Artificial Intelligence for Industrial Applications

Lectures:
1. Introduction to Industrial AI & Project Lifecycle: Industry vs. Academia requirements. Data & Model Versioning. Project structure and reproducibility.
2. Data-Centric AI: Importance of data quality. Techniques for data acquisition, cleaning, and annotation. Dataset balance and bias.
3. Fundamentals of Machine Learning: From statistics to neural networks. Decision Trees, SVM, and classifier evaluation basics (Confusion Matrix, Precision/Recall).
4. Deep Learning for 1D Signals: Time-series processing. Autoencoders for Anomaly Detection (Unsupervised Learning). Spectral analysis and feature extraction.
5. Computer Vision Fundamentals (2D): Image representation. Preprocessing and the role of Data Augmentation for generalization. Batching strategies.
6. Convolutional Neural Networks (CNN) & Classification: Architectures (ResNet, EfficientNet). Convolution and pooling principles. Transfer Learning and Fine-tuning in practice.
7. Semantic Segmentation & Object Detection: U-Net and YOLO architectures. Loss functions for segmentation (Dice Loss, IoU). Detection vs. Instance vs. Semantic Segmentation.
8. 3D Deep Learning & Volumetric Data: 3D data specifics (Voxel vs. Point Cloud). DICOM/NIfTI formats. The nnU-Net framework and automated network configuration.
9. Generative AI & LLMs in Industry: Transformer architecture and Attention mechanism. Running Local LLMs (Privacy & Security). Prompt Engineering for automation.
10. Validation & Inference: Real-world performance metrics. Overfitting vs. Underfitting. Strategies for evaluating model robustness on unseen data.
11. Neural Network Optimization: Reducing computational cost. Quantization (INT8 vs FP16) and Pruning. Hardware-aware optimization (TensorRT).
12. MLOps & Containerization: Docker in ML context. Dependency management. CI/CD principles for automated retraining and deployment.
13. Edge Computing: Hardware specifics (NVIDIA Jetson, FPGA). Energy efficiency and latency. Inference application architecture. Ethics, Legislation & Future of AI: The AI Act and EU regulations. Liability for AI decisions. Trends in industrial automation.

Laboratories:
1. Environment setup. Preparation of a Git project for data versioning. Preparation of a basic pipeline. Working with annotation tools.
2. Data preparation and preprocessing. Exploratory data analysis. Data cleaning methods.
3. Machine learning methods. Classifiers.
4. Training neural networks on 1D data. Feature extraction. Architectures for anomaly detection.
5. Preprocessing methods for 2D data. Data augmentation. Data loaders. Annotations and working with them.
6. Training architectures on 2D data. Transfer learning. Architectures for classification.
7. Training architectures on 2D data. Architectures for segmentation. Binary and multi-class segmentation.
8. Preparation of 3D data. Data normalization. Resampling. Visualization of slices. Training the nnU-Net architecture for 3D segmentation.
9. Setup and preparation of local LLMs. Installation. Runtime optimization of local language models. Integration into Python scripts.
10. Preparation of a robust inference pipeline. Data preparation. Evaluation with a trained model. Processing of outputs.
11. Optimization of models for inference. Model optimization methods. Network pruning. Utilization and comparison for individual accelerators.
12. Containerization of applications. Use of MLOps methods. Deployment of containers on different hardware.
13. Pipeline deployment. Preparation for production. Edge computing. Project consultations.