Ročníky

Předchozí přednášky

• Storm Visser:

  Artificial Immune Systems: Multiclass Classification and Hybrid Methods

  15.10.2026 14:00Posluchárna S8, Matematicko-fyzikální fakulta, Malostranské náměstí 2

  Storm will explain the basics of Artificial Immune Systems, discuss some of the recent advances in the field, including multiclass and fine-grained classification, as well as potential future work

• Isam Vrce:

  Reinforcement Learning Built for Acceleration: Sparse Topology Control, Stability, and Hardware-Aware Training

  01.10.2026 14:00Posluchárna S8, Matematicko-fyzikální fakulta, Malostranské náměstí 2

  Isam will explore how hardware-aware semi-structured sparsity can make deep reinforcement learning more computationally efficient while preserving final performance, using sparse training from scratch, sparse topology adaptation, and replay-aware mask updates

• Noe Lallouet:

  Multi-objective Neural Architecture Design Using Monte Carlo Search

  14.05.2026 14:00Posluchárna S8, Matematicko-fyzikální fakulta, Malostranské náměstí 2

  In this talk, Noe investigates the use of Monte Carlo search for building neural network models satisfying multiple, possibly contradictory, objectives.

• Dario Vajda:

  Scaling LLM Alignment Beyond Supervised Fine-Tuning: Solving the Data Reliance Bottleneck

  30.04.2026 14:00Posluchárna S8, Matematicko-fyzikální fakulta, Malostranské náměstí 2

  After reviewing LLM fundamentals and the limitations of Supervised Fine-Tuning, Dario explores how advanced RL-based alignment techniques like RLHF and DeepSeek's GRPO thrive in verifiable environments, culminating in his research demonstrating how Direct Preference Optimization outperforms classic supervised translation models.

• Benjamin Moore:

  Automated and Robust Shape Optimisation of Industrial Wastewater Systems Using Multi-Objective Bayesian Methods

  16.04.2026 14:00Posluchárna S8, Matematicko-fyzikální fakulta, Malostranské náměstí 2

  Ben will show how modern machine learning algorithms can be coupled with CFD to enable the efficient multi-objective optimisation of complex industrial systems, and how optimisers can be built to work with noisy data, infeasible designs, and a limited computational budget.

• Marcel Salvenmoser:

  Evolving Prompt Embeddings of Diffusion Models for Optimization and Exploration of Synthetic Images

  02.04.2026 14:00Posluchárna S8, Matematicko-fyzikální fakulta, Malostranské náměstí 2

  Marcel will show how to apply evolutionary algorithms to diffusion model conditioning to both explore diverse generated images via style mixing and to optimize the results using aesthetic, image-quality, and AI-detection scoring functions.

• David Moe:

  Conformal Predictors meet Active Learning: Adaptive Systems for Efficient Industrial Monitoring in Non-Stationary Environments

  19.03.2026 14:00Posluchárna S8, Matematicko-fyzikální fakulta, Malostranské náměstí 2

  David will discuss how conformal predictors and active learning can be integrated into uncertainty-aware adaptive systems that maintain reliability under distribution shift, enabling efficient and scalable monitoring in real industrial environments.

• Ondřej Maršálek and Pavol Šimko:

  Machine-learned interatomic potentials for molecular dynamics: committees and foundation models

  19.02.2026 14:00Posluchárna S8, Matematicko-fyzikální fakulta, Malostranské náměstí 2

  Ond?ej will give an introduction to machine-learned interatomic potentials and their application in molecular simulations and Pavol will explain their recent work on uncertainty quantification for message-passing graph neural network potentials using committees and foundation models

  [LINK]