Search results for: deep q-network

Authors: Hansraj Riteesh Bookun, Cassandra Monique Hidajat

Abstract:

This pilot study investigated the efficacy of a newly designed shoe which will act as an intermittent pneumatic compression device to augment venous flow in the lower limb. The aim was to assess the degree with which a wearable intermittent compression device can increase the venous flow in the popliteal vein. Background: Deep venous thrombosis and chronic venous insufficiency are relatively common problems with significant morbidity and mortality. While mechanical and chemical thromboprophylaxis measures are in place in hospital environments (in the form of TED stockings, intermittent pneumatic compression devices, analgesia, antiplatelet and anticoagulant agents), there are limited options in a community setting. Additionally, many individuals are poorly tolerant of graduated compression stockings due to the difficulty in putting them on, their constant tightness and increased associated discomfort in warm weather. These factors may hinder the management of their chronic venous insufficiency. Method: The device is lightweight, easy to wear and comfortable, with a self-contained power source. It features a Bluetooth transmitter and can be controlled with a smartphone. It is externally almost indistinguishable from a normal shoe. During activation, two bladders are inflated -one overlying the metatarsal heads and the second at the pedal arch. The resulting cyclical increase in pressure squeezes blood into the deep venous system. This will decrease periods of stasis and potentially reduce the risk of deep venous thrombosis. The shoe was fitted to 2 healthy participants and the peak systolic velocity of flow in the popliteal vein was measured during and prior to intermittent compression phases. Assessments of total flow volume were also performed. All haemodynamic assessments were performed with ultrasound by a licensed sonographer. Results: Mean peak systolic velocity of 3.5 cm/s with standard deviation of 1.3 cm/s were obtained. There was a three fold increase in mean peak systolic velocity and five fold increase in total flow volume. Conclusion: The device augments venous flow in the leg significantly. This may contribute to lowered thromboembolic risk during periods of prolonged travel or immobility. This device may also serve as an adjunct in the treatment of chronic venous insufficiency. The study will be replicated on a larger scale in a multi—centre trial.

Keywords: venous, intermittent compression, shoe, wearable device

Procedia PDF Downloads 194

Authors: Tom O'Mahony

Abstract:

This article considers the design of a student-centered learning environment in engineering education. The learning environment integrates a number of components, including project-based learning, collaborative learning, two-stage assignments, active learning lectures, and a flipped-classroom. Together these elements place the individual learner and their learning at the center of the environment by focusing on understanding, enhancing relevance, applying learning, obtaining rich feedback, making choices, and taking responsibility. The evolution of this environment from 2014 to the present day is outlined. The impact of this environment on learners and their learning is evaluated via student questionnaires that consist of both open and closed-ended questions. The closed questions indicate that students found the learning environment to be really interesting and enjoyable (rated as 4.7 on a 5 point scale) and encouraged students to adopt a deep approach towards studying the course materials (rated as 4.0 on a 5 point scale). A content analysis of the open-ended questions provides evidence that the project, active learning lectures, and flipped classroom all contribute to the success of this environment. Furthermore, this analysis indicates that the two-stage assessment process, in which feedback is provided between a draft and final assignment, is the key component and the dominant theme. A limitation of the study is the small class size (less than 20 learners per year), but, to some degree, this is compensated for by the longitudinal nature of the study.

Keywords: deep approaches, formative assessment, project-based learning, student-centered learning

Procedia PDF Downloads 112

Authors: Omar Elharrouss, Youssef Hmamouche, Assia Kamal Idrissi, Btissam El Khamlichi, Amal El Fallah-Seghrouchni

Abstract:

Edge detection is represented as one of the most challenging tasks in computer vision, due to the complexity of detecting the edges or boundaries in real-world images that contains objects of different types and scales like trees, building as well as various backgrounds. Edge detection is represented also as a key task for many computer vision applications. Using a set of backbones as well as attention modules, deep-learning-based methods improved the detection of edges compared with the traditional methods like Sobel and Canny. However, images of complex scenes still represent a challenge for these methods. Also, the detected edges using the existing approaches suffer from non-refined results while the image output contains many erroneous edges. To overcome this, n this paper, by using the mechanism of residual learning, a refined edge detection network is proposed (RED-Net). By maintaining the high resolution of edges during the training process, and conserving the resolution of the edge image during the network stage, we make the pooling outputs at each stage connected with the output of the previous layer. Also, after each layer, we use an affined batch normalization layer as an erosion operation for the homogeneous region in the image. The proposed methods are evaluated using the most challenging datasets including BSDS500, NYUD, and Multicue. The obtained results outperform the designed edge detection networks in terms of performance metrics and quality of output images.

Keywords: edge detection, convolutional neural networks, deep learning, scale-representation, backbone

Procedia PDF Downloads 102

Authors: Neritan Shkodrani, Klearta Rrushi, Anxhela Shaha

Abstract:

Generally, the design of axial load capacity of deep foundations is based on the data provided from field tests, such as SPT (Standard Penetration Test) and CPT (Cone Penetration Test) tests. This paper reports the results of axial load capacity analysis of drilled shafts at a construction site at Semani, in Fier county, Fier prefecture in Albania. In this case, the axial load capacity analyses are based on the data of 416 SPT tests and 12 CPTU tests, which are carried out in this site construction using 12 boreholes (10 borings of a depth 30.0 m and 2 borings of a depth of 80.0m). The considered foundation widths range from 0.5m to 2.5 m and foundation embedment lengths is fixed at a value of 25m. SPT – based analytical methods from the Japanese practice of design (Building Standard Law of Japan) and CPT – based analytical Eslami and Fellenius methods are used for obtaining axial ultimate load capacity of drilled shafts. The considered drilled shaft (25m long and 0.5m - 2.5m in diameter) is analyzed for the soil conditions of each borehole. The values obtained from sets of calculations are shown in different charts. Then the reported axial load capacity values acquired from SPT and CPTU data are compared and some conclusions are found related to the mentioned methods of calculations.

