Valerii Likhosherstov
Papers
1
Total Citations
2
H-Index
1
About
Valerii Likhosherstov is a researcher whose work sits at the intersection of deep learning, reinforcement learning, and efficient attention mechanisms. He is best known for advancing vision-based reinforcement learning (RL), a field grappling with high-dimensional pixel inputs and the risk of observational overfitting. In his influential 2021 paper, "Unlocking Pixels for Reinforcement Learning via Implicit Attention," Likhosherstov introduced a novel implicit attention mechanism that enables RL agents to focus on task-relevant features while discarding spurious correlations. This work addresses a critical bottleneck in training agents directly from raw pixels, offering a more robust and sample-efficient path to learning. While his citation count is still growing, the conceptual impact of his approach is significant, providing a foundation for future research in sample-efficient, vision-based RL. His contributions are particularly valuable for students and researchers seeking to bridge the gap between high-dimensional sensory input and effective policy learning in complex, real-world environments.
Research Focus
Key Achievements
Top Papers
- 1Unlocking Pixels for Reinforcement Learning via Implicit Attention2 citations · 2021