter. Results: In the field of knowledge discovery advanced methods and applications could be performed to produce operation and patient related correlations. So, new knowledge was generated by finding causal relations between the operational equipment, the medical instances and patient specific history by a dependency ranking process. After transformation in association rules logically based representations were available for the clinical experts to evaluate the new knowledge. The structured data sets take account of about 80 parameters as special characteristic features per patient. For different extended patient groups (100, 300, 500), as well one target value as well multi-target values were set for the subgroup analysis. So the newly generated hypotheses could be interpreted regarding the dependency or independency of patient number. Conclusions: The aim and the advantage of such a semi-automatically self-learning process are the extensions of the knowledge base by finding new parameter correlations. The discovered knowledge is transformed into association rules and serves as rule-based representation of the knowledge in the knowledge base. Even more, than one goal parameter of interest can be considered by the semi-automated learning process. With ranking procedures, the most strong premises and also conjunctive associated conditions can be found to conclude the interested goal parameter. So the knowledge, hidden in structured tables or lists can be extracted as rule-based representation. This is a real assistance power for the communication with the clinical experts.

Keywords: an expert system, knowledge-based support, ophthalmic decision support, self-learning methods

Procedia PDF Downloads 252

Authors: Sithandazile Masuku

Abstract:

Introduction: Public health approaches to offender mental health are driven by international policies and frameworks in response to the disproportionately large representation of people with mental health problems within the offender pathway compared to the general population. Public health service innovations include mental health courts in the US, restorative models in Singapore and, liaison and diversion services in Australia, the UK, and some other European countries. Mental health nurses are at the forefront of offender health service innovations. In the U.K. context, police custody has been identified as an early point within the offender pathway where nurses can improve outcomes by offering assessments and share information with criminal justice partners. This scope of nursing practice has introduced challenges related to skills and support required for nurses working at the interface of health and the criminal justice system. Parallel literature exploring experiences of nurses working in forensic settings suggests the presence of compassion fatigue, burnout and vicarious trauma that may impede risk harm to the nurses in these settings. Published research explores mainly service-level outcomes including monitoring of figures indicative of a reduction in offending behavior. There is minimal research exploring the experiences of liaison and diversion nurses who are situated away from a supportive clinical environment and engaged in complex autonomous decision-making. Aim: This paper will share qualitative findings (in progress) from a PhD study that aims to explore the experiences of liaison and diversion nurses in one service in the U.K. Methodology: This is a qualitative interview study conducted using an Interpretative Phenomenological Analysis to gain an in-depth analysis of lived experiences. Methods: A purposive sampling technique was used to recruit n=8 mental health nurses registered with the UK professional body, Nursing and Midwifery Council, from one UK Liaison and Diversion service. All participants were interviewed online via video call using semi-structured interview topic guide. Data were recorded and transcribed verbatim. Data were analysed using the seven steps of the Interpretative Phenomenological Analysis data analysis method. Emerging Findings Analysis to date has identified pertinent themes: • Difficulties of meaning-making for nurses because of the complexity of their boundary spanning role. • Emotional burden experienced in a highly emotive and fast-changing environment. • Stress and difficulties with role identity impacting on individual nurses’ ability to be resilient. • Challenges to wellbeing related to a sense of isolation when making complex decisions. Conclusion Emerging findings have highlighted the lived experiences of nurses working in liaison and diversion as challenging. The nature of the custody environment has an impact on role identity and decision making. Nurses left feeling isolated and unsupported are less resilient and may go on to experience compassion fatigue. The findings from this study thus far point to a need to connect nurses working in these boundary spanning roles with a supportive infrastructure where the complexity of their role is acknowledged, and they can be connected with a health agenda. In doing this, the nurses would be protected from harm and the likelihood of sustained positive outcomes for service users is optimised.

Keywords: liaison and diversion, nurse experiences, offender health, staff wellbeing

Procedia PDF Downloads 129

Authors: Manaranjan Pradhan, Shailaja Grover, U. Dinesh Kumar

Abstract:

Farming community in India, as well as other parts of the world, is one of the highly stressed communities due to reasons such as increasing input costs (cost of seeds, fertilizers, pesticide), droughts, reduced revenue leading to farmer suicides. Lack of integrated farm advisory system in India adds to the farmers problems. Farmers need right information during the early stages of crop’s lifecycle to prevent damage and loss in revenue. In this paper, we use deep learning techniques to develop an early warning system for detection of crop diseases using images taken by farmers using their smart phone. The research work leads to building a smart assistant using analytics and big data which could help the farmers with early diagnosis of the crop diseases and corrective actions. The classical approach for crop disease management has been to identify diseases at crop level. Recently, ImageNet Classification using the convolutional neural network (CNN) has been successfully used to identify diseases at individual plant level. Our model uses convolution filters, max pooling, dense layers and dropouts (to avoid overfitting). The models are built for binary classification (healthy or not healthy) and multi class classification (identifying which disease). Transfer learning is used to modify the weights of parameters learnt through ImageNet dataset and apply them on crop diseases, which reduces number of epochs to learn. One shot learning is used to learn from very few images, while data augmentation techniques are used to improve accuracy with images taken from farms by using techniques such as rotation, zoom, shift and blurred images. Models built using combination of these techniques are more robust for deploying in the real world. Our model is validated using tomato crop. In India, tomato is affected by 10 different diseases. Our model achieves an accuracy of more than 95% in correctly classifying the diseases. The main contribution of our research is to create a personal assistant for farmers for managing plant disease, although the model was validated using tomato crop, it can be easily extended to other crops. The advancement of technology in computing and availability of large data has made possible the success of deep learning applications in computer vision, natural language processing, image recognition, etc. With these robust models and huge smartphone penetration, feasibility of implementation of these models is high resulting in timely advise to the farmers and thus increasing the farmers' income and reducing the input costs.

Keywords: analytics in agriculture, CNN, crop disease detection, data augmentation, image recognition, one shot learning, transfer learning

Procedia PDF Downloads 115

Authors: E. K. Tripsanas, D. Spanos, I. Oikonomopoulos, K. Stathopoulou, A. S. Abdelsamad, A. Pagoulatos

Abstract:

Patraikos Gulf is located offshore western Greece, and it is limited to the west by the Zante, Cephalonia, and Lefkas islands. The Plio/Pleistocene sequence is characterized by two depocenters, the east and west Patraikos basins separated from each other by a prominent sill. This study is based on the Plio/Pleistocene seismic stratigraphy analysis of a newly acquired 3D PSDM (Pre-Stack depth migration) seismic survey in the west Patraikos Basin and few 2D seismic profiles throughout the entire Patraikos Gulf. The eastern Patraikos Basin, although completely buried today with water depths less than 100 m, it was a deep basin during Pliocene ( > 2 km of Pliocene-Pleistocene sediments) and appears to have gathered most of Achelous River discharges. The west Patraikos Gulf was shallower ( < 1300 m of Pliocene-Pleistocene sediments) and characterized by a hummocky relief due to thrust-belt tectonics and Miocene to Pleistocene halokinetic processes. The transition from Pliocene to Miocene is expressed by a widespread erosional unconformity with evidence of fluvial drainage patterns. This indicates that west Patraikos Basin was aerially exposed during the Messinian Salinity Crisis. Continuous to semi-continuous, parallel reflections in the lower, early- to mid-Pliocene seismic packet provides evidence that the re-connection of the Mediterranean Sea with the Atlantic Ocean during Zanclean resulted in the flooding of the west Patraikos basin and the domination of hemipelagic sedimentation interrupted by occasional gravity flows. This is evident in amplitude and semblance horizon slices, which clearly show the presence of long-running, meandering submarine channels sourced from the southeast (northwest Peloponnese) and north. The long-running nature of the submarine channels suggests mobile efficient turbidity currents, probably due to the participation of a sufficient amount of clay minerals in their suspended load. The upper seismic section in the study area mainly consists of several successions of clinoforms, interpreted as progradational delta complexes of Achelous River. This sudden change from marine to shallow marine sedimentary processes is attributed to climatic changes and eustatic perturbations since late Pliocene onwards (~ 2.6 Ma) and/or a switch of Achelous River from the east Patraikos Basin to the west Patraikos Basin. The deltaic seismic unit consists of four delta complexes. The first two complexes result in the infill of topographic depressions and smoothing of an initial hummocky bathymetry. The distribution of the upper two delta complexes is controlled by compensational stacking. Amplitude and semblance horizon slices depict the development of several almost straight and short (a few km long) distributary submarine channels at the delta slopes and proximal prodeltaic plains with lobate sand-sheet deposits at their mouths. Such channels are interpreted to result from low-efficiency turbidity currents with low content in clay minerals. Such a differentiation in the nature of the gravity flows is attributed to the switch of the sediment supply from clay-rich sediments derived from the draining of flysch formations of the Ionian and Gavrovo zones, to the draining of poor in clay minerals carbonate formations of Gavrovo zone through the Achelous River.