Keywords: deep foundations, drilled shafts, axial load capacity, ultimate load capacity, allowable load capacity, SPT test, CPTU test

Procedia PDF Downloads 104

Authors: Chandu Rathnayake, Isuri Anuradha

Abstract:

Early detection and accurate diagnosis of lung diseases play a crucial role in improving patient prognosis. However, conventional diagnostic methods heavily rely on subjective symptom assessments and medical imaging, often causing delays in diagnosis and treatment. To overcome this challenge, we propose a novel lung disease prediction system that integrates patient symptoms and X-ray images to provide a comprehensive and reliable diagnosis.In this project, develop a mobile application specifically designed for detecting lung diseases. Our application leverages both patient symptoms and X-ray images to facilitate diagnosis. By combining these two sources of information, our application delivers a more accurate and comprehensive assessment of the patient's condition, minimizing the risk of misdiagnosis. Our primary aim is to create a user-friendly and accessible tool, particularly important given the current circumstances where many patients face limitations in visiting healthcare facilities. To achieve this, we employ several state-of-the-art algorithms. Firstly, the Decision Tree algorithm is utilized for efficient symptom-based classification. It analyzes patient symptoms and creates a tree-like model to predict the presence of specific lung diseases. Secondly, we employ the Random Forest algorithm, which enhances predictive power by aggregating multiple decision trees. This ensemble technique improves the accuracy and robustness of the diagnosis. Furthermore, we incorporate a deep learning model using Convolutional Neural Network (CNN) with the RestNet50 pre-trained model. CNNs are well-suited for image analysis and feature extraction. By training CNN on a large dataset of X-ray images, it learns to identify patterns and features indicative of lung diseases. The RestNet50 architecture, known for its excellent performance in image recognition tasks, enhances the efficiency and accuracy of our deep learning model. By combining the outputs of the decision tree-based algorithms and the deep learning model, our mobile application generates a comprehensive lung disease prediction. The application provides users with an intuitive interface to input their symptoms and upload X-ray images for analysis. The prediction generated by the system offers valuable insights into the likelihood of various lung diseases, enabling individuals to take appropriate actions and seek timely medical attention. Our proposed mobile application has significant potential to address the rising prevalence of lung diseases, particularly among young individuals with smoking addictions. By providing a quick and user-friendly approach to assessing lung health, our application empowers individuals to monitor their well-being conveniently. This solution also offers immense value in the context of limited access to healthcare facilities, enabling timely detection and intervention. In conclusion, our research presents a comprehensive lung disease prediction system that combines patient symptoms and X-ray images using advanced algorithms. By developing a mobile application, we provide an accessible tool for individuals to assess their lung health conveniently. This solution has the potential to make a significant impact on the early detection and management of lung diseases, benefiting both patients and healthcare providers.

Keywords: CNN, random forest, decision tree, machine learning, deep learning

Procedia PDF Downloads 73

Authors: Rajesh Kumar Gautam, Debabrata Seth

Abstract:

Thioflavin-T (ThT) play a key role of an important biologically active fluorescent sensor for amyloid fibrils. ThT molecule has been developed a method to detect the analysis of different type of diseases such as neurodegenerative disorders, Alzheimer’s, Parkinson’s, and type II diabetes. ThT was used as a fluorescent marker to detect the formation of amyloid fibril. In the presence of amyloid fibril, ThT becomes highly fluorescent. ThT undergoes twisting motion around C-C bonds of the two adjacent benzothiazole and dimethylaniline aromatic rings, which is predominantly affected by the micro-viscosity of the local environment. The present study articulates photophysics and torsional dynamics of biologically active molecule ThT in the presence of deep-eutectic solvents (DESs). DESs are environment-friendly, low cost and biodegradable alternatives to the ionic liquids. DES resembles ionic liquids, but the constituents of a DES include a hydrogen bond donor and acceptor species, in addition to ions. Due to the presence of the H-bonding network within a DES, it exhibits structural heterogeneity. Herein, we have prepared two different DESs by mixing urea with choline chloride and N, N-diethyl ethanol ammonium chloride at ~ 340 K. It was reported that deep eutectic mixture of choline chloride with urea gave a liquid with a freezing point of 12°C. We have experimented by taking two different concentrations of ThT. It was observed that at higher concentration of ThT (50 µM) it forms aggregates in DES. The photophysics of ThT as a function of temperature have been explored by using steady-state, and picoseconds time-resolved fluorescence emission spectroscopic techniques. From the spectroscopic analysis, we have observed that with rising temperature the fluorescence quantum yields and lifetime values of ThT molecule gradually decreases; this is the cumulative effect of thermal quenching and increase in the rate of the torsional rate constant. The fluorescence quantum yield and fluorescence lifetime decay values were always higher for DES-II (urea & N, N-diethyl ethanol ammonium chloride) than those for DES-I (urea & choline chloride). This was mainly due to the presence of structural heterogeneity of the medium. This was further confirmed by comparison with the activation energy of viscous flow with the activation energy of non-radiative decay. ThT molecule in less viscous media undergoes a very fast twisting process and leads to deactivation from the photoexcited state. In this system, the torsional motion increases with increasing temperature. We have concluded that beside bulk viscosity of the media, structural heterogeneity of the medium play crucial role to guide the photophysics of ThT in DESs. The analysis of the experimental data was carried out in the temperature range 288 ≤ T = 333K. The present articulate is to obtain an insight into the DESs as media for studying various photophysical processes of amyloid fibrils sensing molecule of ThT.

Keywords: deep eutectic solvent, photophysics, Thioflavin T, the torsional rate constant

Procedia PDF Downloads 163

Authors: Pitigalage Chamath Chandira Peiris

Abstract:

A widely researched domain in the field of image processing in recent times has been single image super-resolution, which tries to restore a high-resolution image from a single low-resolution image. Many more single image super-resolution efforts have been completed utilizing equally traditional and deep learning methodologies, as well as a variety of other methodologies. Deep learning-based super-resolution methods, in particular, have received significant interest. As of now, the most advanced image restoration approaches are based on convolutional neural networks; nevertheless, only a few efforts have been performed using Transformers, which have demonstrated excellent performance on high-level vision tasks. The effectiveness of CNN-based algorithms in image super-resolution has been impressive. However, these methods cannot completely capture the non-local features of the data. Enhancer is a simple yet powerful Transformer-based approach for enhancing the resolution of images. A method for single image super-resolution was developed in this study, which utilized an efficient and effective transformer design. This proposed architecture makes use of a locally enhanced window transformer block to alleviate the enormous computational load associated with non-overlapping window-based self-attention. Additionally, it incorporates depth-wise convolution in the feed-forward network to enhance its ability to capture local context. This study is assessed by comparing the results obtained for popular datasets to those obtained by other techniques in the domain.

