Embedded Files
Reliable Machine & Reinforcement Learning
I offer a part time scientific soft. eng. position @ MIM University of Warsaw.
I present a selection of research topics pursued within the group below
Safety in Reinforcement Learning
Safety in Reinforcement Learning
We are interested in studying the safety aspects of reinforcement learning algorithms. The problem is how to train agents successfully performing the prescribed task while conserving safety constraints. Algorithms are evaluated using the set of challenging safety-gym environments https://github.com/openai/safety-gym . We study efficient safety RL algorithms based on the state-of-the-art off-policy RL for raw vector observations and safety for visual RL control.
Neural ODEs, physics informed ML
Neural ODEs, physics informed ML
We study efficient algorithms for training Deep Neural ODE models introduced by R. Chen et al. "Neural Ordinary Differential Equations", NeurIPS 2018. Over the years, neural ODE has become one of the major deep-learning models for time-series data. Among the advantages are memory efficiency, the possibility of continuous interpolation, and modeling partially observable time series. In particular, we generalize the approach of Latent ODE from subsequent work.
Reachability of deep NN & Neural ODEs
Reachability of deep NN & Neural ODEs
In light of the existence of adversarial examples, it is crucial to study deep neural network architectures' stability properties and estimate the set overestimating the reachable set of outputs taken by the NN. Our goal is to design an alternative approach to, e.g., the GoTube algorithm introduced in S. Gruenbacher et al. "GoTube: Scalable Statistical Verification of Continuous-Depth Models", AAAI 2022.
Offer
Offer
part-time contract (at least 20 hrs weekly commitment) with the university. Salary 5800 gross (negotiable);
Contract for a year (non-renewable at this time);
Possibility of applying for scientific grants, work towards publications on major CS conferences;
work towards MSc or PhD thesis;
Access to personal computer and computational resources.
Requirements
Requirements
At least bachelor or engineer title in computer science/mathematics/physics;
passion for research;
proficient in Python programming;
know fundamentals of machine learning, neural networks and reinforcement learning algorithms (e.g. the Deep Learning book);
capability of working individually and self-study, available to start as soon as possible.
Nice to have
Nice to have
practical experience with deployment of machine learning;
publication record in conferences/journals;
knowledge of Julia programming language.
Application
Application
Apply by sending CV, a cover letter addressed to the Dean of the Faculty of Mathematics, Informatics and Mechanics, University of Warsaw to cyranka (at) mimuw.edu.pl using the tag [nawa] in the title.
Page updated
Google Sites
Report abuse