Keywords: sequence stratigraphy, basin analysis, river deltas, submarine channels

Procedia PDF Downloads 316

Authors: Juan Sebastian Huertas Piedrahita, Jaime Arturo Lopez Duque, Eduardo Luis Perez Londoño, Julián S. Rodríguez

Abstract:

The process of generating computer vision is called artificial vision. The artificial vision is a branch of artificial intelligence that allows the obtaining, processing, and analysis of any type of information especially the ones obtained through digital images. Actually the artificial vision is used in manufacturing areas for quality control and production, as these processes can be realized through counting algorithms, positioning, and recognition of objects that can be measured by a single camera (or more). On the other hand, the companies use assembly lines formed by conveyor systems with actuators on them for moving pieces from one location to another in their production. These devices must be previously programmed for their good performance and must have a programmed logic routine. Nowadays the production is the main target of every industry, quality, and the fast elaboration of the different stages and processes in the chain of production of any product or service being offered. The principal base of this project is to program a computer that recognizes geometric figures (circle, square, and triangle) through a camera, each one with a different color and link it with a group of conveyor systems to organize the mentioned figures in cubicles, which differ from one another also by having different colors. This project bases on artificial vision, therefore the methodology needed to develop this project must be strict, this one is detailed below: 1. Methodology: 1.1 The software used in this project is QT Creator which is linked with Open CV libraries. Together, these tools perform to realize the respective program to identify colors and forms directly from the camera to the computer. 1.2 Imagery acquisition: To start using the libraries of Open CV is necessary to acquire images, which can be captured by a computer’s web camera or a different specialized camera. 1.3 The recognition of RGB colors is realized by code, crossing the matrices of the captured images and comparing pixels, identifying the primary colors which are red, green, and blue. 1.4 To detect forms it is necessary to realize the segmentation of the images, so the first step is converting the image from RGB to grayscale, to work with the dark tones of the image, then the image is binarized which means having the figure of the image in a white tone with a black background. Finally, we find the contours of the figure in the image to detect the quantity of edges to identify which figure it is. 1.5 After the color and figure have been identified, the program links with the conveyor systems, which through the actuators will classify the figures in their respective cubicles. Conclusions: The Open CV library is a useful tool for projects in which an interface between a computer and the environment is required since the camera obtains external characteristics and realizes any process. With the program for this project any type of assembly line can be optimized because images from the environment can be obtained and the process would be more accurate.

Keywords: artificial intelligence, artificial vision, binarized, grayscale, images, RGB

Procedia PDF Downloads 376

Authors: Julio Jesus Salazar, Julio Jesus De Lama

Abstract:

the fundamental objective of hydrological studies applied to the engineering field is to determine the statistically consistent volumes or water flows that, in each case, allow us to size or design a series of elements or structures to effectively manage and develop a river basin. To determine these values, there are several ways of working within the framework of traditional hydrology: (1) Study each of the factors that influence the hydrological cycle, (2) Study the historical behavior of the hydrology of the area, (3) Study the historical behavior of hydrologically similar zones, and (4) Other studies (rain simulators or experimental basins). Of course, this range of studies in a certain basin is very varied and complex and presents the difficulty of collecting the data in real time. In this complex space, the study of variables can only be overcome by collecting and transmitting data to decision centers through the Internet of things and artificial intelligence. Thus, this research work implemented the learning project of the sub-basin of the Shullcas river in the Andean basin of the Mantaro river in Peru. The sensor firmware to collect and communicate hydrological parameter data was programmed and tested in similar basins of the European Union. The Machine Learning applications was programmed to choose the algorithms that direct the best solution to the determination of the rainfall-runoff relationship captured in the different polygons of the sub-basin. Tests were carried out in the mountains of Europe, and in the sub-basins of the Shullcas river (Huancayo) and the Yauli river (Jauja) with heights close to 5000 m.a.s.l., giving the following conclusions: to guarantee a correct communication, the distance between devices should not pass the 15 km. It is advisable to minimize the energy consumption of the devices and avoid collisions between packages, the distances oscillate between 5 and 10 km, in this way the transmission power can be reduced and a higher bitrate can be used. In case the communication elements of the devices of the network (internet of things) installed in the basin do not have good visibility between them, the distance should be reduced to the range of 1-3 km. The energy efficiency of the Atmel microcontrollers present in Arduino is not adequate to meet the requirements of system autonomy. To increase the autonomy of the system, it is recommended to use low consumption systems, such as the Ashton Raggatt McDougall or ARM Cortex L (Ultra Low Power) microcontrollers or even the Cortex M; and high-performance direct current (DC) to direct current (DC) converters. The Machine Learning System has initiated the learning of the Shullcas system to generate the best hydrology of the sub-basin. This will improve as machine learning and the data entered in the big data coincide every second. This will provide services to each of the applications of the complex system to return the best data of determined flows.

Keywords: hydrology, internet of things, machine learning, river basin

Procedia PDF Downloads 155

Authors: Anjalika Fernando, Banuka Athuraliya

Abstract:

This paper presents a comprehensive approach for converting Sinhala Sign Language (SSL) into grammatically correct sentences using Natural Language Processing (NLP) techniques in real-time. While previous studies have explored various aspects of SSL translation, the research gap lies in the absence of grammar checking for SSL. This work aims to bridge this gap by proposing a two-stage methodology that leverages deep learning models to detect signs and translate them into coherent sentences, ensuring grammatical accuracy. The first stage of the approach involves the utilization of a Long Short-Term Memory (LSTM) deep learning model to recognize and interpret SSL signs. By training the LSTM model on a dataset of SSL gestures, it learns to accurately classify and translate these signs into textual representations. The LSTM model achieves a commendable accuracy rate of 94%, demonstrating its effectiveness in accurately recognizing and translating SSL gestures. Building upon the successful recognition and translation of SSL signs, the second stage of the methodology focuses on improving the grammatical correctness of the translated sentences. The project employs a Neural Machine Translation (NMT) architecture, consisting of an encoder and decoder with LSTM components, to enhance the syntactical structure of the generated sentences. By training the NMT model on a parallel corpus of Sinhala wrong sentences and their corresponding grammatically correct translations, it learns to generate coherent and grammatically accurate sentences. The NMT model achieves an impressive accuracy rate of 98%, affirming its capability to produce linguistically sound translations. The proposed approach offers significant contributions to the field of SSL translation and grammar correction. Addressing the critical issue of grammar checking, it enhances the usability and reliability of SSL translation systems, facilitating effective communication between hearing-impaired and non-sign language users. Furthermore, the integration of deep learning techniques, such as LSTM and NMT, ensures the accuracy and robustness of the translation process. This research holds great potential for practical applications, including educational platforms, accessibility tools, and communication aids for the hearing-impaired. Furthermore, it lays the foundation for future advancements in SSL translation systems, fostering inclusive and equal opportunities for the deaf community. Future work includes expanding the existing datasets to further improve the accuracy and generalization of the SSL translation system. Additionally, the development of a dedicated mobile application would enhance the accessibility and convenience of SSL translation on handheld devices. Furthermore, efforts will be made to enhance the current application for educational purposes, enabling individuals to learn and practice SSL more effectively. Another area of future exploration involves enabling two-way communication, allowing seamless interaction between sign-language users and non-sign-language users.In conclusion, this paper presents a novel approach for converting Sinhala Sign Language gestures into grammatically correct sentences using NLP techniques in real time. The two-stage methodology, comprising an LSTM model for sign detection and translation and an NMT model for grammar correction, achieves high accuracy rates of 94% and 98%, respectively. By addressing the lack of grammar checking in existing SSL translation research, this work contributes significantly to the development of more accurate and reliable SSL translation systems, thereby fostering effective communication and inclusivity for the hearing-impaired community