Keywords: single image super resolution, computer vision, vision transformers, image restoration

Procedia PDF Downloads 105

Authors: M. P. Rajeeshkumar, K. V. Aneesh Kumar, J. L. Otero-Ferrer, A. Lombarte, M. Hashim, N. Saravanane, V. N.Sanjeevan, V. M. Tuset

Abstract:

The anglerfishes are widely distributed from shallow to deep-water habitats and are highly diverse in morphology, behaviour, and niche occupancy patterns. To understand this interspecific variability and degree of niche overlap, we performed a functional analysis of five species inhabiting Indian waters where diversity of deep-sea anglerfishes is very high. The sensory capacities (otolith shape and eye size) were also studied to improve the understanding of coexistence of species. The analyses of fish body and otolith shape clustered species in two morphotypes related to phylogenetic lineages: i) Malthopsis lutea, Lophiodes lugubri and Halieutea coccinea were characterized by a dorso-ventrally flattened body with high swimming ability and relative small otoliths, and ii) Chaunax spp. were distinguished by their higher body depth, lower swimming efficiency, and relative big otoliths. The sensory organs did not show a pattern linked to depth distribution of species. However, the larger eye size in M. lutea suggested a nocturnal feeding activity, whereas Chaunax spp. had a large mouth and deeper body in response to different ecological niches. Therefore, the present study supports the hypothesis of spatial and temporal segregation of anglerfishes in the Indian waters, which can be explained from a functional approach and understanding from sensory capabilities.

Keywords: functional traits, otoliths, niche overlap, fishes, Indian waters

Procedia PDF Downloads 136

Authors: Chia-Yi Liu, Bo-Bin Xiao, Wen-Bin Lin, Hsiang-Ning Wu, Liang-Hsun Huang

Abstract:

In order to reduce the risk of radiation damage that personnel may suffer during operations in the radiation environment, the use of automated guided vehicles to assist or replace on-site personnel in the radiation environment has become a key technology and has become an important trend. In this paper, we demonstrate our proof of concept for autonomous self-learning radiation source searcher in an unknown environment without a map. The research uses GPU version of Asynchronous Advantage Actor-Critic network (GA3C) of deep reinforcement learning to search for radiation sources. The searcher network, based on GA3C architecture, has self-directed learned and improved how search the anomalous radiation source by training 1 million episodes under three simulation environments. In each episode of training, the radiation source position, the radiation source intensity, starting position, are all set randomly in one simulation environment. The input for searcher network is the fused data from a 2D laser scanner and a RGB-D camera as well as the value of the radiation detector. The output actions are the linear and angular velocities. The searcher network is trained in a simulation environment to accelerate the learning process. The well-performance searcher network is deployed to the real unmanned vehicle, Dashgo E2, which mounts LIDAR of YDLIDAR G4, RGB-D camera of Intel D455, and radiation detector made by Institute of Nuclear Energy Research. In the field experiment, the unmanned vehicle is enable to search out the radiation source of the 18.5MBq Na-22 by itself and avoid obstacles simultaneously without human interference.

Keywords: deep reinforcement learning, GA3C, source searching, source detection

Procedia PDF Downloads 114

Authors: Levent Taylan Ozgur Yildirim, Qihao Qian, John Yilin Wang

Abstract:

A critical review of assessments of geological carbon dioxide (CO2) storage resources in Pennsylvania and the surrounding region was completed with a focus on the studies of Midwest Regional Carbon Sequestration Partnership (MRCSP), United States Department of Energy (US-DOE), and United States Geological Survey (USGS). Pennsylvania Geological Survey participated in the MRCSP Phase I research to characterize potential storage formations in Pennsylvania. The MRCSP’s volumetric method estimated ~89 gigatonnes (Gt) of total CO2 storage resources in deep saline formations, depleted oil and gas reservoirs, coals, and shales in Pennsylvania. Meanwhile, the US-DOE calculated storage efficiency factors using log-odds normal distribution and Monte Carlo sampling, revealing contingent storage resources of ~18 Gt to ~20 Gt in deep saline formations, depleted oil and gas reservoirs, and coals in Pennsylvania. Additionally, the USGS employed Beta-PERT distribution and Monte Carlo sampling to determine buoyant and residual storage efficiency factors, resulting in 20 Gt of contingent storage resources across four storage assessment units in Appalachian Basin. However, few studies have explored CO2 storage resources in shales in the region, yielding inconclusive findings. This article provides a critical and most up to date review and analysis of geological CO2 storage resources in Pennsylvania and the region.

Keywords: carbon capture and storage, geological CO2 storage, pennsylvania, appalachian basin

Procedia PDF Downloads 54

Authors: Barnabás Körmöndi

Abstract:

The recent crises have affected societies worldwide, resulting in the irresponsible exploitation of natural resources and the unattainability of sustainability. Regions that are economically underdeveloped, such as the Bodrogköz in Eastern Hungary and Slovakia, experience these issues more severely. The aim of this study is to analyze the natural and social resources of the Bodrogköz area for the planning of complex development based on ecotourism. The objective is to develop ecotourism opportunities in this least developed area of the borderland of Hungary and Slovakia. The study utilizes desk research, deep interviews, focus group meetings, and remote sensing methods. Desk research is aimed at providing a comprehensive understanding of the area, while deep interviews and focus group meetings were conducted to understand the stakeholders' perspectives on the potential for ecotourism. Remote sensing methods were used to better understand changes in the natural environment. The study identified the potential for ecotourism development in the Bodrogköz area due to its near-natural habitats along its bordering rivers and rich cultural heritage. The analysis revealed that ecotourism could promote the region's sustainable development, which is essential for its economic growth. Additionally, the study identified the possible threats to the natural environment during ecotourism development and suggested strategies to mitigate these threats. This study highlights the significance of ecotourism in promoting sustainable development in underdeveloped areas such as the Bodrogköz. It provides a basis for future research on ecotourism development and sustainable planning in similar regions. The analysis is based on the data collected through desk research, deep interviews, focus group meetings, and remote sensing. The assessment was conducted through content analysis, which allowed for the identification of themes and patterns in the data. The study addressed the question of how to develop ecotourism in the least developed area of the borderland of Hungary and Slovakia and promote sustainable development in the region. In conclusion, the study highlights the potential for ecotourism development in Bodrogköz and identifies the natural and social resources that contribute to its development. The study emphasizes the need for sustainable development to promote economic growth and mitigate any environmental threats. The findings can inform the development of future strategic plans for ecotourism, promoting sustainable development in underdeveloped regions.

