Research map / Reinforcement learning / Visual and representation RL
Adaptive Auxiliary Task Weighting for Reinforcement Learning
Xingyu Lin, Harjatin Baweja, George Kantor, David Held
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 auxiliary tasks with gradient balancing | A2C / DDPG+HER · Atari and visual manipulation | sample efficiency (learning curves) | Improved |
| adapt auxiliary task weights online with OL-AUX-5 instead of fixed balanced weights | auxiliary-task RL · Atari and visual manipulation | sample efficiency (learning curves) | Improved |
| combine all auxiliary tasks with OL-AUX instead of using a single auxiliary task | OL-AUX · visual manipulation tasks | sample efficiency (learning curves) | Improved |
| use a single auxiliary task with gradient balancing | baseline RL · visual manipulation tasks | sample efficiency (learning curves) | No clear effect |
| use 5-step instead of 1-step weight update (ablation) | OL-AUX · RL environments (Fig. 6) | sample efficiency (learning curves) | Improved |
| add long-term reasoning (N-step objective) with same update frequency and data (ablation) | OL-AUX one-step ablation · RL environments (Fig. 6) | sample efficiency (learning curves) | Improved |
| add gradient smoothing (update w every 5 steps with more data) to one-step update (ablation) | OL-AUX-1 · RL environments (Fig. 6) | sample efficiency (learning curves) | Improved |
| use OL-AUX instead of fixed weights with an adversarial auxiliary loss | gradient descent · toy quadratic problem | sample efficiency (learning curves) | Improved |
| use OL-AUX derived weight rule instead of thresholded cosine similarity | auxiliary-task RL · Atari and visual manipulation | sample efficiency (learning curves) | Improved |
Closest papers by meaning
- Continual Auxiliary Task Learning (NeurIPS 2021 · similarity 0.90)
- Boosting Reinforcement Learning with Strongly Delayed Feedback Through Auxiliary Short Delays (ICML 2024 · similarity 0.87)
- ARS: Adaptive Reward Scaling for Multi-Task Reinforcement Learning (ICML 2025 · similarity 0.86)
- Task-Aware Virtual Training: Enhancing Generalization in Meta-Reinforcement Learning for Out-of-Distribution Tasks (ICML 2025 · similarity 0.85)
- Hard Tasks First: Multi-Task Reinforcement Learning Through Task Scheduling (ICML 2024 · similarity 0.85)
- Discovering Multiple Solutions from a Single Task in Offline Reinforcement Learning (ICML 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.