Keywords: Sinhala sign language, sign Language, NLP, LSTM, NMT

Procedia PDF Downloads 96

Authors: Matthew Yeager, Christopher Willy, John Bischoff

Abstract:

The conceptualization and design phases of a system lifecycle consume a significant amount of the lifecycle budget in the form of direct tasking and capital, as well as the implicit costs associated with unforeseeable design errors that are only realized during downstream phases. Ad hoc or iterative approaches to generating system requirements oftentimes fail to consider the full array of feasible systems or product designs for a variety of reasons, including, but not limited to: initial conceptualization that oftentimes incorporates a priori or legacy features; the inability to capture, communicate and accommodate stakeholder preferences; inadequate technical designs and/or feasibility studies; and locally-, but not globally-, optimized subsystems and components. These design pitfalls can beget unanticipated developmental or system alterations with added costs, risks and support activities, heightening the risk for suboptimal system performance, premature obsolescence or forgone development. Supported by rapid advances in learning algorithms and hardware technology, sensors and sensor systems have become commonplace in both commercial and industrial products. The evolving array of hardware components (i.e. sensors, CPUs, modular / auxiliary access, etc…) as well as recognition, data fusion and communication protocols have all become increasingly complex and critical for design engineers during both concpetualization and implementation. This work seeks to develop and utilize a non-deterministic approach for sensor system design within the multi-attribute tradespace exploration (MATE) paradigm, a technique that incorporates decision theory into model-based techniques in order to explore complex design environments and discover better system designs. Developed to address the inherent design constraints in complex aerospace systems, MATE techniques enable project engineers to examine all viable system designs, assess attribute utility and system performance, and better align with stakeholder requirements. Whereas such previous work has been focused on aerospace systems and conducted in a deterministic fashion, this study addresses a wider array of system design elements by incorporating both traditional tradespace elements (e.g. hardware components) as well as popular multi-sensor data fusion models and techniques. Furthermore, statistical performance features to this model-based MATE approach will enable non-deterministic techniques for various commercial systems that range in application, complexity and system behavior, demonstrating a significant utility within the realm of formal systems decision-making.

Keywords: multi-attribute tradespace exploration, data fusion, sensors, systems engineering, system design

Procedia PDF Downloads 176

Authors: Yiea Funk Te, Daniel Mueller, Irena Pletikosa Cvijikj

Abstract:

Small and medium enterprises (SMEs) play an important role in the economy of many countries. When the overall world economy is considered, SMEs represent 95% of all businesses in the world, accounting for 66% of the total employment. Existing studies show that the current business environment is characterized as highly turbulent and strongly influenced by modern information and communication technologies, thus forcing SMEs to experience more severe challenges in maintaining their existence and expanding their business. To support SMEs at improving their competitiveness, researchers recently turned their focus on applying data mining techniques to build risk and growth prediction models. However, data used to assess risk and growth indicators is primarily obtained via questionnaires, which is very laborious and time-consuming, or is provided by financial institutes, thus highly sensitive to privacy issues. Recently, web mining (WM) has emerged as a new approach towards obtaining valuable insights in the business world. WM enables automatic and large scale collection and analysis of potentially valuable data from various online platforms, including companies’ websites. While WM methods have been frequently studied to anticipate growth of sales volume for e-commerce platforms, their application for assessment of SME risk and growth indicators is still scarce. Considering that a vast proportion of SMEs own a website, WM bears a great potential in revealing valuable information hidden in SME websites, which can further be used to understand SME risk and growth indicators, as well as to enhance current SME risk and growth prediction models. This study aims at developing an automated system to collect business-relevant data from the Web and predict future growth trends of SMEs by means of WM and data mining techniques. The envisioned system should serve as an 'early recognition system' for future growth opportunities. In an initial step, we examine how structured and semi-structured Web data in governmental or SME websites can be used to explain the success of SMEs. WM methods are applied to extract Web data in a form of additional input features for the growth prediction model. The data on SMEs provided by a large Swiss insurance company is used as ground truth data (i.e. growth-labeled data) to train the growth prediction model. Different machine learning classification algorithms such as the Support Vector Machine, Random Forest and Artificial Neural Network are applied and compared, with the goal to optimize the prediction performance. The results are compared to those from previous studies, in order to assess the contribution of growth indicators retrieved from the Web for increasing the predictive power of the model.

Keywords: data mining, SME growth, success factors, web mining

Procedia PDF Downloads 264

Authors: Burak Oksuzoglu, Savas Yildirim, Ferhat Kutlu

Abstract:

Opinion mining (OM) is one of the natural language processing (NLP) problems to determine the polarity of opinions, mostly represented on a positive-neutral-negative axis. The data for OM is usually collected from various social media platforms. In an era where social media has considerable control over companies’ futures, it’s worth understanding social media and taking actions accordingly. OM comes to the fore here as the scale of the discussion about companies increases, and it becomes unfeasible to gauge opinion on individual levels. Thus, the companies opt to automize this process by applying machine learning (ML) approaches to their data. For the last two decades, OM or sentiment analysis (SA) has been mainly performed by applying ML classification algorithms such as support vector machines (SVM) and Naïve Bayes to a bag of n-gram representations of textual data. With the advent of deep learning and its apparent success in NLP, traditional methods have become obsolete. Transfer learning paradigm that has been commonly used in computer vision (CV) problems started to shape NLP approaches and language models (LM) lately. This gave a sudden rise to the usage of the pretrained language model (PTM), which contains language representations that are obtained by training it on the large datasets using self-supervised learning objectives. The PTMs are further fine-tuned by a specialized downstream task dataset to produce efficient models for various NLP tasks such as OM, NER (Named-Entity Recognition), Question Answering (QA), and so forth. In this study, the traditional and modern NLP approaches have been evaluated for OM by using a sizable corpus belonging to a large private company containing about 76,000 comments in Turkish: SVM with a bag of n-grams, and two chosen pre-trained models, multilingual universal sentence encoder (MUSE) and bidirectional encoder representations from transformers (BERT). The MUSE model is a multilingual model that supports 16 languages, including Turkish, and it is based on convolutional neural networks. The BERT is a monolingual model in our case and transformers-based neural networks. It uses a masked language model and next sentence prediction tasks that allow the bidirectional training of the transformers. During the training phase of the architecture, pre-processing operations such as morphological parsing, stemming, and spelling correction was not used since the experiments showed that their contribution to the model performance was found insignificant even though Turkish is a highly agglutinative and inflective language. The results show that usage of deep learning methods with pre-trained models and fine-tuning achieve about 11% improvement over SVM for OM. The BERT model achieved around 94% prediction accuracy while the MUSE model achieved around 88% and SVM did around 83%. The MUSE multilingual model shows better results than SVM, but it still performs worse than the monolingual BERT model.