Keywords: ecotourism, natural resources, remote sensing, social development

Procedia PDF Downloads 64

Authors: B. Elshafei, X. Mao

Abstract:

The demand for renewable energy is significantly increasing, major investments are being supplied to the wind power generation industry as a leading source of clean energy. The wind energy sector is entirely dependable and driven by the prediction of wind speed, which by the nature of wind is very stochastic and widely random. This s0tudy employs deep multi-fidelity Gaussian process regression, used to predict wind speeds for medium term time horizons. Data of the RUNE experiment in the west coast of Denmark were provided by the Technical University of Denmark, which represent the wind speed across the study area from the period between December 2015 and March 2016. The study aims to investigate the effect of pre-processing the data by denoising the signal using empirical wavelet transform (EWT) and engaging the vector components of wind speed to increase the number of input data layers for data fusion using deep multi-fidelity Gaussian process regression (GPR). The outcomes were compared using root mean square error (RMSE) and the results demonstrated a significant increase in the accuracy of predictions which demonstrated that using vector components of the wind speed as additional predictors exhibits more accurate predictions than strategies that ignore them, reflecting the importance of the inclusion of all sub data and pre-processing signals for wind speed forecasting models.

Keywords: data fusion, Gaussian process regression, signal denoise, temporal extrapolation

Procedia PDF Downloads 136

Authors: Lin Po-Hsi, Sheu Ming-Thau

Abstract:

Urea cycle disorders (UCD) are rare genetic metabolic disorders that compromise the body's urea cycle. Sodium phenylbutyrate (SPB) is a medication commonly administered in tablet or powder form to lower ammonia levels. Nonetheless, its high sodium content poses risks to sodium-sensitive UCD patients. This necessitates the creation of an alternative drug formulation to mitigate sodium load and optimize drug delivery for UCD patients. This study focused on crafting a novel oral drug formulation for UCD, leveraging choline bicarbonate and phenylbutyric acid. The active pharmaceutical ingredient-ionic liquids (API-ILs) and therapeutic deep eutectic systems (THEDES) were formed by combining these with choline chloride. These systems display characteristics like maintaining a liquid state at room temperature and exhibiting enhanced solubility. This in turn amplifies drug dissolution rate, permeability, and ultimately oral bioavailability. Incorporating choline-based phenylbutyric acid as a substitute for traditional SPB can effectively curtail the sodium load in UCD patients. Our in vitro dissolution experiments revealed that the ILs and DESs, synthesized using choline bicarbonate and choline chloride with phenylbutyric acid, surpassed commercial tablets in dissolution speed. Pharmacokinetic evaluations in SD rats indicated a notable uptick in the oral bioavailability of phenylbutyric acid, underscoring the efficacy of choline salt ILs in augmenting its bioavailability. Additional in vitro intestinal permeability tests on SD rats authenticated that the ILs, formulated with choline bicarbonate and phenylbutyric acid, demonstrate superior permeability compared to their sodium and acid counterparts. To conclude, choline salt ILs developed from choline bicarbonate and phenylbutyric acid present a promising avenue for UCD treatment, with the added benefit of reduced sodium load. They also hold merit in formulation engineering. The sustained-release capabilities of DESs position them favorably for drug delivery, while the low toxicity and cost-effectiveness of choline chloride signal potential in formulation engineering. Overall, this drug formulation heralds a prospective therapeutic avenue for UCD patients.

Keywords: phenylbutyric acid, sodium phenylbutyrate, choline salt, ionic liquids, deep eutectic systems, oral bioavailability

Procedia PDF Downloads 116

Authors: Sara ElElimy, Samir Moustafa

Abstract:

Mobile network operators start to face many challenges in the digital era, especially with high demands from customers. Since mobile network operators are considered a source of big data, traditional techniques are not effective with new era of big data, Internet of things (IoT) and 5G; as a result, handling effectively different big datasets becomes a vital task for operators with the continuous growth of data and moving from long term evolution (LTE) to 5G. So, there is an urgent need for effective Big data analytics to predict future demands, traffic, and network performance to full fill the requirements of the fifth generation of mobile network technology. In this paper, we introduce data science techniques using machine learning and deep learning algorithms: the autoregressive integrated moving average (ARIMA), Bayesian-based curve fitting, and recurrent neural network (RNN) are employed for a data-driven application to mobile network operators. The main framework included in models are identification parameters of each model, estimation, prediction, and final data-driven application of this prediction from business and network performance applications. These models are applied to Telecom Italia Big Data challenge call detail records (CDRs) datasets. The performance of these models is found out using a specific well-known evaluation criteria shows that ARIMA (machine learning-based model) is more accurate as a predictive model in such a dataset than the RNN (deep learning model).

Keywords: big data analytics, machine learning, CDRs, 5G

Procedia PDF Downloads 139

Authors: Padmalochan Panda, R. Ramaseshan

Abstract:

Aluminum nitride (AlN) is a potential candidate for semiconductor industry due to its wide band gap (6.2 eV), high thermal conductivity and low thermal coefficient of expansion. A-plane oriented AlN film finds an important role in deep UV-LED with higher isotropic light extraction efficiency. Also, Cr-doped AlN films exhibit dilute magnetic semiconductor property with high Curie temperature (300 K), and thus compatible with modern day microelectronics. In this work, highly a-axis oriented wurtzite AlN and Al1-xMxN (M = Cr, Ti) films have synthesized by reactive co-sputtering technique at different concentration. Crystal structure of these films is studied by Grazing incidence X-ray diffraction (GIXRD) and Transmission electron microscopy (TEM). Identification of binding energy and concentration (x) in these films is carried out by X-ray photoelectron spectroscopy (XPS). Local crystal structure around the Cr and Ti atom of these films are investigated by X-ray absorption spectroscopy (XAS). It is found that Cr and Ti replace the Al atom in AlN lattice and the bond lengths in first and second coordination sphere with N and Al, respectively, decrease concerning doping concentration due to strong p-d hybridization. The nano-indentation hardness of Cr and Ti-doped AlN films seems to increase from 17.5 GPa (AlN) to around 23 and 27.5 GPa, respectively. An-isotropic optical properties of these films are studied by the Spectroscopic Ellipsometry technique. Refractive index and extinction coefficient of these films are enhanced in normal dispersion region as compared to the parent AlN film. The optical band gap energies also seem to vary between deep UV to UV regions with the addition of Cr, thus by bringing out the usefulness of these films in the area of optoelectronic device applications.

