Research map / Reinforcement learning / Policy gradient and actor-critic
Adapting to Reward Progressivity via Spectral Reinforcement Learning
Michael Dann, John Thangarajah
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 |
|---|---|---|---|
| Use Spectral DQN instead of Pop-Art or DQN+TC | Ponglantis · Ponglantis | mean episodic return | Improved |
| Use Spectral DQN instead of Pop-Art or DQN+TC | Exponential Pong · Exponential Pong | mean episodic return | Improved |
| Use Spectral DQN instead of both baselines | Atari agents · Bowling | mean episodic return | Improved |
| Use Spectral DQN instead of both baselines | Atari agents · Chopper Command | mean episodic return | Improved |
| Use Spectral DQN instead of both baselines | Atari agents · Video Pinball | mean episodic return | Improved |
| Use Spectral DQN instead of DQN+TC | Atari agents · Centipede | mean episodic return | Hurt |
| Use Spectral DQN instead of the baselines | Atari agents · the other standard Atari games | mean episodic return | No clear effect |
| Use DQN+TC instead of Pop-Art | unclipped-reward DQN · Ponglantis | mean episodic return | Improved |
| Set spectral-frequency loss weights to b^i instead of balancing frequencies (ablation) | Spectral DQN · Ponglantis and Exponential Pong | mean episodic return | Hurt |
| Remove target-scale correction for hidden-layer gradients (set w_i = 1) (ablation) | Spectral DQN · standard Atari games | mean episodic return: 4 of the 6 games | Hurt |
| Scale Atlantis-phase perceived rewards down by a factor 10 | DQN+TC · Ponglantis Easier | mean episodic return | Improved |
| Scale Atlantis-phase perceived rewards down by a factor 100 | Pop-Art · Ponglantis Even Easier | mean episodic return | Improved |
Closest papers by meaning
- Spectral Normalisation for Deep Reinforcement Learning: An Optimisation Perspective (ICML 2021 · similarity 0.90)
- Graph-Theoretic Intrinsic Reward: Guiding RL with Effective Resistance (ICLR 2026 · similarity 0.88)
- Addressing Sparse-Rewards in RL with Scalable Hierarchical Novel Eigen Options (NeurIPS 2026 · similarity 0.88)
- Rainbow: Combining Improvements in Deep Reinforcement Learning (AAAI 2018 · similarity 0.87)
- Reward learning from human preferences and demonstrations in Atari (NeurIPS 2018 · similarity 0.87)
- Revisiting Rainbow: Promoting more insightful and inclusive deep reinforcement learning research (ICML 2021 · similarity 0.87)
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.