Keywords: BERT, MUSE, opinion mining, pretrained language model, SVM, Turkish

Procedia PDF Downloads 139

Authors: Anishur Rahman, Maria Areias, Diogo Simões, Ana Figeuiredo, Filipa Figueiredo, João Nunes

Abstract:

The circular economy (CE) has gained increased awareness among academics, businesses, and decision-makers as it stimulates resource circularity in the production and consumption systems. A large epistemological study has explored the principles of CE, but scant attention eagerly focused on analysing how CE is evaluated, consented to, and enforced using economic metabolism data and business intelligent framework. Economic metabolism involves the ongoing exchange of materials and energy within and across socio-economic systems and requires the assessment of vast amounts of data to provide quantitative analysis related to effective resource management. Limited concern, the present work has focused on the regional flows pilot region from Portugal. By addressing this gap, this study aims to promote eco-innovation and sustainability in the regions of Intermunicipal Communities Região de Coimbra, Viseu Dão Lafões and Beiras e Serra da Estrela, using this data to find precise synergies in terms of material flows and give companies a competitive advantage in form of valuable waste destinations, access to new resources and new markets, cost reduction and risk sharing benefits. In our work, emphasis on applying artificial intelligence (AI) and, more specifically, on implementing state-of-the-art deep learning algorithms is placed, contributing to construction a business intelligent approach. With the emergence of new approaches generally highlighted under the sub-heading of AI and machine learning (ML), the methods for statistical analysis of complex and uncertain production systems are facing significant changes. Therefore, various definitions of AI and its differences from traditional statistics are presented, and furthermore, ML is introduced to identify its place in data science and the differences in topics such as big data analytics and in production problems that using AI and ML are identified. A lifecycle-based approach is then taken to analyse the use of different methods in each phase to identify the most useful technologies and unifying attributes of AI in manufacturing. Most of macroeconomic metabolisms models are mainly direct to contexts of large metropolis, neglecting rural territories, so within this project, a dynamic decision support model coupled with artificial intelligence tools and information platforms will be developed, focused on the reality of these transition zones between the rural and urban. Thus, a real decision support tool is under development, which will surpass the scientific developments carried out to date and will allow to overcome imitations related to the availability and reliability of data.

Keywords: circular economy, artificial intelligence, economic metabolisms, machine learning

Procedia PDF Downloads 66

Authors: Javier Carroquino, Nieves García-Casarejos, Pilar Gargallo, F. Javier García-Ramos

Abstract:

The energy used in agriculture is a source of global emissions of greenhouse gases. The two main types of this energy are electricity for pumping and diesel for agricultural machinery. In order to reduce these emissions, the European project LIFE REWIND addresses the supply of this demand from renewable sources. First of all, comprehensive data on energy demand and available renewable resources have been obtained in several case studies. Secondly, a set of simulations and optimizations have been performed, in search of the best configuration and sizing, both from an economic and emission reduction point of view. For this purpose, it was used software based on genetic algorithms. Thirdly, a prototype has been designed and installed, that it is being used for the validation in a real case. Finally, throughout a year of operation, various technical and economic parameters are being measured for further analysis. The prototype is not connected to the utility grid, avoiding the cost and environmental impact of a grid extension. The system includes three kinds of photovoltaic fields. One is located on a fixed structure on the terrain. Another one is floating on an irrigation raft. The last one is mounted on a two axis solar tracker. Each has its own solar inverter. The total amount of nominal power is 44 kW. A lead acid battery with 120 kWh of capacity carries out the energy storage. Three isolated inverters support a three phase, 400 V 50 Hz micro-grid, the same characteristics of the utility grid. An advanced control subsystem has been constructed, using free hardware and software. The electricity produced feeds a set of seven pumps used for purification, elevation and pressurization of water in a drip irrigation system located in a vineyard. Since the irrigation season does not include the whole year, as well as a small oversize of the generator, there is an amount of surplus energy. With this surplus, a hydrolyser produces on site hydrogen by electrolysis of water. An off-road vehicle with fuel cell feeds on that hydrogen and carries people in the vineyard. The only emission of the process is high purity water. On the one hand, the results show the technical and economic feasibility of stand-alone renewable energy systems to feed seasonal pumping. In this way, the economic costs, the environmental impacts and the landscape impacts of grid extensions are avoided. The use of diesel gensets and their associated emissions are also avoided. On the other hand, it is shown that it is possible to replace diesel in agricultural machinery, substituting it for electricity or hydrogen of 100% renewable origin and produced on the farm itself, without any external energy input. In addition, it is expected to obtain positive effects on the rural economy and employment, which will be quantified through interviews.

Keywords: drip irrigation, greenhouse gases, hydrogen, renewable energy, vineyard

Procedia PDF Downloads 339

Authors: Vanessa Roulet, Marc Turini, Juliane Hey, Stéphanie Bord-Romeu, Emilie Bonzom, Mahmoud Badawi, Mohammed-Amine Chakir, Valérie Simon, Vanessa Viotti, Jérémie Gautier, Françoise Le Boulaire, Catherine Coignard, Claire Vincent, Sandrine Greaume, Isabelle Voisin

Abstract:

Background: Beckman Coulter, Inc. has recently developed fully automated assays for the detection of HBsAg on a new immunoassay platform. The objective of this European multicenter study was to evaluate the performance of the ACCESS HBsAg and ACCESS HBsAg Confirmatory assays† on the recently CE-marked DxI 9000 ACCESS Immunoassay Analyzer. Methods: The clinical specificity of the ACCESS HBsAg and HBsAg Confirmatory assays was determined using HBsAg-negative samples from blood donors and hospitalized patients. The clinical sensitivity was determined using presumed HBsAg-positive samples. Sample HBsAg status was determined using a CE-marked HBsAg assay (Abbott ARCHITECT HBsAg Qualitative II, Roche Elecsys HBsAg II, or Abbott PRISM HBsAg assay) and a CE-marked HBsAg confirmatory assay (Abbott ARCHITECT HBsAg Qualitative II Confirmatory or Abbott PRISM HBsAg Confirmatory assay) according to manufacturer package inserts and pre-determined testing algorithms. False initial reactive rate was determined on fresh hospitalized patient samples. The sensitivity for the early detection of HBV infection was assessed internally on thirty (30) seroconversion panels. Results: Clinical specificity was 99.95% (95% CI, 99.86 – 99.99%) on 6047 blood donors and 99.71% (95%CI, 99.15 – 99.94%) on 1023 hospitalized patient samples. A total of six (6) samples were found false positive with the ACCESS HBsAg assay. None were confirmed for the presence of HBsAg with the ACCESS HBsAg Confirmatory assay. Clinical sensitivity on 455 HBsAg-positive samples was 100.00% (95% CI, 99.19 – 100.00%) for the ACCESS HBsAg assay alone and for the ACCESS HBsAg Confirmatory assay. The false initial reactive rate on 821 fresh hospitalized patient samples was 0.24% (95% CI, 0.03 – 0.87%). Results obtained on 30 seroconversion panels demonstrated that the ACCESS HBsAg assay had equivalent sensitivity performances compared to the Abbott ARCHITECT HBsAg Qualitative II assay with an average bleed difference since first reactive bleed of 0.13. All bleeds found reactive in ACCESS HBsAg assay were confirmed in ACCESS HBsAg Confirmatory assay. Conclusion: The newly developed ACCESS HBsAg and ACCESS HBsAg Confirmatory assays from Beckman Coulter have demonstrated high clinical sensitivity and specificity, equivalent to currently marketed HBsAg assays, as well as a low false initial reactive rate. †Pending achievement of CE compliance; not yet available for in vitro diagnostic use. 2023-11317 Beckman Coulter and the Beckman Coulter product and service marks mentioned herein are trademarks or registered trademarks of Beckman Coulter, Inc. in the United States and other countries. All other trademarks are the property of their respective owners.

