Scholar
Roberta Raileanu
Google Scholar ID: 462OoekAAAAJ
Research Scientist at Google DeepMind, Honorary Lecturer at UCL
Citations & Impact
Citations
17,185
H-index
31
i10-index
41
Publications
20
Co-authors
29
list available
Publications
Resume (English only)
Research Experience
- Senior Staff Research Scientist at Google DeepMind (Open-Endedness team)
- Former Research Scientist at Meta GenAI, where she led the AI Scientist team to accelerate end-to-end AI R&D using AI agents
- Founded and led the Tool Use team for Llama 3 at Meta, enabling tool usage (e.g., search, Python, text-to-image, Wolfram Alpha) and zero-shot generalization to new tools
- Her LLM agent research underpins widely used Meta products like Meta AI, Data Analyst, AI Studio, and Ads Business Agent, serving hundreds of millions of users
- Interned during PhD at DeepMind, Microsoft Research, and Facebook AI Research
Background
- Currently a Senior Staff Research Scientist at Google DeepMind in the Open-Endedness team
- Honorary Lecturer (Adjunct Professor) at UCL, closely collaborating with the UCL-DARK Lab and advising PhD students
- Co-developed and currently co-teaching a new course on Open-Endedness and General Intelligence
- Broadly interested in designing ML algorithms that make robust sequential decisions in complex environments while continuously acquiring new skills and knowledge
- Long-term goal: develop models that can solve a wide range of new tasks with little or no supervision or training
- Deeply committed to ensuring such systems are reliable, trustworthy, and aligned with human intentions
- Work draws from reinforcement learning, open-ended learning, self-supervised learning, and NLP
- Currently focuses on the intersection of large language models (LLMs) and decision-making/RL—especially augmenting LLMs with planning, reasoning, actions, tool use, and goal-directed behavior, and teaching them via human feedback in an open-ended manner
- Also interested in improving generalization and robustness of autonomous agents and large pretrained models