Keywords: ellipsometry, GIXRD, hardness, XAS

Procedia PDF Downloads 114

Authors: Erick C. Valverde, Wansu Lim

Abstract:

Facial emotion recognition (FER) system has been recently developed for more advanced computer vision applications. The ability to identify human emotions would enable smart healthcare facility to diagnose mental health illnesses (e.g., depression and stress) as well as better human social interactions with smart technologies. The FER system involves two steps: 1) face detection task and 2) facial emotion recognition task. It classifies the human expression in various categories such as angry, disgust, fear, happy, sad, surprise, and neutral. This system requires intensive research to address issues with human diversity, various unique human expressions, and variety of human facial features due to age differences. These issues generally affect the ability of the FER system to detect human emotions with high accuracy. Early stage of FER systems used simple supervised classification task algorithms like K-nearest neighbors (KNN) and artificial neural networks (ANN). These conventional FER systems have issues with low accuracy due to its inefficiency to extract significant features of several human emotions. To increase the accuracy of FER systems, deep learning (DL)-based methods, like convolutional neural networks (CNN), are proposed. These methods can find more complex features in the human face by means of the deeper connections within its architectures. However, the inference speed and computational costs of a DL-based FER system is often disregarded in exchange for higher accuracy results. To cope with this drawback, an optimized DL-based FER system is proposed in this study.An extreme version of Inception V3, known as Xception model, is leveraged by applying different network optimization methods. Specifically, network pruning and quantization are used to enable lower computational costs and reduce memory usage, respectively. To support low resource requirements, a 68-landmark face detector from Dlib is used in the early step of the FER system.Furthermore, a DL compiler is utilized to incorporate advanced optimization techniques to the Xception model to improve the inference speed of the FER system. In comparison to VGG-Net and ResNet50, the proposed optimized DL-based FER system experimentally demonstrates the objectives of the network optimization methods used. As a result, the proposed approach can be used to create an efficient and real-time FER system.

Keywords: deep learning, face detection, facial emotion recognition, network optimization methods

Procedia PDF Downloads 118

Authors: Anne Schattel, Mitja Echim, Christof Büskens

Abstract:

Trajectory planning for deep space missions has become a recent topic of great interest. Flying to space objects like asteroids provides two main challenges. One is to find rare earth elements, the other to gain scientific knowledge of the origin of the world. Due to the enormous spatial distances such explorer missions have to be performed unmanned and autonomously. The mathematical field of optimization and optimal control can be used to realize autonomous missions while protecting recourses and making them safer. The resulting algorithms may be applied to other, earth-bound applications like e.g. deep sea navigation and autonomous driving as well. The project KaNaRiA ('Kognitionsbasierte, autonome Navigation am Beispiel des Ressourcenabbaus im All') investigates the possibilities of cognitive autonomous navigation on the example of an asteroid mining mission, including the cruise phase and approach as well as the asteroid rendezvous, landing and surface exploration. To verify and test all methods an interactive, real-time capable simulation using virtual reality is developed under KaNaRiA. This paper focuses on the specific challenge of the guidance during the cruise phase of the spacecraft, i.e. trajectory optimization and optimal control, including first solutions and results. In principle there exist two ways to solve optimal control problems (OCPs), the so called indirect and direct methods. The indirect methods are being studied since several decades and their usage needs advanced skills regarding optimal control theory. The main idea of direct approaches, also known as transcription techniques, is to transform the infinite-dimensional OCP into a finite-dimensional non-linear optimization problem (NLP) via discretization of states and controls. These direct methods are applied in this paper. The resulting high dimensional NLP with constraints can be solved efficiently by special NLP methods, e.g. sequential quadratic programming (SQP) or interior point methods (IP). The movement of the spacecraft due to gravitational influences of the sun and other planets, as well as the thrust commands, is described through ordinary differential equations (ODEs). The competitive mission aims like short flight times and low energy consumption are considered by using a multi-criteria objective function. The resulting non-linear high-dimensional optimization problems are solved by using the software package WORHP ('We Optimize Really Huge Problems'), a software routine combining SQP at an outer level and IP to solve underlying quadratic subproblems. An application-adapted model of impulsive thrusting, as well as a model of an electrically powered spacecraft propulsion system, is introduced. Different priorities and possibilities of a space mission regarding energy cost and flight time duration are investigated by choosing different weighting factors for the multi-criteria objective function. Varying mission trajectories are analyzed and compared, both aiming at different destination asteroids and using different propulsion systems. For the transcription, the robust method of full discretization is used. The results strengthen the need for trajectory optimization as a foundation for autonomous decision making during deep space missions. Simultaneously they show the enormous increase in possibilities for flight maneuvers by being able to consider different and opposite mission objectives.

Keywords: deep space navigation, guidance, multi-objective, non-linear optimization, optimal control, trajectory planning.

Procedia PDF Downloads 412

Authors: K. Ravi, Sabu Subhash

Abstract:

Densely compacted bentonite or bentonite–sand mixture has been identified as a suitable buffer in the deep geological repository (DGR) for the safe disposal of high-level nuclear waste (HLW) due to its favourable physicochemical and hydro-mechanical properties. The addition of sand to the bentonite enhances the thermal conductivity and compaction properties and reduces the drying shrinkage of the buffer material. The buffer material may undergo cyclic wetting and drying upon ingress of groundwater from the surrounding rock mass and from evaporation due to high temperature (50–210 °C) derived from the waste canister. The cycles of changes in temperature may result in thermal history, and the hydro-mechanical properties of the buffer material may be affected. This paper examines the influence of thermal history on the undrained shear strength of bentonite and bentonite-sand mixture. Bentonite from Rajasthan state and sand from the Assam state of India are used in this study. The undrained shear strength values are obtained by conducting unconfined compressive strength (UCS) tests on cylindrical specimens (dry densities 1.30 and 1.5 Mg/m3) of bentonite and bentonite-sand mixture consisting of 30 % bentonite+ 70 % sand. The specimens are preheated at temperatures varying from 50-150 °C for one, two and four hours in hot air oven. The results indicate that the undrained shear strength is increased by the thermal history of the buffer material. The specimens of bentonite-sand mixture exhibited more increase in strength compared to the pure bentonite specimens. This indicates that the sand content of the mixture plays a vital role in taking the thermal stresses of the bentonite buffer in DGR conditions.

Keywords: bentonite, deep geological repository, thermal history, undrained shear strength

Procedia PDF Downloads 345

Authors: Ozgu Hafizoglu

Abstract:

Analogy is an essential tool of human cognition that enables connecting diffuse and diverse systems with attributional, deep structural, casual relations that are essential to learning, to innovation in artificial worlds, and to discovery in science. Cognitive Model of Analogy (CMA) leads and creates information pattern transfer within and between domains and disciplines in science. This paper demonstrates the Cognitive Model of Analogy (CMA) as an evolutionary approach to scientific research. The model puts forward the challenges of deep uncertainty about the future, emphasizing the need for flexibility of the system in order to enable reasoning methodology to adapt to changing conditions. In this paper, the model of analogical reasoning is created based on brain cells, their fractal, and operational forms within the system itself. Visualization techniques are used to show correspondences. Distinct phases of the problem-solving processes are divided thusly: encoding, mapping, inference, and response. The system is revealed relevant to brain activation considering each of these phases with an emphasis on achieving a better visualization of the brain cells: glial cells, axons, axon terminals, and neurons, relative to matching conditions of analogical reasoning and relational information. It’s found that encoding, mapping, inference, and response processes in four-term analogical reasoning are corresponding with the fractal and operational forms of brain cells: glial, axons, and neurons.

