Seminarium "Uczenie maszynowe"

7 kwietnia 2022 12:15
Piotr Kozakowski (Uniwerystet Warszawski)
Entropy-Regularized Planning
Recent works have shown the effectiveness of entropy regularization in Monte Carlo Tree Search (MCTS). In this presentation I will first introduce the framework of Maximum Entropy Reinforcement Learning and show how it can be …

31 marca 2022 12:15
Adam Izdebski (Uniwerystet Warszawski)
Generative Modelling with Optimization for Molecule Discovery
In recent years, discovering novel drug-like molecules become a common application of generative models. However, it is much harder to generate novel molecules that are at the same time optimized for being promising drug candidates. During …

17 marca 2022 12:15
Spyros Mouselinos (University of Warsaw)
Measuring CLEVRness: Black-box Testing of Visual Reasoning Models
How can we measure the reasoning capabilities of intelligence systems? Visual question answering provides a convenient framework for testing the model's abilities by interrogating the model through questions about the scene. However, despite scores of …

3 marca 2022 12:15
Michał Zawalski (Uniwerystet Warszawski)
Gentle introduction to multi-agent reinforcement learning
Recent studies show some impressive applications of reinforcement learning algorithms in sequential decision-making problems. In my talk, I will focus on problems that involve controlling a group of agents, i.e. multi-agent reinforcement learning. Though that …

27 stycznia 2022 12:15
Jakub Świątkowski (Uniwerystet Warszawski)
Tutorial on deep learning generative models for speech synthesis
Speech synthesis has important applications in virtual assistants, voice interfaces, and accessibility. There has been rapid progress in the quality of speech synthesis systems in recent years thanks to deep learning generative models. In recent months, fully end-to-end neural generative approaches achieved for the first time human-level performance on challenging …

16 grudnia 2021 12:15
Jan Ludziejewski (Uniwerystet Warszawski)
Towards Generative Music
The OpenAI Jukebox was a groundbreaking model in sound generation and is still considered to be the state-of-the-art in the music modeling task. It consists of two separate networks, Vector Quantization Variational Autoencoder, which strongly compresses …

2 grudnia 2021 12:15
Piotr Tempczyk (Uniwersytet Warszawski)
LIDL: Local Intrinsic Dimension estimation using approximate Likelihood
Understanding how neural networks work is one of the most important questions in machine learning research. Their performance is connected with the shape of the data manifold. The structure of this manifold can be explored …

18 listopada 2021 12:15
Michał Zając (Uniwersytet Jagielloński)
Continual World: Continual learning meets reinforcement learning
First, I'll introduce the setup of continual learning. I'll talk about how various methods mitigate catastrophic forgetting and improve forward transfer. I'll also show what are the trade-offs and requirements for continual learning methods. In …

4 listopada 2021 12:15
Konrad Czechowski (Uniwerystet Warszawski)
Subgoal Search For Complex Reasoning Tasks
I will present our publication accepted to NeurIPS 2021. We proposed a method that improves search guided by neural networks in  combinatorially complex domains. Its key component is a learned  subgoal generator that produces a diversity of …

21 października 2021 12:15
Łukasz Kuciński (IM PAN)
Catalytic Role of Noise and Necessity of Inductive Biases in the Emergence of Compositional Communication
 We will talk about our recent compositionality paper accepted at NeurIPS 2021. Communication is compositional if complex signals can be represented as a combi-nation of simpler subparts. In this paper, we theoretically show that inductive biases …

17 czerwca 2021 12:15
Sebastian Jaszczur
Sparsity in Efficient Transformers
Large Transformer models yield impressive results on many tasks, but are expensive to train, or even fine-tune, and so slow at decoding that their use and study becomes out of reach. We address this problem …

6 maja 2021 12:15
Piotr Kozakowski
Q-Value Weighted Regression: Reinforcement Learning with Limited Data
 Sample efficiency is a major challenge in the current Reinforcement Learning (RL) systems. Another is robustness - it is hard to find one RL algorithm that will perform well in a variety of settings. I …

25 marca 2021 12:15
Bartłomiej Polaczyk
Loss landscape of deep neural networks
Finding the global minimum of a general non-convex function is in general an NP-complete problem. The optimization objective of deep neural networks (DNN) is not only non-convex but even non-smooth in case of ReLU activation …

18 marca 2021 12:15
Piotr Tempczyk
LIDL: Local Intrinsic Dimension estimation using Likelihood
We investigate the problem of local intrinsic dimension (LID)estimation. LID of the data is the minimal number of coordinates which are necessary to describe the data point and its neighborhood without significant information loss. Existing methods for …

4 marca 2021 12:15
Kajetan Janiak
Modelling rigid body dynamics with physics-informed neural networks
Functions describing physical systems of rigid bodies lie in a low-dimensional subspace of functions that neural networks can represent. If we could construct a NN architecture, that for any set of parameters yields dynamics of …

4 lutego 2021 12:15
Jacek Cyranka
Safety
This is a "reading group" type of seminar - expect less polished presentation, more of a learning effort for the interested participants.

26 listopada 2020 12:15
Piotr Januszewski
Decision making in uncertainty: Introduction to Multi-Armed Bandits algorithms
There are many reasons to care about bandit problems. Decision-making with uncertainty is a challenge we all face, and bandits provide a simple model of this dilemma we can study. Bandit problems have practical applications …

12 listopada 2020 12:15
Tomasz Danel (Jagiellonian University)
Machine Learning in Computer-Aided Drug Design
Drug discovery is a long and expensive process. Currently, pharmaceutical companies begin to acknowledge the potential of machine learning methods. These methods enable the effective exploration of the chemical space and the rapid identification of …

29 października 2020 12:15
Michał Garmulewicz
Towards structured world models for control
In recent years model-based RL has broadly lived up to its promise and has kept bringing outstanding benchmark performance, especially in the limited data regime.  Secondly, there is a growing interest in using structural priors …