Michael Panchenko | TransferLab — appliedAI Institute

Michael Panchenko

Mischa is a researcher with background in physics and mathematics who decided to change course and go into AI (for the sake of falsifiability of ideas). On his path since then he has worked on multiple projects in ML and data analysis and as a bonus gained some experience DevOps and in developing production grade solutions.

Mischa's work

Training

Practical anomaly detection

An two-day introduction to unsupervised ML techniques for anomaly detection and their strengths and weaknesses in different application areas.

Software

sensAI: The Python Library for Sensible AI

sensAI is a high-level AI toolkit with a specific focus on rapid experimentation for machine learning applications. It provides a unifying …

Dec 5, 2023

Software

Tianshou: An elegant deep reinforcement learning library

Tianshou is a deep reinforcement learning library based on pure PyTorch. It caters to the needs of researchers and practitioners alike by …

Reinforcement Learning

Dec 5, 2023

Training

Safe and efficient deep reinforcement learning

Mischa's work

Practical anomaly detection

An two-day introduction to unsupervised ML techniques for anomaly detection and their strengths and weaknesses in different application areas.

sensAI: The Python Library for Sensible AI

sensAI is a high-level AI toolkit with a specific focus on rapid experimentation for machine learning applications. It provides a unifying …

Tianshou: An elegant deep reinforcement learning library

Tianshou is a deep reinforcement learning library based on pure PyTorch. It caters to the needs of researchers and practitioners alike by …

Safe and efficient deep reinforcement learning

This 2-day dive into deep RL for decision and control is suitable for engineers who want to solve real-world control problems with efficient …

Machine Learning and Control

A one-day training introducing concepts and principles from control theory, with a heavy focus on optimal control theory, and presenting its …

Advantage-Induced Policy Alignment

Building on the classic results on reward weighted regression and its more recent adaptation to deep learning, a new algorithm called …

Direct Preference Optimization

With direct preference optimization (DPO), a language model can be aligned with human preferences without using reinforcement learning, …

Augmented Language Models: a survey

A survey of recent advances in augmenting (large) language models with new capabilities such as reasoning, tool use, and more. While the …

Representation learning with BYOL and SimSiam

BYOL was the first work to show how useful low-dimensional representations can be when learned in an unsupervised way without negative …

Mixture of Decision Trees for Interpretable Machine Learning

A linear gating function together with multiple expert decision trees is trained by expectation-maximization and results in a new, fully …

Critic Regularized Regression

A simple but powerful algorithm for offline reinforcement learning, which can be seen as a combination of behavior cloning and Q-learning, …

Deep Reinforcement Learning at the Edge of the Statistical Precipice

A sound statistical analysis of the performance of RL algorithms is performed, with outcomes that force a reconsideration of the ranking of …

Aleatoric and epistemic uncertainty in machine learning

This talk introduces the notions of aleatoric and epistemic uncertainty for probabilistic models and gives an overview of recent …

DictBERT: Dictionary Description Knowledge Enhanced Language Model Pre-training via Contrastive Learning

External knowledge can be injected into pre-trained language models (here specifically BERT) by training an additional language model on …

Git Re-Basin: Merging Models modulo Permutation Symmetries

Two neural networks $A$ and $B$ of the same architecture and trained by SGD are compared after finding a suitable permutation $\pi$ that …

Methods and issues in explainable AI

A two-day workshop for ML practitioners wishing to make their models better understandable for themselves and decision makers.

Modeling Long Sequences with Structured State Spaces

Building on ideas of using state-spaces and linear ODEs for sequence modeling in optimal control, a new layer for sequence processing is …

Chatterjee Coefficient

A new measure of independence between random variables, beyond just measuring linear dependence like correlation coefficients is proposed. …

A Fine-Grained Analysis on Distribution Shift

The topic of how different models behave under distribution shift is studied experimentally, after a framework for quantifying distribution …

Comparing Distributions by Measuring Differences that Affect Decision Making

A new family of divergences for distributions is introduced, generalizing KL and others, together with an efficient estimator for it. A …

Neural Collapse in deep classifiers during Terminal Phase of Training

Neural collapse refers to the observation that the last two layers of neural networks that were trained for a long time take a very simple, …

Label Encoding for Regression Networks

Regression is reformulated to predicting a binary code of length M with M binary classifiers. With carefully chosen encoding and decoding …

Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV)

A non-intrusive and user-input based method for interpreting decision-making of trained neural networks is introduced. It is cheap and has …

Learning Strides in Convolutional Neural networks

Instead of finding strides through a discrete search, an analogy of the strides hyperparameter for CNNs can be learned during training with …

Natural, Trust Region and Proximal Policy Optimization

We present an overview of the theory behind three popular and related algorithms for gradient based policy optimization: natural policy …

Calibration of Classifiers

We discuss the calibration of probabilistic classifiers, i.e. the question of whether the output vectors of such classifiers accurately …

Calibration of modern classifiers

An introduction to the pitfalls of uncalibrated classifiers and modern techniques for (re)calibration.

K.A.L.E. - The calibration game

The calibration game simulates resource allocation (medicine) for healthcare. A limited budget constraints prescription to patients with …

kyle - A toolkit for classifier calibration

There exist a number of different methods for calibrating models, and for measuring and visualizing calibration. Given the lack of reliable …

Markov Logic Networks: Construction, Learning and Inference

Markov logic networks provide a simple way for combining two seemingly rather different descriptions of data - first order logic and …