Artificial Inteligence

AI - Overview, paradigms, use-cases and limitations: AI as a system, current capabilities and shortcomings, typical failure modes (hallucinations), success metrics, costs and risks, domain adaptation, and AI pipeline mapping.

Trustworthy, Interpretable & Explainable AI: Ante-hoc vs. post-hoc (SHAP/LIME/counterfactuals), calibration/uncertainty, fairness/bias, auditability, operational quality monitoring.

Reinforcement Learning I – Fundamentals
Value-based (Q-learning, DQN), policy-based, actor-critic; exploration, reward shaping, learning stability, MDP/POMDP.

Reinforcement Learning II – Advanced and Applications
Offline RL, safe RL, multi-agent systems, simulation and policy evaluation; applications in recommendation and control.

Overview and Applied Comparison of Generative Models: VAE, GAN, Flow, Diffusion
Model families, comparison and applicability, multimodal generation and synthetic data, security and evaluation.

LLM vs. SLM: Comparison, architectures, scaling and parameters; hallucinations, factuality and robustness. Decision-making framework for LLM/SLM – model selection and operational robustness (quality/latency/cost/privacy), context window vs. RAG, caching; hallucination typology, factuality/groundedness, and mitigation at the model and decoding levels.

RAG vs. Fine-tuning in Practice - Design Patterns and Evaluation: Indexing (chunking, windows, compression), benchmarking RAG, retrieval methods, re-ranking, context routing; metrics for faithfulness/groundedness (system-centric view); security (content filters, prompt-injection guards).

Agent-based Approaches and Tool-use (Model Context Protocol): Goal decomposition, orchestration of multiple roles, task planning, memory, tool/API calls, self-critique; function/tool calling; MCP – standardized secure tool integration and audit logs; safety boundaries.

Privacy and Collaboration: Federated Learning & Privacy-preserving AI
Federated AI frameworks, aggregation and scaling, differential privacy, local RAG and hybrid querying, on-premise/edge AI.

Efficient Inference Engineering, Production Prompting, and In-Context Learning (ICL): Zero-/few-shot and production deployment; designing safe and measurable prompting/ICL; accelerating inference without critical quality loss – quantization, distillation, pruning, speculative decoding, caching; programmatic prompting, prompt security (prompt-injection/jailbreak), ICL selection and A/B evaluation.

Neuro-symbolic AI and Knowledge Systems (KG – Knowledge Graphs): Linking LLMs with knowledge bases for accuracy and auditability, RDF/OWL, SPARQL, rules and reasoning, probabilistic extensions; LLM+KG, KG-RAG (verification, citation).

AutoML, Meta-learning, and Reproducible Evaluation: HPO, configuration and portfolio selection, meta-learning; reproducibility principles, data protocols and benchmarking across domains, synthetic data/simulation; experiment tracking, drift detection; protocols and reproducibility.