The State Final Examination
The State Final Examination is governed by the CTU Study and Examination Regulations and the Directive of the Dean of the Faculty of Electrical Engineering. The first part consists of the presentation of the master’s thesis, familiarization with the thesis assessment reports, the student’s responses to the comments made in the reports and to questions from the examination committee, followed by a discussion of the submitted thesis and its presentation. In the second part of the State Final Examination, the student answers two questions selected by the examination committee from the following list of thematic areas.
- Empirical Risk Minimization. Generalization theory. PAC learning. Bias-Variance trade-off.
- Model selection and validation methods. Performance metrics.
- Linear models. Support Vector Machines. Kernel methods. Generative learning. Bayesian learning.
- Learning from Tabular data. Ensembling. Model interpretability.
- Learning from Structured data. Graph Neural Networks. Neural-Symbolic methods.
- Causality. Bandit algorithms.
- Markov decision processes. Reinforcement learning.
- Markov decision processes. Reinforcement learning.
- Neural networks. Auto-differentiation. Gradient learning and optimization issues.
- Convolutional and recurrent networks. Generative models. Applications of deep nets in classification, segmentation, detection, regression, reinforcement learning.
AI and Society
- Robustness of ML models. Social choice theory.
- Societal issues of ML/AI: fairness, bias, privacy, safety and security. Existential risks. Vulnerabilities of ML/AI.
- ML/AI policies and regulatory approaches. Ethics of ML/AI development. Misuseability of generative models.
- Cybersecurity.
- Basics of security principles and network protocols. Vulnerability types, Social Engineering, Exploiting.
- Detection of intruders in computers. Hardening. Host based IDS. Network security analysis of attacks.
- Sandboxing and Virtualization. Threat Intelligence. Improve with AI.
- Honeypots. AI for Honeypots. Side-channel attacks. Persistence. Privilege Escalation. Hiding activities.
- Binary exploitation. Stack buffer overflow and defenses. Reverse Engineering. Assistance with AI.
- Automatic attacks with Malware and AI. C&C.
- Steganography. Manual and AI detection of C&C channels.
- Basics of cryptography, Symmetric and Asymmetric encryption.
- Web attacks, SQL injection.
- Attacking AI systems.
Natural Language Processing
- Linguistic layers of language description and their relation to machine translation (morphology, syntax of natural languages).
- Machine translation evaluation. Manual and automatic metrics of quality.
- Alignment of documents, sentences and words in texts.
- Statistical machine translation: Pre-neural and neural approaches.
- Text processing with neural networks (subword units, word embeddings).
- Transformers in machine translation. Word and sentence representations.
- Multi-lingual and multi-modal machine translation.
- Large language models.
Computer Vision
- Correspondences and wide baseline stereo. Harris detector. Laplace operator. SIFT descriptors.3D reconstruction.
- Deep learning in computer vision. Architectures for image recognition, object detection, and semantic segmentation. Foundation models.
- Tracking. Image Retrieval. Self-supervised representation learning.