Keywords: dxi 9000 access immunoassay analyzer, hbsag, hbv, hepatitis b surface antigen, hepatitis b virus, immunoassay

Procedia PDF Downloads 83

Authors: Baha Eddine Aissou, Aichouche Belhadj Aissa

Abstract:

Light detection and ranging (LiDAR) is an active remote sensing technology used for several applications. Airborne LiDAR is becoming an important technology for the acquisition of a highly accurate dense point cloud. A classification of airborne laser scanning (ALS) point cloud is a very important task that still remains a real challenge for many scientists. Support vector machine (SVM) is one of the most used statistical learning algorithms based on kernels. SVM is a non-parametric method, and it is recommended to be used in cases where the data distribution cannot be well modeled by a standard parametric probability density function. Using a kernel, it performs a robust non-linear classification of samples. Often, the data are rarely linearly separable. SVMs are able to map the data into a higher-dimensional space to become linearly separable, which allows performing all the computations in the original space. This is one of the main reasons that SVMs are well suited for high-dimensional classification problems. Only a few training samples, called support vectors, are required. SVM has also shown its potential to cope with uncertainty in data caused by noise and fluctuation, and it is computationally efficient as compared to several other methods. Such properties are particularly suited for remote sensing classification problems and explain their recent adoption. In this poster, the SVM classification of ALS LiDAR data is proposed. Firstly, connected component analysis is applied for clustering the point cloud. Secondly, the resulting clusters are incorporated in the SVM classifier. Radial basic function (RFB) kernel is used due to the few numbers of parameters (C and γ) that needs to be chosen, which decreases the computation time. In order to optimize the classification rates, the parameters selection is explored. It consists to find the parameters (C and γ) leading to the best overall accuracy using grid search and 5-fold cross-validation. The exploited LiDAR point cloud is provided by the German Society for Photogrammetry, Remote Sensing, and Geoinformation. The ALS data used is characterized by a low density (4-6 points/m²) and is covering an urban area located in residential parts of the city Vaihingen in southern Germany. The class ground and three other classes belonging to roof superstructures are considered, i.e., a total of 4 classes. The training and test sets are selected randomly several times. The obtained results demonstrated that a parameters selection can orient the selection in a restricted interval of (C and γ) that can be further explored but does not systematically lead to the optimal rates. The SVM classifier with hyper-parameters is compared with the most used classifiers in literature for LiDAR data, random forest, AdaBoost, and decision tree. The comparison showed the superiority of the SVM classifier using parameters selection for LiDAR data compared to other classifiers.

Keywords: classification, airborne LiDAR, parameters selection, support vector machine

Procedia PDF Downloads 146

Authors: Stojan Kravanja, Andrej Ivanič, Tomaž Žula

Abstract:

This contribution focuses on structural optimization in civil engineering using mixed integer non-linear programming (MINLP). MINLP is characterized as a versatile method that can handle both continuous and discrete optimization variables simultaneously. Continuous variables are used to optimize parameters such as dimensions, stresses, masses, or costs, while discrete variables represent binary decisions to determine the presence or absence of structural elements within a structure while also calculating discrete materials and standard sections. The optimization process is divided into three main steps. First, a mechanical superstructure with a variety of different topology-, material- and dimensional alternatives. Next, a MINLP model is formulated to encapsulate the optimization problem. Finally, an optimal solution is searched in the direction of the defined objective function while respecting the structural constraints. The economic or mass objective function of the material and labor costs of a structure is subjected to the constraints known from structural analysis. These constraints include equations for the calculation of internal forces and deflections, as well as equations for the dimensioning of structural components (in accordance with the Eurocode standards). Given the complex, non-convex and highly non-linear nature of optimization problems in civil engineering, the Modified Outer-Approximation/Equality-Relaxation (OA/ER) algorithm is applied. This algorithm alternately solves subproblems of non-linear programming (NLP) and main problems of mixed-integer linear programming (MILP), in this way gradually refines the solution space up to the optimal solution. The NLP corresponds to the continuous optimization of parameters (with fixed topology, discrete materials and standard dimensions, all determined in the previous MILP), while the MILP involves a global approximation to the superstructure of alternatives, where a new topology, materials, standard dimensions are determined. The optimization of a convex problem is stopped when the MILP solution becomes better than the best NLP solution. Otherwise, it is terminated when the NLP solution can no longer be improved. While the OA/ER algorithm, like all other algorithms, does not guarantee global optimality due to the presence of non-convex functions, various modifications, including convexity tests, are implemented in OA/ER to mitigate these difficulties. The effectiveness of the proposed MINLP approach is demonstrated by its application to various structural optimization tasks, such as mass optimization of steel buildings, cost optimization of timber halls, composite floor systems, etc. Special optimization models have been developed for the optimization of these structures. The MINLP optimizations, facilitated by the user-friendly software package MIPSYN, provide insights into a mass or cost-optimal solutions, optimal structural topologies, optimal material and standard cross-section choices, confirming MINLP as a valuable method for the optimization of structures in civil engineering.

Keywords: MINLP, mixed-integer non-linear programming, optimization, structures

Procedia PDF Downloads 44

Authors: Keivan Davami, Michael Munther, Lloyd Hackel

Abstract:

The control of properties and material behavior by implementing thermal-mechanical processes is based on mechanical deformation and annealing according to a precise schedule that will produce a unique and stable combination of grain structure, dislocation substructure, texture, and dispersion of precipitated phases. The authors recently developed a thermal-mechanical technique to stabilize the microstructure of additively manufactured nickel-based superalloys even after exposure to high temperatures. However, the mechanism(s) that controls this stability is still under investigation. Laser peening (LP), also called laser shock peening (LSP), is a shock based (50 ns duration) post-processing technique used for extending performance levels and improving service life of critical components by developing deep levels of plastic deformation, thereby generating high density of dislocations and inducing compressive residual stresses in the surface and deep subsurface of components. These compressive residual stresses are usually accompanied with an increase in hardness and enhance the material’s resistance to surface-related failures such as creep, fatigue, contact damage, and stress corrosion cracking. While the LP process enhances the life span and durability of the material, the induced compressive residual stresses relax at high temperatures (>0.5Tm, where Tm is the absolute melting temperature), limiting the applicability of the technology. At temperatures above 0.5Tm, the compressive residual stresses relax, and yield strength begins to drop dramatically. The principal reason is the increasing rate of solid-state diffusion, which affects both the dislocations and the microstructural barriers. Dislocation configurations commonly recover by mechanisms such as climbing and recombining rapidly at high temperatures. Furthermore, precipitates coarsen, and grains grow; virtually all of the available microstructural barriers become ineffective.Our results indicate that by using “cyclic” treatments with sequential LP and annealing steps, the compressive stresses survive, and the microstructure is stable after exposure to temperatures exceeding 0.5Tm for a long period of time. When the laser peening process is combined with annealing, dislocations formed as a result of LPand precipitates formed during annealing have a complex interaction that provides further stability at high temperatures. From a scientific point of view, this research lays the groundwork for studying a variety of physical, materials science, and mechanical engineering concepts. This research could lead to metals operating at higher sustained temperatures enabling improved system efficiencies. The strengthening of metals by a variety of means (alloying, work hardening, and other processes) has been of interest for a wide range of applications. However, the mechanistic understanding of the often complex processes of interactionsbetween dislocations with solute atoms and with precipitates during plastic deformation have largely remained scattered in the literature. In this research, the elucidation of the actual mechanisms involved in the novel cyclic LP/annealing processes as a scientific pursuit is investigated through parallel studies of dislocation theory and the implementation of advanced experimental tools. The results of this research help with the validation of a novel laser processing technique for high temperature applications. This will greatly expand the applications of the laser peening technology originally devised only for temperatures lower than half of the melting temperature.

Keywords: laser shock peening, mechanical properties, indentation, high temperature stability

Procedia PDF Downloads 145

Authors: Sergio J. Calleja, Adria Roca, José D. Santotoribio

Abstract:

