Research map / Reinforcement learning / Visual and representation RL
Widening the Pipeline in Human-Guided Reinforcement Learning with Explanation and Context-Aware Data Augmentation
Lin Guan, Mudit Verma, Sihang Guo, Ruohan Zhang, Subbarao Kambhampati
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add human visual explanation and context-aware data augmentation to evaluative feedback | Efficient DQN with human feedback · Pixel-Taxi and Pong | environment and feedback samples to learn a near-optimal policy: 35% less environment samples/ feedback samples | Improved |
| Add human visual explanation and context-aware data augmentation to evaluative feedback | Efficient DQN with human feedback · Enduro-1000 | score at a given environment sample count: over 60 by using 80k samples compared to 120k samples used by DQN-TAMER, an over 30% improvement | Improved |
| Add human visual explanation to evaluative feedback | Efficient DQN with human feedback · MR Level 1 | policy performance | No clear effect |
| Add human visual explanation to evaluative feedback | Efficient DQN with human feedback · Asterix and MR Level 1 | score | Improved |
| Replace context-aware augmentation with Gaussian blurring or random cropping | EXPAND · Atari games and Pixel-Taxi | policy performance | Hurt |
| Use Ex-AGIL's separate attention prediction network and mask the policy input | DQN-Feedback · Pong | policy performance | Improved |
| Use Ex-AGIL's separate attention prediction network and mask the policy input | DQN-Feedback · visually complex environments | policy performance | Hurt |
| Add Attention-Align's attention alignment loss | Efficient DQN with feedback · all five tasks | usable policy learning | Hurt |
| Remove the invariance loss (ablation) | EXPAND · synthetic-oracle tasks | policy performance | Hurt |
| Remove augmented data from the advantage loss (ablation) | EXPAND · synthetic-oracle tasks | policy performance | Hurt |
| Increase augmentations per evaluated state from 5 to 12 (ablation) | EXPAND · Pixel-Taxi and Pong | policy performance | Improved |
| Use fixed Gaussian filters instead of randomly sampled filters (ablation) | Context-agnostic Gaussian-blurring augmentation · Atari games and Pixel-Taxi | policy performance | No clear effect |
Closest papers by meaning
- GUIDE: Real-Time Human-Shaped Agents (NeurIPS 2024 · similarity 0.88)
- Leveraging Sub-Optimal Data for Human-in-the-Loop Reinforcement Learning (ICLR 2025 · similarity 0.86)
- Explainable Reinforcement Learning from Human Feedback to Improve Alignment (NeurIPS 2025 · similarity 0.86)
- Exploring Data Scaling Trends and Effects in Reinforcement Learning from Human Feedback (NeurIPS 2025 · similarity 0.86)
- Enhancing Rating-Based Reinforcement Learning to Effectively Leverage Feedback from Large Vision-Language Models (ICML 2025 · similarity 0.86)
- VLM Agents Generate Their Own Memories: Distilling Experience into Embodied Programs of Thought (NeurIPS 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
