PeopleofSuperintelligence

Researcher

Andrei A. Rusu

Also known as Andrei Rusu, Andrei-Alexandru Rusu

Andrei A. Rusu is an AI researcher whose work spans text classification, reinforcement learning, continual learning, and language-model training. He and collaborators published a rank-based text classifier, then developed policy distillation, progressive networks, and a method for retaining skills across tasks. Later, he studied transfer across Atari game variations and asynchronous training for language models. In 2025, he taught a Google DeepMind course and spoke as a Google DeepMind researcher at Romanian AI Days; a 2026 patent application named him as an inventor on distributed path composition.

Works
25
Citations
43,520
h-index
16
Badges

Publication numbers from OpenAlex, as of 18 October 2026.

Timeline

  1. 2010

    Text categorization by rank

    In 2010, he and Liviu P. Dinu publish a paper on rank-distance aggregation as a fixed rule for combining classifiers in text categorization.

  2. 2015

    Distilling reinforcement policies

    In 2015, he and colleagues introduce policy distillation, a method for extracting a reinforcement-learning agent’s policy and training a smaller, more efficient network.

  3. 2016

    Progressive networks for robots

    In 2016, he and colleagues introduce progressive neural networks, which reuse earlier learned features through lateral connections. In a second 2016 study, they transfer learned policies from simulation to a real robot using progressive networks.

  4. 2017

    Protecting earlier skills

    In 2017, he and colleagues publish elastic weight consolidation, which protects knowledge of earlier tasks by slowing learning on important network weights. They demonstrate sequential learning on MNIST and Atari games.

  5. 2022

    Transfer across Atari games

    In 2022, he and colleagues use Atari game curricula as a transfer benchmark for deep reinforcement learning without manually engineering tasks. Their analysis finds that game variations and combinations affect learning and transfer.

  6. 2024–2026

    Distributed learning and path composition

    In 2024, he coauthors an asynchronous Local-SGD study for language-model training. Tests up to 150 million parameters match synchronous training on perplexity per update and improve wall-clock time. In March and April 2025, he teaches a Google DeepMind course on large language models, and in September 2025 he speaks at Romanian AI Days as a Google DeepMind researcher. In March 2026, a patent application assigned to DeepMind names him as an inventor on neural networks with distributed path composition.

Most-cited works

  1. Human-level control through deep reinforcement learning2015 · Nature · 30,201 citationsWith Alex Graves, Daan Wierstra, David Silver, Demis Hassabis, Ioannis Antonoglou and Koray Kavukcuoglu
  2. Overcoming catastrophic forgetting in neural networks2017 · Proceedings of the National Academy of Sciences · 7,721 citationsWith Demis Hassabis
  3. Progressive Neural Networks2016 · 1,307 citationsWith Koray Kavukcuoglu
  4. Meta-Learning with Latent Embedding Optimization2018 · 1,068 citationsWith Oriol Vinyals and Simon Osindero
  5. PathNet: Evolution Channels Gradient Descent in Super Neural Networks2017 · 619 citationsWith Alexander Pritzel and Daan Wierstra

From OpenAlex, as of 18 October 2026.

  1. The median h-index is 54 at universities and 20 at frontier AI labsIncludes Andrei A. Rusu. The median h-index is 54 at universities and 20 at frontier AI labs. We compare publishing careers, citations and moves to estimate how universities and frontier labs contribute to AI research.
  2. Google has the highest combined h-index among AI organizationsIncludes Google. Our estimates of combined and median h-index at AI organizations, using current affiliations and OpenAlex records.
  3. Google alumni founded or led 82 other organizations, the most among the labs we compareIncludes Google. We trace 189 organizations founded or led by AI lab alumni, and 196 career moves between the labs.
  4. Google had the most people shaping AI every year from 2014 to 2025Includes Google. Our estimates of which organizations employed the most people shaping AI from 1961 to 2025, with a year-by-year chart.

Colleagues

People with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Andrei A. Rusu yet. Send us a free-licensed photo.

  1. Publications - NLP Group, University of Bucharest nlp.unibuc.ro. Checked 18 October 2026.
  2. Policy Distillation arxiv.org. Checked 18 October 2026.
  3. Progressive Neural Networks arxiv.org. Checked 18 October 2026.
  4. Sim-to-Real Robot Learning from Pixels with Progressive Nets arxiv.org. Checked 18 October 2026.
  5. Overcoming catastrophic forgetting in neural networks pmc.ncbi.nlm.nih.gov. Checked 18 October 2026.
  6. Probing Transfer in Deep Reinforcement Learning without Task Engineering arxiv.org. Checked 18 October 2026.
  7. Asynchronous Local-SGD Training for Language Modeling deepmind.google. Checked 18 October 2026.
  8. Courses - AI for Science Masters ai.aims.ac.za. Checked 18 October 2026.
  9. Iasi2025 - Romanian AI Days days.airomania.eu. Checked 18 October 2026.
  10. US20260080263A1 - Neural networks with distributed path composition - Google Patents patents.google.com. Checked 18 October 2026.