Pan-African Conference on Artificial Intelligence: First Conference, PanAfriCon AI 2022, Addis Ababa, Ethiopia, October 4–5, 2022, Revised Selected ... in Computer and Information Science, 1800)

This volume consitutes selected papers presented during the First Pan-African Conference on Artificial Intelligence, PanAfriCon AI 2022, held in Addis Ababa, Ethiopia, in October 2022. The 16 papers were thoroughly reviewed and selected from the 49 qualified submissions. The papers are organized in the following topical sections: ​AI in public health; agriculture; algorithmic optimization; human-machine interaction; economy and security. Preface Address of the Prime Minister of the Federal Democratic Republic of Ethiopia Address of the Head of Cyber Security, United Arab Emirates Government Address of the Director General EAII Address of the Core Organizing Committee Member Organization Contents AI in Public Health Robust Cough Analysis System for Diagnosis of Tuberculosis Using Artificial Neural Network 1 Introduction 1.1 Overview of Tuberculosis 1.2 Diagnosis Using Cough Sound 2 Datasets Preparation 3 Cough Detection and Classification Methods 3.1 Cough Detection Method 3.2 Background Noise and Silence Removal 3.3 Amplitude Normalization 3.4 Pre-emphasis and Segmentation 3.5 Feature Extraction 3.6 Cough Detection Learning Algorithms 3.7 Cough Classification 4 Implementation Result and Discussion 4.1 The Datasets Preparation Result and Discussion 4.2 Result and Discussion on Pre-processing Phases 4.3 Results of Feature Extraction Process 4.4 Hyper-parameters Optimization of the Models 4.5 Performance Comparison of the Selected ANN and SVM Models 4.6 Research Contributions 5 Conclusion and Recommendation References A Multi-input Architecture for the Classification of Skin Lesions Using ResNets and Metadata 1 Introduction 2 Related Works 3 Ethical Approval 4 Dataset 5 Methodology 5.1 Image Pre-processing 5.2 ResNet Image Feature Extractor Architecture 5.3 CNN Metadata Feature Extractor Architecture 5.4 Representing Tabular Metadata Feature Vectors as Images 6 Experimental Results 7 Conclusion 8 Future Work References Data Management Strategy for AI Deployment in Ethiopian Healthcare System 1 Introduction 1.1 Data Management 1.2 Data Management and AI Applications 2 Data Management in Ethiopian Healthcare 3 Study Design 4 Results and Discussion 5 Conclusion References AI-Based Heart Disease and Brain Stroke Prediction Using Multi-modal Patient Data 1 Introduction 2 Methods 2.1 Heart Disease Prediction Model 2.2 Brain Stroke Prediction Using Patient Information and Laboratory Data 2.3 Brain Stroke Prediction Using MRI Image Data 3 Results 3.1 Heart Disease Prediction 3.2 Brain Stroke Prediction Using Clinical Data 3.3 Brain Stroke Prediction Using MRI Brain Image Data 4 Discussion 5 Conclusion A Appendix: User Interface References Lung Tumor Detection and Recognition Using Deep Convolutional Neural Networks 1 Introduction 2 Background and Literature Review 2.1 Deep Learning 2.2 Tumors 2.3 Related Work 3 Specifications and Experimental Methods 3.1 Datasets 3.2 Data Augmentation 3.3 Models 4 Studies 4.1 Study 1: Transfer Learning 4.2 Study 2: Transfer Learning on the Advanced Dataset 4.3 Study 3: Majority Voting Ensemble Learning 4.4 Study 4: Soft Voting Ensemble Learning 5 Results and Discussions 5.1 Study 1: Transfer Learning 5.2 Study 2: Transfer Learning on Advanced dataset 5.3 Study 3: Majority Voting Ensemble Learning 5.4 Study 4: Soft Voting Ensemble Learning 5.5 Discussions 6 Conclusions References Agriculture Tomato Leaf Disease Detection and Classification Using Custom Modified AlexNet*-4pt 1 Introduction 2 Related Works 3 Materials and Methods 3.1 Dataset 3.2 Methodology 4 Results and Discussions 4.1 Results for Different Experimental Setup 4.2 Evaluation of Experimental Results 5 Conclusion References Wheat Yield Prediction Using Machine Learning: A Survey 1 Introduction 2 Related Works 3 Methods 3.1 Research Questions 3.2 Search Strategies 4 Machine Learning in Wheat Yield Prediction 4.1 Remote Sensing Based Wheat Yield Prediction 4.2 Environmental Factors Based Wheat Yield Prediction 4.3 Genomic and Phenology Based Wheat Yield Prediction 5 Discussion 6 Conclusion References Algorithmic Optimization Small Training Datasets for Deep Learning Based Medical Diagnosis 1 Introduction 2 State-of-the-Art 2.1 Plasmodium Variants and Classification Task 2.2 Limits of Classical Malaria Diagnosis 2.3 Malaria CAD 2.4 Binary Classification 2.5 Multi-stage Classification 3 Malaria Datasets 3.1 Training, Validation