Pancreatic endocrine diseases include pathologies like insulin resistance (IR), prediabetes, and type 2 diabetes mellitus (DM2). Some of them are highly prevalent in the U.S.—40% of U.S. adults have IR, 38% of U.S. adults have prediabetes, and 12% of U.S. adults have DM2—, as reported by the National Center for Biotechnology Information (NCBI). Building upon this imperative, the objective of the present study was to develop a non-invasive test for the assessment of the patient’s pancreatic endocrine function and to evaluate its accuracy in detecting various pancreatic endocrine diseases, such as IR, prediabetes, and DM2. This approach to a routine blood and urine test is based around serum and urine biomarkers. It is made by the combination of several independent public algorithms, such as the Adult Treatment Panel III (ATP-III), triglycerides and glucose (TyG) index, homeostasis model assessment-insulin resistance (HOMA-IR), HOMA-2, and the quantitative insulin-sensitivity check index (QUICKI). Additionally, it incorporates essential measurements such as the creatinine clearance, estimated glomerular filtration rate (eGFR), urine albumin-to-creatinine ratio (ACR), and urinalysis, which are helpful to achieve a full image of the patient’s pancreatic endocrine disease. To evaluate the estimated accuracy of this test, an iterative process was performed by a machine learning (ML) algorithm, with a training set of 9,391 patients. The sensitivity achieved was 97.98% and the specificity was 99.13%. Consequently, the area under the receiver operating characteristic (AUROC) curve, the positive predictive value (PPV), and the negative predictive value (NPV) were 92.48%, 99.12%, and 98.00%, respectively. The algorithm was validated with a randomized controlled trial (RCT) with a target sample size (n) of 314 patients. However, 50 patients were initially excluded from the study, because they had ongoing clinically diagnosed pathologies, symptoms or signs, so the n dropped to 264 patients. Then, 110 patients were excluded because they didn’t show up at the clinical facility for any of the follow-up visits—this is a critical point to improve for the upcoming RCT, since the cost of each patient is very high and for this RCT almost a third of the patients already tested were lost—, so the new n consisted of 154 patients. After that, 2 patients were excluded, because some of their laboratory parameters and/or clinical information were wrong or incorrect. Thus, a final n of 152 patients was achieved. In this validation set, the results obtained were: 100.00% sensitivity, 100.00% specificity, 100.00% AUROC, 100.00% PPV, and 100.00% NPV. These results suggest that this approach to a routine blood and urine test holds promise in providing timely and accurate diagnoses of pancreatic endocrine diseases, particularly among individuals aged 40 and above. Given the current epidemiological state of these type of diseases, these findings underscore the significance of early detection. Furthermore, they advocate for further exploration, prompting the intention to conduct a clinical trial involving 26,000 participants (from March 2025 to December 2026).

Keywords: algorithm, diabetes, laboratory medicine, non-invasive

Procedia PDF Downloads 24

Authors: Venencia Albert, Arulselvi Subramanian, Hara Prasad Pati, Asok K. Mukhophadhyay

Abstract:

Acute traumatic coagulopathy in an endogenous dysregulation of the intrinsic coagulation system in response to the injury, associated with three-fold risk of poor outcome, and is more amenable to corrective interventions, subsequent to early identification and management. Multiple definitions for stratification of the patients' risk for early acute coagulopathy have been proposed, with considerable variations in the defining criteria, including several trauma-scoring systems based on prehospital data. We aimed to develop a clinically relevant definition for acute coagulopathy of trauma based on conventional coagulation assays and to assess its efficacy in comparison to recently established prehospital prediction models. Methodology: Retrospective data of all trauma patients (n = 490) presented to our level I trauma center, in 2014, was extracted. Receiver operating characteristic curve analysis was done to establish cut-offs for conventional coagulation assays for identification of patients with acute traumatic coagulopathy was done. Prospectively data of (n = 100) adult trauma patients was collected and cohort was stratified by the established definition and classified as "coagulopathic" or "non-coagulopathic" and correlated with the Prediction of acute coagulopathy of trauma score and Trauma-Induced Coagulopathy Clinical Score for identifying trauma coagulopathy and subsequent risk for mortality. Results: Data of 490 trauma patients (average age 31.85±9.04; 86.7% males) was extracted. 53.3% had head injury, 26.6% had fractures, 7.5% had chest and abdominal injury. Acute traumatic coagulopathy was defined as international normalized ratio ≥ 1.19; prothrombin time ≥ 15.5 s; activated partial thromboplastin time ≥ 29 s. Of the 100 adult trauma patients (average age 36.5±14.2; 94% males), 63% had early coagulopathy based on our conventional coagulation assay definition. Overall prediction of acute coagulopathy of trauma score was 118.7±58.5 and trauma-induced coagulopathy clinical score was 3(0-8). Both the scores were higher in coagulopathic than non-coagulopathic patients (prediction of acute coagulopathy of trauma score 123.2±8.3 vs. 110.9±6.8, p-value = 0.31; trauma-induced coagulopathy clinical score 4(3-8) vs. 3(0-8), p-value = 0.89), but not statistically significant. Overall mortality was 41%. Mortality rate was significantly higher in coagulopathic than non-coagulopathic patients (75.5% vs. 54.2%, p-value = 0.04). High prediction of acute coagulopathy of trauma score also significantly associated with mortality (134.2±9.95 vs. 107.8±6.82, p-value = 0.02), whereas trauma-induced coagulopathy clinical score did not vary be survivors and non-survivors. Conclusion: Early coagulopathy was seen in 63% of trauma patients, which was significantly associated with mortality. Acute traumatic coagulopathy defined by conventional coagulation assays (international normalized ratio ≥ 1.19; prothrombin time ≥ 15.5 s; activated partial thromboplastin time ≥ 29 s) demonstrated good ability to identify coagulopathy and subsequent mortality, in comparison to the prehospital parameter-based scoring systems. Prediction of acute coagulopathy of trauma score may be more suited for predicting mortality rather than early coagulopathy. In emergency trauma situations, where immediate corrective measures need to be taken, complex multivariable scoring algorithms may cause delay, whereas coagulation parameters and conventional coagulation tests will give highly specific results.

Keywords: trauma, coagulopathy, prediction, model

Procedia PDF Downloads 174

Authors: Todd H. Buxton, Yong G. Lai

Abstract:

Temperature management in regulated rivers can involve significant expenditures of water to meet the cold-water requirements of species in summer. For this purpose, flows released from Lewiston Dam on the Trinity River in Northern California are 12.7 cms with temperatures around 11oC in July through September to provide adult spring Chinook cold water to hold in deep pools and mature until spawning in fall. The releases are more than double the flow and 10oC colder temperatures than the natural conditions before the dam was built. The high, cold releases provide springers the habitat they require but may suppress the stream food base and limit future populations of salmon by reducing the juvenile fish size and survival to adults via the positive relationship between the two. Field and modeling research was undertaken to explore whether lowering summer releases from Lewiston Dam may promote thermal stratification in river pools so that both the cold-water needs of adult salmon and warmer water requirements of other organisms in the stream biome may be met. For this investigation, a three-dimensional (3D) computational fluid dynamics (CFD) model was developed and validated with field measurements in two deep pools on the Trinity River. Modeling and field observations were then used to identify the flows and temperatures that may form and maintain thermal stratification under different meteorologic conditions. Under low flows, a pool was found to be well mixed and thermally homogenous until temperatures began to stratify shortly after sunrise. Stratification then strengthened through the day until shading from trees and mountains cooled the inlet flow and decayed the thermal gradient, which collapsed shortly before sunset and returned the pool to a well-mixed state. This diurnal process of stratification formation and destruction was closely predicted by the 3D CFD model. Both the model and field observations indicate that thermal stratification maintained the coldest temperatures of the day at ≥2m depth in a pool and provided water that was around 8oC warmer in the upper 2m of the pool. Results further indicate that the stratified pool under low flows provided almost the same daily average temperatures as when flows were an order of magnitude higher and stratification was prevented, indicating significant water savings may be realized in regulated streams while also providing a diversity in water temperatures the ecosystem requires. With confidence in the 3D CFD model, the model is now being applied to a dozen pools in the Trinity River to understand how pool bathymetry influences thermal stratification under variable flows and diurnal temperature variations. This knowledge will be used to expand the results to 52 pools in a 64 km reach below Lewiston Dam that meet the depth criteria (≥2 m) for spring Chinook holding. From this, rating curves will be developed to relate discharge to the volume of pool habitat that provides springers the temperature (<15.6oC daily average), velocity (0.15 to 0.4 m/s) and depths that accommodate the escapement target for spring Chinook (6,000 adults) under maximum fish densities measured in other streams (3.1 m3/fish) during the holding time of year (May through August). Flow releases that meet these goals will be evaluated for water savings relative to the current flow regime and their influence on indicator species, including the Foothill Yellow-Legged Frog, and aspects of the stream biome that support salmon populations, including macroinvertebrate production and juvenile Chinook growth rates.

