Juan Rocamonde

Research Fellow

FAR AI

Biography

I am currently doing research on artificial intelligence with FAR AI. My current focus is on RL reward specification from natural language feedback. My general interest is understanding and solving intelligence, and how that relates to learning, knowledge, and interpretability, both for humans and machines.

Interests

Artificial Intelligence
Deep Reinforcement Learning
Natural Language Processing
Interpretability

Education

MASt in Mathematics, 2022
University of Cambridge
BSc in Theoretical Physics, 2021
University College London

Featured Publications

Adam Gleave, Mohammad Taufeeque, Juan Rocamonde, Erik Jenner, Steven H. Wang, Sam Toyer, Maximilian Ernestus, Nora Belrose, Scott Emmons, Stuart Russell

November, 2022

imitation: Clean Imitation Learning Implementations

imitation provides clean open-source implementations of imitation and reward learning algorithms in PyTorch.

Juan Rocamonde, Louie Corpe, Gustavs Zilgalvis, Maria Avramidou, Jon Butterworth

July, 2022 In SciPost Phys.

Picking the low-hanging fruit: testing new physics at scale with active learning

The Contur Oracle is a ML method for testing the compatibility of proposed physics theories with existing experimental data from particle colliders. The technique makes it possible to test physics models that were impossible to probe before, and allows for large-scale testing of new physics theories.

A. Buckley, J. M. Butterworth, L. Corpe, M. Habedank, D. Huang, D. Yallup, M. Altakach, G. Bassman, I. Lagwankar, Juan Rocamonde, H. Saunders, B. Waugh, G. Zilgalvis

May, 2021 In SciPost Phys. Core

Testing new physics models with global comparisons to collider measurements: the Contur toolkit

The Contur package allows users to test new models against a bank of LHC measurements. This manual accompanies the first user-facing version, with the hope that external groups will be able to run their own Contur studies.

Contact

juan@far.ai