and Test Dataset 3.2 Thin Blood Smear Dataset 4 Data Preprocessing 4.1 General Preprocessing Steps 4.2 Data Augmentation by Zooming 5 Deep Learning Models 5.1 Reference Model Architectures 5.2 Improved Model Architecture 5.3 Transfer Learning Architecture 6 Neural Network Training 6.1 Training Environment 6.2 Hyperparameters 6.3 Training Sessions 6.4 Duration of Training 7 Results 7.1 Comparison of the Training Sessions 7.2 Optimum Hyperparameters 7.3 Comparison of the FP Rates at a Fixed TP Rate 8 Application 8.1 Application on Thick Blood Smear Images 8.2 Application on Thin Blood Smear Images 9 Conclusions References Using Generative Adversarial Networks for Single Image Super-Resolution 1 Introduction 1.1 Related Work 1.2 Contribution 2 Methods 2.1 Generator and Discriminator Networks 2.2 Loss Function 2.3 Training Details 2.4 Training Experiments 2.5 Testing Details 3 Results and Analysis 3.1 Investigating the Perceptual Loss 3.2 Investigating the Performance of Final Networks 4 Conclusion References Model Compression Techniques in Deep Neural Networks 1 Introduction 2 Model Compression Methods 2.1 Generic Paradigms 2.2 Pruning 2.3 Knowledge Distillation 2.4 Quantization 2.5 Summary of Model Compression Against Generic Paradigms 2.6 Other Compression Methods 3 Hybrid Model Compression Methods 3.1 Pruning and Quantization 3.2 Pruning and Knowledge Distillation 3.3 Quantization and Knowledge Distillation 4 Model Compression Beyond Size Reduction 4.1 Reducing Overfitting 4.2 Explainability 4.3 Neural Architecture Search 4.4 Algorithmic Fairness 4.5 Security 5 Findings 5.1 Pruning 5.2 Knowledge Distillation 5.3 Quantization 6 Factors Affecting Choice of Methods 7 Open Research Problems 7.1 Knowledge Distillation (KD) 7.2 Hybrid Methods 7.3 Model Compression Beyond Size Reduction 7.4 Miscellaneous 8 Conclusion References Convolution Filter Equivariance/Invariance in Convolutional Neural Networks: A Survey 1 Introduction 2 Related Work 3 Method 4 Symmetry Transformations and Equivariance 4.1 Group Theory 4.2 Equivariance 5 Equivariant Convolutional Neural Networks 5.1 Translation Equivariance 5.2 Rotation, Reflection and Scale Equivariance 6 Application 7 Summary 8 Conclusion References Human-Machine Interaction Deep Learning Models for Audio Processing Applications Under Resource-Constrained Devices: A Survey 1 Introduction 2 Deep Learning Architectures 2.1 Deep Feed-Forward Neural Network (DFFNN) 2.2 Deep Recurrent Neural Networks 2.3 Convolutional Neural Networks (CNN) 2.4 Autoencoders 3 Performance of Squeezed DL Models in Resource-Constrained Devices 3.1 Comparison of Classical Versus Squeezed Deep Feedforward Models 3.2 Comparison of Classical Versus Squeezed Deep Restricted Boltzmann Models 3.3 Comparison of Classical Versus Compressed RNN Models 3.4 Comparison of Classical Versus Squeezed CNN Models 4 Challenges in Deploying DL Models on Resource-Constrained Devices 4.1 On-Device Learning 4.2 Limited Dataset Availability 4.3 Robustness 4.4 Interpretability 5 Conclusion References Offline Handwritten Amharic Character Recognition Using Few-Shot Learning 1 Introduction 2 Related Work 3 Methodology 3.1 Dataset Preparation 3.2 Prototypical Networks 4 Results and Discussion 5 Conclusion References Morpheme Based Amharic-Kistanigna Bi-directional Machine Translation Using Deep Learning 1 Introduction 2 The Nature of the Languages 3 Related Works 4 Methods 4.1 Preparing Parallel Corpus 4.2 Preprocessing 4.3 One Hot Vector Representation 4.4 Embedding 4.5 Encoder-Decoder Model 5 Experiments and Results 5.1 Experimental Setup 5.2 Results of the Experiment 5.3 Discussions 6 Conclusion References Economy and Security Benefits and Challenges of Industry 4.0 in African Emerging Economies 1 Introduction 2 Background 2.1 Industry 4.0 Key Technologies 2.2 Key Factors for the Realization and National Preparedness of Industry 4.0 2.3 Benefits of Industry 4.0 2.4 Challenges of Industry 4.0 2.5 Industry 4.0 in Africa 3 Case Study - Analysis of the Challenges and Benefits of Implementing Industry 4.0 3.1 Manufacturing Capability 3.2 Research and Development 3.3 Human Capital 3.4 Information Technology Infrastructure 4 Conclusions References Classification and Detection of Prohibited Objects in X-Ray Baggage Security Images 1 Introduction 2 Related Work 3 Methodology 3.1 Data Gathering and Understanding 3.2 Data Preparation and Preprocess 3.3 Model Building 3.4 Image Enhancement and Model Operationalize 3.5 Model Evaluation Parameter 4 Experimental Result and Discussion 5 Conclusion References Author Index