Keywords: 3D CFD modeling, flow regulation, thermal stratification, chinook salmon, foothill yellow-legged frogs, water managment

Procedia PDF Downloads 61

Authors: Sharon Hovav, Ilya Levner, Oren Nahum, Istvan Szabo

Abstract:

In modern large-scale logistic centers (e.g., big automated warehouses), complex logistic operations performed by human staff (pickers) need to be coordinated with the operations of automated facilities (robots, conveyors, cranes, lifts, flow racks, etc.). The efficiency of advanced logistic centers strongly depends on optimizing picking technologies in synch with the facility/product layout, as well as on optimal distribution of material flows (products) in the system. The challenge is to develop a mathematical operations research (OR) tool that will optimize system cost-effectiveness. In this work, we propose a model that describes an automatic logistic center consisting of a set of workstations located at several galleries (floors), with each station containing a known number of flow racks. The requirements of each product and the working capacity of stations served by a given set of workers (pickers) are assumed as predetermined. The goal of the model is to maximize system efficiency. The proposed model includes two echelons. The first is the setting of the (optimal) number of workstations needed to create the total processing/logistic system, subject to picker capacities. The second echelon deals with the assignment of the products to the workstations and flow racks, aimed to achieve maximal throughputs of picked products over the entire system given picker capacities and budget constraints. The solutions to the problems at the two echelons interact to balance the overall load in the flow racks and maximize overall efficiency. We have developed an operations research model within each echelon. In the first echelon, the problem of calculating the optimal number of workstations is formulated as a non-standard bin-packing problem with capacity constraints for each bin. The problem arising in the second echelon is presented as a constrained product-workstation-flow rack assignment problem with non-standard mini-max criteria in which the workload maximum is calculated across all workstations in the center and the exterior minimum is calculated across all possible product-workstation-flow rack assignments. The OR problems arising in each echelon are proved to be NP-hard. Consequently, we find and develop heuristic and approximation solution algorithms based on exploiting and improving local optimums. The LC model considered in this work is highly dynamic and is recalculated periodically based on updated demand forecasts that reflect market trends, technological changes, seasonality, and the introduction of new items. The suggested two-echelon approach and the min-max balancing scheme are shown to work effectively on illustrative examples and real-life logistic data.

Keywords: logistics center, product-workstation, assignment, maximum performance, load balancing, fast algorithm

Procedia PDF Downloads 225

Authors: Dian Zhang, Brendan Heery, Maria O’Neill, Ciprian Briciu-Burghina, Noel E. O’Connor, Fiona Regan

Abstract:

Recent developments in sensors, wireless data communication and the cloud computing have brought the sensor web to a whole new generation. The introduction of the concept of ‘Internet of Thing (IoT)’ has brought the sensor research into a new level, which involves the developing of long lasting, low cost, environment friendly and smart sensors; new wireless data communication technologies; big data analytics algorithms and cloud based solutions that are tailored to large scale smart sensor network. The next generation of smart sensor network consists of several layers: physical layer, where all the smart sensors resident and data pre-processes occur, either on the sensor itself or field gateway; data transmission layer, where data and instructions exchanges happen; the data process layer, where meaningful information is extracted and organized from the pre-process data stream. There are many definitions of smart sensor, however, to summarize all these definitions, a smart sensor must be Intelligent and Adaptable. In future large scale sensor network, collected data are far too large for traditional applications to send, store or process. The sensor unit must be intelligent that pre-processes collected data locally on board (this process may occur on field gateway depends on the sensor network structure). In this case study, three smart sensing methods, corresponding to simple thresholding, statistical model and machine learning based MoPBAS method, are introduced and their strength and weakness are discussed as an introduction to the smart sensing concept. Data fusion, the integration of data and knowledge from multiple sources, are key components of the next generation smart sensor network. For example, in the water level monitoring system, weather forecast can be extracted from external sources and if a heavy rainfall is expected, the server can send instructions to the sensor notes to, for instance, increase the sampling rate or switch on the sleeping mode vice versa. In this paper, we describe the deployment of 11 affordable water level sensors in the Dublin catchment. The objective of this paper is to use the deployed river level sensor network at the Dodder catchment in Dublin, Ireland as a case study to give a vision of the next generation of a smart sensor network for flood monitoring to assist agencies in making decisions about deploying resources in the case of a severe flood event. Some of the deployed sensors are located alongside traditional water level sensors for validation purposes. Using the 11 deployed river level sensors in a network as a case study, a vision of the next generation of smart sensor network is proposed. Each key component of the smart sensor network is discussed, which hopefully inspires the researchers who are working in the sensor research domain.

Keywords: smart sensing, internet of things, water level sensor, flooding

Procedia PDF Downloads 379

Authors: Mian Huang, Chi Ma, Junyu Lin, William Lu

Abstract:

Introduction: Bone health attracts more attention recently and an intelligent question and answer (QA) chatbot for osteoporosis is helpful for science popularization. With Generative Pretrained Transformer (GPT) technology developing, we build an osteoporosis corpus dataset and then fine-tune LLaMA, a famous open-source GPT foundation large language model(LLM), on our self-constructed osteoporosis corpus. Evaluated by clinical orthopedic experts, our fine-tuned model outperforms vanilla LLaMA on osteoporosis QA task in Chinese. Three-dimensional quantitative computed tomography (QCT) measured bone mineral density (BMD) is considered as more accurate than DXA for BMD measurement in recent years. We develop an automatic Phantom-less QCT(PL-QCT) that is more efficient for BMD measurement since no need of an external phantom for calibration. Combined with LLM on osteoporosis, our PL-QCT provides efficient and accurate BMD measurement for our chatbot users. Material and Methods: We build an osteoporosis corpus containing about 30,000 Chinese literatures whose titles are related to osteoporosis. The whole process is done automatically, including crawling literatures in .pdf format, localizing text/figure/table region by layout segmentation algorithm and recognizing text by OCR algorithm. We train our model by continuous pre-training with Low-rank Adaptation (LoRA, rank=10) technology to adapt LLaMA-7B model to osteoporosis domain, whose basic principle is to mask the next word in the text and make the model predict that word. The loss function is defined as cross-entropy between the predicted and ground-truth word. Experiment is implemented on single NVIDIA A800 GPU for 15 days. Our automatic PL-QCT BMD measurement adopt AI-associated region-of-interest (ROI) generation algorithm for localizing vertebrae-parallel cylinder in cancellous bone. Due to no phantom for BMD calibration, we calculate ROI BMD by CT-BMD of personal muscle and fat. Results & Discussion: Clinical orthopaedic experts are invited to design 5 osteoporosis questions in Chinese, evaluating performance of vanilla LLaMA and our fine-tuned model. Our model outperforms LLaMA on over 80% of these questions, understanding ‘Expert Consensus on Osteoporosis’, ‘QCT for osteoporosis diagnosis’ and ‘Effect of age on osteoporosis’. Detailed results are shown in appendix. Future work may be done by training a larger LLM on the whole orthopaedics with more high-quality domain data, or a multi-modal GPT combining and understanding X-ray and medical text for orthopaedic computer-aided-diagnosis. However, GPT model gives unexpected outputs sometimes, such as repetitive text or seemingly normal but wrong answer (called ‘hallucination’). Even though GPT give correct answers, it cannot be considered as valid clinical diagnoses instead of clinical doctors. The PL-QCT BMD system provided by Bone’s QCT(Bone’s Technology(Shenzhen) Limited) achieves 0.1448mg/cm2(spine) and 0.0002 mg/cm2(hip) mean absolute error(MAE) and linear correlation coefficient R2=0.9970(spine) and R2=0.9991(hip)(compared to QCT-Pro(Mindways)) on 155 patients in three-center clinical trial in Guangzhou, China. Conclusion: This study builds a Chinese osteoporosis corpus and develops a fine-tuned and domain-adapted LLM as well as a PL-QCT BMD measurement system. Our fine-tuned GPT model shows better capability than LLaMA model on most testing questions on osteoporosis. Combined with our PL-QCT BMD system, we are looking forward to providing science popularization and early morning screening for potential osteoporotic patients.