Keywords: analogy, analogical reasoning, cognitive model, brain and glials

Procedia PDF Downloads 186

Authors: Xu Jiaqiao

Abstract:

Text sentiment analysis is an important branch of natural language processing. This technology is widely used in public opinion analysis and web surfing recommendations. At present, the mainstream sentiment analysis methods include three parts: sentiment analysis based on a sentiment dictionary, based on traditional machine learning, and based on deep learning. This paper mainly analyzes and compares the advantages and disadvantages of the SVM method of traditional machine learning and the Long Short-term Memory (LSTM) method of deep learning in the field of Chinese sentiment analysis, using Chinese comments on Sina Microblog as the data set. Firstly, this paper classifies and adds labels to the original comment dataset obtained by the web crawler, and then uses Jieba word segmentation to classify the original dataset and remove stop words. After that, this paper extracts text feature vectors and builds document word vectors to facilitate the training of the model. Finally, SVM and LSTM models are trained respectively. After accuracy calculation, it can be obtained that the accuracy of the LSTM model is 85.80%, while the accuracy of SVM is 91.07%. But at the same time, LSTM operation only needs 2.57 seconds, SVM model needs 6.06 seconds. Therefore, this paper concludes that: compared with the SVM model, the LSTM model is worse in accuracy but faster in processing speed.

Keywords: sentiment analysis, support vector machine, long short-term memory, Chinese microblog comments

Procedia PDF Downloads 94

Authors: Hichem Sassi

Abstract:

Pneumonia, being a respiratory infection, has garnered global attention due to its rapid transmission and relatively high mortality rates. Timely detection and treatment play a crucial role in significantly reducing mortality associated with pneumonia. Presently, X-ray diagnosis stands out as a reasonably effective method. However, the manual scrutiny of a patient's X-ray chest radiograph by a proficient practitioner usually requires 5 to 15 minutes. In situations where cases are concentrated, this places immense pressure on clinicians for timely diagnosis. Relying solely on the visual acumen of imaging doctors proves to be inefficient, particularly given the low speed of manual analysis. Therefore, the integration of artificial intelligence into the clinical image diagnosis of pneumonia becomes imperative. Additionally, AI recognition is notably rapid, with convolutional neural networks (CNNs) demonstrating superior performance compared to human counterparts in image identification tasks. To conduct our study, we utilized a dataset comprising chest X-ray images obtained from Kaggle, encompassing a total of 5216 training images and 624 test images, categorized into two classes: normal and pneumonia. Employing five mainstream network algorithms, we undertook a comprehensive analysis to classify these diseases within the dataset, subsequently comparing the results. The integration of artificial intelligence, particularly through improved network architectures, stands as a transformative step towards more efficient and accurate clinical diagnoses across various medical domains.

Keywords: deep learning, computer vision, pneumonia, models, comparative study

Procedia PDF Downloads 64

Authors: Yongjian Gu

Abstract:

Ocean thermal energy is one of the ocean energy sources. It is a renewable, sustainable, and green energy source. Ocean thermal energy conversion (OTEC) applies the ocean temperature gradient between the warmer surface seawater and the cooler deep seawater to run a heat engine and produce a useful power output. Unfortunately, the ocean temperature gradient is not big. Even in the tropical and equatorial regions, the surface water temperature can only reach up to 28oC and the deep water temperature can be as low as 4oC. The thermal efficiency of the OTEC plants, therefore, is low. In order to improve the plant thermal efficiency by using the limited ocean temperature gradient, some OTEC plants use the method of adding more equipment for better heat recovery, such as heat exchangers, pumps, etc. Obviously, the method will increase the plant's complexity and cost. The more important impact of the method is the additional equipment needs to consume power too, which may have an adverse effect on the plant net power output, in turn, the plant thermal efficiency. In the paper, the author first describes varied OTEC plants and the practice of using the method of adding more equipment for improving the plant's thermal efficiency. Then the author proposes a parameter, plant back works ratio ϕ, for measuring if the added equipment is appropriate for the plant thermal efficiency improvement. Finally, in the paper, the author presents examples to illustrate the application of the back work ratio ϕ as a key parameter in the OTEC plant design and operation.

Keywords: ocean thermal energy, ocean thermal energy conversion (OTEC), OTEC plant, plant back work ratio ϕ

Procedia PDF Downloads 196

Authors: Jian Li

Abstract:

The uncertainty quantification and inversion problems of subsurface oil-water phase flow usually require extensive repeated forward calculations for new runs with changed conditions. To reduce the computational time, various forms of surrogate models have been built. Related research shows that deep learning has emerged as an effective surrogate model, while most surrogate models with deep learning are purely data-driven, which always leads to poor robustness and abnormal results. To guarantee the model more consistent with the physical laws, a coupled theory-guided convolutional neural network (TgCNN) based surrogate model is built to facilitate computation efficiency under the premise of satisfactory accuracy. The model is a convolutional neural network based on multi-well reservoir simulation. The core notion of this proposed method is to bridge two separate blocks on top of an overall network. They underlie the TgCNN model in a coupled form, which reflects the coupling nature of pressure and water saturation in the two-phase flow equation. The model is driven by not only labeled data but also scientific theories, including governing equations, stochastic parameterization, boundary, and initial conditions, well conditions, and expert knowledge. The results show that the TgCNN-based surrogate model exhibits satisfactory accuracy and efficiency in subsurface oil-water phase flow under multi-well conditions.

