Rafael Pardinas

Applied Research Scientist

AI Frontier Research

Rafael, at ServiceNow Research, integrates his expertise in software engineering with his proficiency in distributed systems. This combination of skills strengthens his Machine Learning research, linking complex system architecture with practical algorithm deployment. His role spans both the fundamental and applied dimensions of this field.

Rafael, holding a Master’s in Computer Science and a Bachelor’s in Physics, is a core contributor to the fields of Reinforcement Learning (RL) and Natural Language Processing (NLP) at ServiceNow. His academic foundation supports his research activities, which have recently evolved. For the last 4 years, he has been engaged in Deep Reinforcement Learning, with a focus on Offline RL, Policy Optimisation, and RL-driven Energy-Based Models. At the end of 2022, Rafael’s research direction shifted to emphasise Reinforcement Learning from Human Feedback (RLHF) and Reinforcement Learning from AI Feedback (RLAIF). Central to this shift is his work on Reward Modelling, an integral part of his efforts in RLHF and RLAIF, aimed at refining learning algorithms through advanced feedback interpretation and response systems.

Interests

Reinforcement Learning
RLHF
Large Language Models

Publications

TapeAgents: a Holistic Framework for Agent Development and Optimization. Dzmitry Bahdanau, Nicolas Gontier, Gabriel Huang, Ehsan Kamalloo, Rafael Pardinas, Alexandre Piche, Torsten Scholak, Oleh Shliazhko, Jordan Prince Tremblay, Karam Ghanem, Soham Parikh, Mitul Tiwari, Quaizar Vohra. At ArXiv, 2024.

PDF Cite Code Video

Leveraging Human Preferences to Master Poetry. Rafael Pardinas, Gabriel Huang, David Vazquez, Alexandre Piche. At AAAI Workshops, 2023.

PDF Cite

Implicit Offline Reinforcement Learning via Supervised Learning. Alexandre Piche, Rafael Pardinas, David Vazquez, Igor Mordatch, Christopher Pal. At Workshop at the Neural Information Processing Systems (NeurIPS), 2022.

PDF Cite Code

A Probabilistic Perspective on Reinforcement Learning via Supervised Learning. Alexandre Piche, Rafael Pardinas, David Vazquez, Christopher Pal. At Workshop at the International Conference on Learning Representations (ICLR), 2022.

PDF Cite Code

LOOC: Localize Overlapping Objects with Count Supervision. Issam H. Laradji, Rafael Pardinas, Pau Rodriguez, David Vazquez. At International Conference on Image Processing (ICIP), 2020.

PDF Cite Code

Objects of violence: synthetic data for practical ML in human rights investigations. Lachlan Kermode, Jan Freyberg, Alican Akturk, Robert Trafford, Denis Kocetkov, Rafael Pardinas, Eyal Weizman, Julien Cornebise. At Workshop at the Neural Information Processing Systems (NeurIPS), 2019.

PDF Cite