Research map / Probabilistic and scientific modeling / Variational and generative sampling
Transfer of Value Functions via Variational Methods
Andrea Tirinzoni, Rafael Rodriguez Sanchez, Marcello Restelli
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 |
|---|---|---|---|
| transfer via a variational posterior over value-function weights (GVT, MGVT) instead of no transfer | RL with epsilon-greedy / DDQN · Rooms, Cartpole, Mountain Car, Maze Navigation | expected return vs iterations | Improved |
| use more than 1 mixture component in the posterior (3-MGVT vs 1-MGVT) (ablation) | MGVT · Rooms problem | expected return | No clear effect |
| use 3 mixture components instead of 1 (ablation) | MGVT · Maze Navigation | expected return | Improved |
| replace Gaussian prior/posterior with a Gaussian mixture (ablation) | Gaussian variational transfer · Rooms with shifted source distribution, Mountain Car, Maze | expected return | Improved |
| replace Gaussian with mixture on easy tasks (ablation) | variational transfer · Cartpole | expected return | No clear effect |
| use variational posterior with Thompson-style resampling instead of fine-tuning a random source Q-function | MGVT · Rooms, Cartpole, Mountain Car | expected return vs iterations | Improved |
| use fine-tuning of a random source instead of MGVT in complex mazes | fine-tuning baseline · Maze Navigation | expected return | Hurt |
| meta-train MAML on only the fixed source tasks instead of the full task distribution | MAML · navigation tasks | expected return vs gradient steps | Hurt |
Closest papers by meaning
- Reinforcement Learning with General Value Function Approximation: Provably Efficient Approach via Bounded Eluder Dimension (NeurIPS 2020 · similarity 0.88)
- Smoothed Action Value Functions for Learning Gaussian Policies (ICML 2018 · similarity 0.88)
- Deep Exploration via Randomized Value Functions (JMLR 2019 · similarity 0.88)
- Information-Theoretic Considerations in Batch Reinforcement Learning (ICML 2019 · similarity 0.87)
- Enhancing Value Function Estimation through First-Order State-Action Dynamics in Offline Reinforcement Learning (ICML 2024 · similarity 0.86)
- Learning the Target Network in Function Space (ICML 2024 · similarity 0.86)
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.
