Steven Lin

Berkeley College, Academia Sinica

Papers

3

Total Citations

302

H-Index

3

About

Steven Lin is a leading researcher in reinforcement learning, whose work focuses on enabling autonomous agents to learn broad, general-purpose skill repertoires directly from raw sensory input. His major contributions center on self-supervised and goal-conditioned learning, where agents set their own objectives to master diverse behaviors without manually engineered rewards. Lin’s most influential paper, "Visual Reinforcement Learning with Imagined Goals" (2018, 183 citations), pioneered a method for agents to learn skills by imagining and pursuing goals in visual environments, dramatically expanding the scope of autonomous learning. He further advanced this paradigm with "Skew-Fit: State-Covering Self-Supervised Reinforcement Learning" (2019, 66 citations), which introduced a technique for agents to efficiently explore and cover their state space, and "Planning with Goal-Conditioned Policies" (2019, 53 citations), which combined planning with learned policies to solve long-horizon tasks. Through these works, Lin has shaped how modern AI systems acquire flexible, reusable behaviors, making him a key figure in the push toward more general and capable autonomous agents.

Research Focus

Key Achievements

3
H-Index
3
Papers
302
Total Citations
101
Avg Citations/Paper
🏆 Most Cited Paper
Visual Reinforcement Learning with Imagined Goals
183 citations · 2018
📈 Most Prolific Year: 2019 (2 Papers)
🤝 Key Collaborators: 6
🏛 Institutions: Berkeley College, Academia Sinica

Top Papers

  1. 1
  2. 2
  3. 3

Key Collaborators

Contact & Links

Available for collaboration
Content generated · 13 days ago