Roughly every Thursday afternoon. ML, RL, DL and statistics. We schedule talks on many domains and of varying difficulty levels, ranging from basic statistics to recent advances in RL. If you are interested in participating, drop us a line, we promise we won’t spam you.
Normalizing Flows for Policy Representation in Reinforcement Learning
In part 3 on Normalizing Flows, we will discuss how Reinforcement Learning could benefit from this class of methods for policy representation: Using Normalizing Flows to represent stochastic policies in reinforcement learning: I assume that everyone has a basic understanding of RL. Therefore, I will only briefly review the fundamentals (state, action, environment, …)
Emerging Applications of Normalizing Flows in Reinforcement Learning
An Introduction to Normalizing Flows
Bayesian ML and Probabilistic Programming: Introduction
In this seminar, we go back to the very basics with Surya, who will (re)visit the fundamentals of the Bayesian approach to ML, including Maximum Likelihood and Maximum A Posteriori Estimation. He will introduce approximate Bayesian inference using PyMC3 (a Probabilistic Programming Language), which will serve as a springboard for understanding future material in this stream, such as Variational Methods, Graphical models, etc.
A soft introduction to causal inference
‘Correlation does not imply causation’ is a famous saying in statistics. The goal of many (scientific) studies is to draw conclusions on whether a so-called “treatment” has some effect or not, e.g. assigning a drug, or showing homepage A or homepage B to a visitor. As mentioned, it is in general not sufficient to observe statistical association of phenomena to infer causal relationships, but it is possible to make assumptions allowing for it.
Problems and solutions in offline learning
Current reinforcement learning algorithms require tons of data to solve new tasks. This “hunger for data” is especially onerous in high stake scenarios like healthcare or education, where on-policy data collection is infeasible. This motivates the Batch RL setting where the learner is only given a batch of transition tuples but allowed no further interactions with the environment.
Solving PDEs with neural networks
We’ll be talking about PDEs and how to solve them numerically using NNs. If you have no idea what those are you should probably read a little on it because we won’t be introducing them too much. Check the 3Blue1Brown video for some cool visuals. To say that PDEs are important in Science and Engineering would be a ridiculous understatement.
Isolation Forests: The good, the bad and the ugly
Anomaly detection is one of the main methods behind numerous real life machine learning use cases such as predictive maintenance, network intrusion detection, system health monitoring, fraud detection and novelty detection. Because of the high relevance and sensitivity of many application areas, robustness and reliability are main concerns when designing anomaly detection systems.
The calibrated classifier
Informed decision making based on classifiers requires that the confidence in their predictions reflect the actual error rates. When this happens, one speaks of a calibrated model. Recent work showed that expressive neural networks are able to overfit the cross-entropy loss without losing accuracy, thus producing overconfident (i.e. miscalibrated) models. We analyse several definitions of calibration and the relationships between them, look into related empirical measures and their usefulness, and explore several algorithms to improve calibration.
The Levenshtein Transformer