A5001084719- Particle physics theoretical and experimental studies
- Particle Detector Development and Performance
- Reinforcement Learning in Robotics
- Computational Physics and Python Applications
- Robotic Locomotion and Control
- Distributed and Parallel Computing Systems
- Robot Manipulation and Learning
University of Cambridge
2022
Institute of Mathematical Statistics
2022
University College London
2021
imitation provides open-source implementations of and reward learning algorithms in PyTorch. We include three inverse reinforcement (IRL) algorithms, a preference comparison algorithm. The have been benchmarked against previous results, automated tests cover 98% the code. Moreover, are implemented modular fashion, making it simple to develop novel framework. Our source code, including documentation examples, is available at https://github.com/HumanCompatibleAI/imitation
Reinforcement learning (RL) requires either manually specifying a reward function, which is often infeasible, or model from large amount of human feedback, very expensive. We study more sample-efficient alternative: using pretrained vision-language models (VLMs) as zero-shot (RMs) to specify tasks via natural language. propose and general approach VLMs models, we call VLM-RMs. use VLM-RMs based on CLIP train MuJoCo humanoid learn complex without specified such kneeling, doing the splits,...
Since the discovery of Higgs boson, testing many possible extensions to Standard Model has become a key challenge in particle physics. This paper discusses new method for predicting compatibility physics theories with existing experimental data from colliders. Using machine learning, technique obtained comparable results previous methods (>90% precision and recall) only fraction their computing resources (<10%). makes it test models that were impossible probe before, allows large-scale...
Measurements at particle collider experiments, even if primarily aimed understanding Standard Model processes, can have a high degree of model independence, and implicitly contain information about potential contributions from physics beyond the Model. The Contur package allows users to benefit hundreds measurements preserved in Rivet library test new models against bank LHC date. This method has proven be very effective several recent publications team, but ultimately, for this approach...