Keywords: GPT, phantom-less QCT, large language model, osteoporosis

Procedia PDF Downloads 66

Authors: Angelis P. Barlampas

Abstract:

Learning Objectives: The purpose of this study is to inform briefly the reader about the applications of AI in chest radiology. Background: Currently, there are 190 FDA-approved radiology AI applications, with 42 (22%) pertaining specifically to thoracic radiology. Imaging findings OR Procedure details Aids of AI in chest radiology1: Detects and segments pulmonary nodules. Subtracts bone to provide an unobstructed view of the underlying lung parenchyma and provides further information on nodule characteristics, such as nodule location, nodule two-dimensional size or three dimensional (3D) volume, change in nodule size over time, attenuation data (i.e., mean, minimum, and/or maximum Hounsfield units [HU]), morphological assessments, or combinations of the above. Reclassifies indeterminate pulmonary nodules into low or high risk with higher accuracy than conventional risk models. Detects pleural effusion . Differentiates tension pneumothorax from nontension pneumothorax. Detects cardiomegaly, calcification, consolidation, mediastinal widening, atelectasis, fibrosis and pneumoperitoneum. Localises automatically vertebrae segments, labels ribs and detects rib fractures. Measures the distance from the tube tip to the carina and localizes both endotracheal tubes and central vascular lines. Detects consolidation and progression of parenchymal diseases such as pulmonary fibrosis or chronic obstructive pulmonary disease (COPD).Can evaluate lobar volumes. Identifies and labels pulmonary bronchi and vasculature and quantifies air-trapping. Offers emphysema evaluation. Provides functional respiratory imaging, whereby high-resolution CT images are post-processed to quantify airflow by lung region and may be used to quantify key biomarkers such as airway resistance, air-trapping, ventilation mapping, lung and lobar volume, and blood vessel and airway volume. Assesses the lung parenchyma by way of density evaluation. Provides percentages of tissues within defined attenuation (HU) ranges besides furnishing automated lung segmentation and lung volume information. Improves image quality for noisy images with built-in denoising function. Detects emphysema, a common condition seen in patients with history of smoking and hyperdense or opacified regions, thereby aiding in the diagnosis of certain pathologies, such as COVID-19 pneumonia. It aids in cardiac segmentation and calcium detection, aorta segmentation and diameter measurements, and vertebral body segmentation and density measurements. Conclusion: The future is yet to come, but AI already is a helpful tool for the daily practice in radiology. It is assumed, that the continuing progression of the computerized systems and the improvements in software algorithms , will redder AI into the second hand of the radiologist.

Keywords: artificial intelligence, chest imaging, nodule detection, automated diagnoses

Procedia PDF Downloads 69

Authors: Joel A. Gordon, Nazmiye Balta-Ozkan, Seyed Ali Nabavi

Abstract:

Hydrogen-fueled technologies are rapidly advancing as a critical component of the low-carbon energy transition. In countries historically reliant on natural gas for home heating, such as the UK, hydrogen may prove fundamental for decarbonizing the residential sector, alongside other technologies such as heat pumps and district heat networks. While the UK government is set to take a long-term policy decision on the role of domestic hydrogen by 2026, there are considerable uncertainties regarding consumer preferences for ‘hydrogen homes’ (i.e., hydrogen-fueled appliances for space heating, hot water, and cooking. In comparison to other hydrogen energy technologies, such as road transport applications, to date, few studies have engaged with the social acceptance aspects of the domestic hydrogen transition, resulting in a stark knowledge deficit and pronounced risk to policymaking efforts. In response, this study aims to safeguard against undesirable policy measures by revealing the underlying relationships between the factors of domestic hydrogen acceptance and their respective dimensions: attitudinal, socio-political, community, market, and behavioral acceptance. The study employs an online survey (n=~2100) to gauge how different UK householders perceive the proposition of switching from natural gas to hydrogen-fueled appliances. In addition to accounting for housing characteristics (i.e., housing tenure, property type and number of occupants per dwelling) and several other socio-structural variables (e.g. age, gender, and location), the study explores the impacts of consumer heterogeneity on hydrogen acceptance by recruiting respondents from across five distinct groups: (1) fuel poor householders, (2) technology engaged householders, (3) environmentally engaged householders, (4) technology and environmentally engaged householders, and (5) a baseline group (n=~700) which filters out each of the smaller targeted groups (n=~350). This research design reflects the notion that supporting a socially fair and efficient transition to hydrogen will require parallel engagement with potential early adopters and demographic groups impacted by fuel poverty while also accounting strongly for public attitudes towards net zero. Employing a second-order multigroup confirmatory factor analysis (CFA) in Mplus, the proposed hydrogen acceptance model is tested to fit the data through a partial least squares (PLS) approach. In addition to testing differences between and within groups, the findings provide policymakers with critical insights regarding the significance of knowledge and awareness, safety perceptions, perceived community impacts, cost factors, and trust in key actors and stakeholders as potential explanatory factors of hydrogen acceptance. Preliminary results suggest that knowledge and awareness of hydrogen are positively associated with support for domestic hydrogen at the household, community, and national levels. However, with the exception of technology and/or environmentally engaged citizens, much of the population remains unfamiliar with hydrogen and somewhat skeptical of its application in homes. Knowledge and awareness present as critical to facilitating positive safety perceptions, alongside higher levels of trust and more favorable expectations for community benefits, appliance performance, and potential cost savings. Based on these preliminary findings, policymakers should be put on red alert about diffusing hydrogen into the public consciousness in alignment with energy security, fuel poverty, and net-zero agendas.

Keywords: hydrogen homes, social acceptance, consumer heterogeneity, heat decarbonization

Procedia PDF Downloads 110

Authors: Elham Bagheri, Yalda Mohsenzadeh

Abstract:

Image memorability refers to the phenomenon where certain images are more likely to be remembered by humans than others. It is a quantifiable and intrinsic attribute of an image. Understanding how visual perception and memory interact is important in both cognitive science and artificial intelligence. It reveals the complex processes that support human cognition and helps to improve machine learning algorithms by mimicking the brain's efficient data processing and storage mechanisms. To explore the computational underpinnings of image memorability, this study examines the relationship between an image's reconstruction error, distinctiveness in latent space, and its memorability score. A trained autoencoder is used to replicate human-like memorability assessment inspired by the visual memory game employed in memorability estimations. This study leverages a VGG-based autoencoder that is pre-trained on the vast ImageNet dataset, enabling it to recognize patterns and features that are common to a wide and diverse range of images. An empirical analysis is conducted using the MemCat dataset, which includes 10,000 images from five broad categories: animals, sports, food, landscapes, and vehicles, along with their corresponding memorability scores. The memorability score assigned to each image represents the probability of that image being remembered by participants after a single exposure. The autoencoder is finetuned for one epoch with a batch size of one, attempting to create a scenario similar to human memorability experiments where memorability is quantified by the likelihood of an image being remembered after being seen only once. The reconstruction error, which is quantified as the difference between the original and reconstructed images, serves as a measure of how well the autoencoder has learned to represent the data. The reconstruction error of each image, the error reduction, and its distinctiveness in latent space are calculated and correlated with the memorability score. Distinctiveness is measured as the Euclidean distance between each image's latent representation and its nearest neighbor within the autoencoder's latent space. Different structural and perceptual loss functions are considered to quantify the reconstruction error. The results indicate that there is a strong correlation between the reconstruction error and the distinctiveness of images and their memorability scores. This suggests that images with more unique distinct features that challenge the autoencoder's compressive capacities are inherently more memorable. There is also a negative correlation between the reduction in reconstruction error compared to the autoencoder pre-trained on ImageNet, which suggests that highly memorable images are harder to reconstruct, probably due to having features that are more difficult to learn by the autoencoder. These insights suggest a new pathway for evaluating image memorability, which could potentially impact industries reliant on visual content and mark a step forward in merging the fields of artificial intelligence and cognitive science. The current research opens avenues for utilizing neural representations as instruments for understanding and predicting visual memory.

Keywords: autoencoder, computational vision, image memorability, image reconstruction, memory retention, reconstruction error, visual perception

Procedia PDF Downloads 80