Keywords: coupled theory-guided convolutional neural network, multi-well conditions, surrogate model, subsurface oil-water phase

Procedia PDF Downloads 86

Authors: Bilal Wadood, Suleman Khan, Sajjad Ahmed

Abstract:

The Late Cretaceous strata (Mughal Kot Formation) exposed in Central Indus Basin, Pakistan is evaluated for establishing sequence stratigraphic framework and potential of hydrocarbon accumulation. The petrographic studies and SEM analysis were carried out to infer the hydrocarbon potential of the rock unit. The petrographic details disclosed 4 microfacies including Pelagic Mudstone, OrbitoidalWackestone, Quartz Arenite, and Quartz Wacke. The lowermost part of the rock unit consists of OrbitoidalWackestone which shows deposition in the middle shelf environment. The Quartz Arenite and Quartz Wacke suggest deposition on the deep slope settings while the Pelagic Mudstone microfacies point toward deposition in the distal deep marine settings. Based on the facies stacking patterns and cyclicity in the chronostratigraphic context, the strata is divided into two 3rd order cycles. One complete sequence i.e Transgressive system tract (TST), Highstand system tract (HST) and Lowstand system tract (LST) are again replaced by another Transgressive system tract and Highstant system tract with no markers of sequence boundary. The LST sands are sandwiched between TST and HST shales but no potential porosity/permeability values have been determined. Microfacies and SEM studies revealed very fewer chances for hydrocarbon accumulation and overall reservoir potential is characterized as low.

Keywords: cycle, deposition, microfacies, reservoir

Procedia PDF Downloads 150

Authors: Yazan Al-Kofahi, Jamal Alqawasmi.

Abstract:

In this study, a comprehensive literature review (SLR) was conducted, with the main goal of assessing the existing literature about how artificial intelligence (AI), machine learning (ML), deep learning (DL) models are used in sustainable architecture applications and issues including thermal comfort satisfaction, energy efficiency, cost prediction and many others issues. For this reason, the search strategy was initiated by using different databases, including Scopus, Springer and Google Scholar. The inclusion criteria were used by two research strings related to DL, ML and sustainable architecture. Moreover, the timeframe for the inclusion of the papers was open, even though most of the papers were conducted in the previous four years. As a paper filtration strategy, conferences and books were excluded from database search results. Using these inclusion and exclusion criteria, the search was conducted, and a sample of 59 papers was selected as the final included papers in the analysis. The data extraction phase was basically to extract the needed data from these papers, which were analyzed and correlated. The results of this SLR showed that there are many applications of ML and DL in Sustainable buildings, and that this topic is currently trendy. It was found that most of the papers focused their discussions on addressing Environmental Sustainability issues and factors using machine learning predictive models, with a particular emphasis on the use of Decision Tree algorithms. Moreover, it was found that the Random Forest repressor demonstrates strong performance across all feature selection groups in terms of cost prediction of the building as a machine-learning predictive model.

Keywords: machine learning, deep learning, artificial intelligence, sustainable building

Procedia PDF Downloads 67

Authors: Ole-Christian Galbo Engstrøm, Erik Schou Dreier, Birthe Møller Jespersen, Kim Steenstrup Pedersen

Abstract:

Grain quality analysis is a multi-parameterized problem that includes a variety of qualitative and quantitative parameters such as grain type classification, damage type classification, and nutrient regression. Currently, these parameters require human inspection, a multitude of instruments employing a variety of sensor technologies, and predictive model types or destructive and slow chemical analysis. This paper investigates the feasibility of applying near-infrared hyperspectral imaging (NIR-HSI) to grain quality analysis. For this study two datasets of NIR hyperspectral images in the wavelength range of 900 nm - 1700 nm have been used. Both datasets contain images of sparsely and densely packed grain kernels. The first dataset contains ~87,000 image crops of bulk wheat samples from 63 harvests where protein value has been determined by the FOSS Infratec NOVA which is the golden industry standard for protein content estimation in bulk samples of cereal grain. The second dataset consists of ~28,000 image crops of bulk grain kernels from seven different wheat varieties and a single rye variety. In the first dataset, protein regression analysis is the problem to solve while variety classification analysis is the problem to solve in the second dataset. Deep convolutional neural networks (CNNs) have the potential to utilize spatio-spectral correlations within a hyperspectral image to simultaneously estimate the qualitative and quantitative parameters. CNNs can autonomously derive meaningful representations of the input data reducing the need for advanced preprocessing techniques required for classical chemometric model types such as artificial neural networks (ANNs) and partial least-squares regression (PLS-R). A comparison between different CNN architectures utilizing 2D and 3D convolution is conducted. These results are compared to the performance of ANNs and PLS-R. Additionally, a variety of preprocessing techniques from image analysis and chemometrics are tested. These include centering, scaling, standard normal variate (SNV), Savitzky-Golay (SG) filtering, and detrending. The results indicate that the combination of NIR-HSI and CNNs has the potential to be the foundation for an automatic system unifying qualitative and quantitative grain quality analysis within a single sensor technology and predictive model type.

Keywords: deep learning, grain analysis, hyperspectral imaging, preprocessing techniques

Procedia PDF Downloads 99

Authors: Mirjana Ruppel, Rajendra Persad, Amit Bahl, Sanja Dogramadzi, Chris Melhuish, Lyndon Smith

Abstract:

Multimodal image registration is a profoundly complex task which is why deep learning has been used widely to address it in recent years. However, two main challenges remain: Firstly, the lack of ground truth data calls for an unsupervised learning approach, which leads to the second challenge of defining a feasible loss function that can compare two images of different modalities to judge their level of alignment. To avoid this issue altogether we implement a generative adversarial network consisting of two registration networks GAB, GBA and two discrimination networks DA, DB connected by spatial transformation layers. GAB learns to generate a deformation field which registers an image of the modality B to an image of the modality A. To do that, it uses the feedback of the discriminator DB which is learning to judge the quality of alignment of the registered image B. GBA and DA learn a mapping from modality A to modality B. Additionally, a cycle-consistency loss is implemented. For this, both registration networks are employed twice, therefore resulting in images ˆA, ˆB which were registered to ˜B, ˜A which were registered to the initial image pair A, B. Thus the resulting and initial images of the same modality can be easily compared. A dataset of liver CT and MRI was used to evaluate the quality of our approach and to compare it against learning and non-learning based registration algorithms. Our approach leads to dice scores of up to 0.80 ± 0.01 and is therefore comparable to and slightly more successful than algorithms like SimpleElastix and VoxelMorph.

