Deep Learning Applications

Description

This book presents a compilation of selected papers from the 17th IEEE International Conference on Machine Learning and Applications (IEEE ICMLA 2018), focusing on use of deep learning technology in application like game playing, medical applications, video analytics, regression/classification, object detection/recognition and robotic control in industrial environments. It highlights novel ways of using deep neural networks to solve real-world problems, and also offers insights into deep learning architectures and algorithms, making it an essential reference guide for academic researchers, professionals, software engineers in industry, and innovative product developers. Preface Contents About the Editors Trends in Deep Learning Applications 1 Introduction 2 Deep Learning in Game Playing 3 Deep Learning in Medical Applications 4 Deep Learning in Video Analytics 5 Deep Learning in Regression and Classification 6 Deep Learning in Object Detection and Recognition 7 Deep Learning in Robotic Automation Reference Quasi-Newton Optimization Methods for Deep Learning Applications 1 Introduction 1.1 Existing Methods 1.2 Motivation 1.3 Applications and Objectives 1.4 Chapter Outline 2 Unconstrained Optimization Problem 3 Optimization Strategies 3.1 Line Search Methods 3.2 Trust-Region Methods 4 Quasi-Newton Optimization Methods 4.1 The BFGS Update 4.2 Line Search L-BFGS Optimization 4.3 Trust-Region Subproblem Solution 5 Application to Image Recognition 5.1 LeNet-5 Convolutional Neural Network Architecture 5.2 MNIST Image Classification Task 5.3 Results 6 Application to Deep Reinforcement Learning 6.1 Reinforcement Learning Problem 6.2 Empirical Risk Minimization in Deep Reinforcement Learning 6.3 L-BFGS Line Search Deep Q-Learning Method 6.4 Convergence Analysis 6.5 Convergence for the Empirical Risk 6.6 Value Optimality 6.7 Computation Time 6.8 Experiments on Atari 2600 Games 6.9 Results and Discussion 7 Conclusions References Medical Image Segmentation Using Deep Neural Networks with Pre-trained Encoders 1 Introduction 2 Network Architectures and Training 3 Angiodysplasia Lesion Segmentation in Endoscopic Videos 3.1 Background 3.2 Dataset Description and Preprocessing 3.3 Results 4 Robotic Instrument Segmentation in Surgical Videos 4.1 Background 4.2 Dataset Description and Preprocessing 4.3 Results 5 Conclusions References Ensemble of 3D Densely Connected Convolutional Network for Diagnosis of Mild Cognitive Impairment and Alzheimer's Disease 1 Introduction 2 Related Work 2.1 Deep Learning for Computer-Aided Diagnosis 2.2 Automatic Recognition of Alzheimer's Disease and Mild Cognitive Impairment 3 Methods 3.1 Data Acquisition and Preprocessing 3.2 Proposed Method 4 Experiments 4.1 Data and Implementation 4.2 Experimental Steps and Evaluation 4.3 Parametric Analyses 4.4 Results 5 Conclusion References Detecting Work Zones in SHRP2 NDS Videos Using Deep Learning Based Computer Vision 1 Introduction 2 Related Work 3 Deep Learning Framework and Architecture Selection 3.1 Deep Learning Framework Selection 3.2 CNN Architecture Selection 4 Data Set Construction and Model Development 4.1 Data Source Selection 4.2 Active Learning via Uncertainty Sampling 5 SNVA Application Design and Development 5.1 Core Software Components 5.2 Video Frame Timestamp Extraction 5.3 Input Pipeline Optimization 5.4 Software Development Environment 5.5 Production SNVA Environment 6 Future Work 6.1 Precision-Oriented Active Learning 6.2 Robust Multi-frame Event Detection Using Bidirectional Recurrent Neural Networks 6.3 Other High-Priority Target Scene Features 7 Conclusion References Action Recognition in Videos Using Multi-stream Convolutional Neural Networks 1 Introduction 2 Related Work 3 Basic Concepts 3.1 Visual Rhythm 3.2 Two-Stream Architecture 4 Proposed Method 4.1 Improved Spatial Stream 4.2 Temporal Stream 4.3 Spatiotemporal Stream 4.4 Stacking 5 Experimental Results 5.1 Datasets 5.2 Results 6 Conclusions References Deep Active Learning for Image Regression 1 Introduction 2 Related Work 2.1 Deep Learning for Regression 2.2 Active Learning for Regression 3 Proposed Framework 3.1 Loss on Labeled Data 3.2 Principle of Expected Model Output Change (EMOC) 3.3 Loss on Unlabeled Data 3.4 Novel Joint Objective Function 3.5 Gradient of the Objective Function 4 Experiments and Results 4.1 Implementation Details 4.2 Datasets and Experimental Setup 4.3 Comparison Baselines and Evaluation Metrics 4.4 Active Learning Performance 4.5 Study of the Active Sampling Criterion 4.6 Study of Number of Active Learning Iterations 5 Conclusion and Future Work References Deep Learning Based Hazard Label Object Detection for Lithium-ion Batteries Using Synthetic and Real Data 1 Introduction 2 Related Work 2.1 Synthetic Data 2.2 Transfer Learning 3 Methodology 3.1 YOLO 3.2 MSER-ODP 4 Dataset Creation 4.1 Generation and Distribution of the Bounding Box Annotated Real Dataset 4.2 Generation and Distribution of the Real Training Set 4.3 Generation and Distribution of the Mixed Training Set 5 Performance Metric 6 Architecture of the CNN Models 6.1 Tiny YOLO and YOLOv2 Model Architectures 6.2 Inception V3 Model Architecture 7 Configuration and Training of the Object Detection Systems 7.1 Configuration and Training of the MSER-ODP 7.2 Configuration and Training of the YOLO Approach 8 Questions and Results 9 Conclusion References Enabling Robust and Autonomous Materialhandling in Logistics Through Applied Deep Learning Algorithms 1 Introduction 1.1 Evolving Automatisation 2 Logistics 2.1 Intralogistics 2.2 Objects 3 Perception Algorithm 3.1 Basic Concept 3.2 Module 1: Detection 3.3 Module 2: Selection 3.4 Module 3: Localization 4 Conclusion References Author Index