Keywords: cycle consistency, deformable multimodal image registration, deep learning, GAN

Procedia PDF Downloads 131

Authors: Fazıl Gökgöz, Fahrettin Filiz

Abstract:

Electricity prices have sophisticated features such as high volatility, nonlinearity and high frequency that make forecasting quite difficult. Electricity price has a volatile and non-random character so that, it is possible to identify the patterns based on the historical data. Intelligent decision-making requires accurate price forecasting for market traders, retailers, and generation companies. So far, many shallow-ANN (artificial neural networks) models have been published in the literature and showed adequate forecasting results. During the last years, neural networks with many hidden layers, which are referred to as DNN (deep neural networks) have been using in the machine learning community. The goal of this study is to investigate electricity price forecasting performance of the shallow-ANN and DNN models for the Turkish day-ahead electricity market. The forecasting accuracy of the models has been evaluated with publicly available data from the Turkish day-ahead electricity market. Both shallow-ANN and DNN approach would give successful result in forecasting problems. Historical load, price and weather temperature data are used as the input variables for the models. The data set includes power consumption measurements gathered between January 2016 and December 2017 with one-hour resolution. In this regard, forecasting studies have been carried out comparatively with shallow-ANN and DNN models for Turkish electricity markets in the related time period. The main contribution of this study is the investigation of different shallow-ANN and DNN models in the field of electricity price forecast. All models are compared regarding their MAE (Mean Absolute Error) and MSE (Mean Square) results. DNN models give better forecasting performance compare to shallow-ANN. Best five MAE results for DNN models are 0.346, 0.372, 0.392, 0,402 and 0.409.

Keywords: deep learning, artificial neural networks, energy price forecasting, turkey

Procedia PDF Downloads 292

Authors: Dongping Fang, Lian Duan, Xiaojing Yuan, Mike Xu, Allyn Klunder, Kevin Tan, Suiting Cao, Yeqing Ji

Abstract:

Accurate prediction of a medical condition with straight clinical evidence is a long-sought topic in the medical management and health insurance field. Although great progress has been made with machine learning algorithms, the medical community is still, to a certain degree, suspicious about the model's accuracy and interpretability. This paper presents an innovative hierarchical attention deep learning model to achieve good prediction and clear interpretability that can be easily understood by medical professionals. This deep learning model uses a hierarchical attention structure that matches naturally with the medical history data structure and reflects the member’s encounter (date of service) sequence. The model attention structure consists of 3 levels: (1) attention on the medical code types (diagnosis codes, procedure codes, lab test results, and prescription drugs), (2) attention on the sequential medical encounters within a type, (3) attention on the medical codes within an encounter and type. This model is applied to predict the occurrence of stage 3 chronic kidney disease (CKD3), using three years’ medical history of Medicare Advantage (MA) members from a top health insurance company. The model takes members’ medical events, both claims and electronic medical record (EMR) data, as input, makes a prediction of CKD3 and calculates the contribution from individual events to the predicted outcome. The model outcome can be easily explained with the clinical evidence identified by the model algorithm. Here are examples: Member A had 36 medical encounters in the past three years: multiple office visits, lab tests and medications. The model predicts member A has a high risk of CKD3 with the following well-contributed clinical events - multiple high ‘Creatinine in Serum or Plasma’ tests and multiple low kidneys functioning ‘Glomerular filtration rate’ tests. Among the abnormal lab tests, more recent results contributed more to the prediction. The model also indicates regular office visits, no abnormal findings of medical examinations, and taking proper medications decreased the CKD3 risk. Member B had 104 medical encounters in the past 3 years and was predicted to have a low risk of CKD3, because the model didn’t identify diagnoses, procedures, or medications related to kidney disease, and many lab test results, including ‘Glomerular filtration rate’ were within the normal range. The model accurately predicts members A and B and provides interpretable clinical evidence that is validated by clinicians. Without extra effort, the interpretation is generated directly from the model and presented together with the occurrence date. Our model uses the medical data in its most raw format without any further data aggregation, transformation, or mapping. This greatly simplifies the data preparation process, mitigates the chance for error and eliminates post-modeling work needed for traditional model explanation. To our knowledge, this is the first paper on an interpretable deep-learning model using a 3-level attention structure, sourcing both EMR and claim data, including all 4 types of medical data, on the entire Medicare population of a big insurance company, and more importantly, directly generating model interpretation to support user decision. In the future, we plan to enrich the model input by adding patients’ demographics and information from free-texted physician notes.

Keywords: deep learning, interpretability, attention, big data, medical conditions

Procedia PDF Downloads 91

Authors: Ja-Keoung Koo, Kensuke Nakamura, Hyohun Kim, Dongwha Shin, Yeonseok Kim, Ji-Su Ahn, Byung-Woo Hong

Abstract:

The machine learning techniques based on a convolutional neural network (CNN) have been actively developed and successfully applied to a variety of image analysis tasks including reconstruction, noise reduction, resolution enhancement, segmentation, motion estimation, object recognition. The classical visual information processing that ranges from low level tasks to high level ones has been widely developed in the deep learning framework. It is generally considered as a challenging problem to derive visual interpretation from high dimensional imagery data. A CNN is a class of feed-forward artificial neural network that usually consists of deep layers the connections of which are established by a series of non-linear operations. The CNN architecture is known to be shift invariant due to its shared weights and translation invariance characteristics. However, it is often computationally intractable to optimize the network in particular with a large number of convolution layers due to a large number of unknowns to be optimized with respect to the training set that is generally required to be large enough to effectively generalize the model under consideration. It is also necessary to limit the size of convolution kernels due to the computational expense despite of the recent development of effective parallel processing machinery, which leads to the use of the constantly small size of the convolution kernels throughout the deep CNN architecture. However, it is often desired to consider different scales in the analysis of visual features at different layers in the network. Thus, we propose a CNN model where different sizes of the convolution kernels are applied at each layer based on the random projection. We apply random filters with varying sizes and associate the filter responses with scalar weights that correspond to the standard deviation of the random filters. We are allowed to use large number of random filters with the cost of one scalar unknown for each filter. The computational cost in the back-propagation procedure does not increase with the larger size of the filters even though the additional computational cost is required in the computation of convolution in the feed-forward procedure. The use of random kernels with varying sizes allows to effectively analyze image features at multiple scales leading to a better generalization. The robustness and effectiveness of the proposed CNN based on random kernels are demonstrated by numerical experiments where the quantitative comparison of the well-known CNN architectures and our models that simply replace the convolution kernels with the random filters is performed. The experimental results indicate that our model achieves better performance with less number of unknown weights. The proposed algorithm has a high potential in the application of a variety of visual tasks based on the CNN framework. Acknowledgement—This work was supported by the MISP (Ministry of Science and ICT), Korea, under the National Program for Excellence in SW (20170001000011001) supervised by IITP, and NRF-2014R1A2A1A11051941, NRF2017R1A2B4006023.

Keywords: deep learning, convolutional neural network, random kernel, random projection, dimensionality reduction, object recognition

Procedia PDF Downloads 290