Notice: This website is continuously updated as new conference proceedings and journal volumes are published. Consequently, the figures presented here may differ from those reported in our publications.
A brief overview of our methodology is provided on the Methods page, along with references to publications documenting our approach and links to the code and data used in each study.
The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates.
Reincarnating Reinforcement Learning: Reusing Prior Computation to Accelerate Progress
Authors: Rishabh Agarwal, Max Schwarzer, Pablo Samuel Castro, Aaron C. Courville, Marc Bellemare
NeurIPS 2022 | Venue PDF | LLM Run Details | Input Tokens: 24,048 Total number of tokens sent to the LLM as input for this paper's analysis. | Output Tokens: 5,203 Total number of tokens produced by the LLM (including reasoning/thinking tokens) for this paper's analysis.
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate reincarnating RL s gains over tabula rasa RL on Atari 2600 games, a challenging locomotion task, and the real-world problem of navigating stratospheric balloons. |
| Researcher Affiliation | Collaboration | 1 Google Research, Brain Team 2 MILA |
| Pseudocode | No | The paper does not contain a pseudocode block or algorithm section. |
| Open Source Code | Yes | Open-sourced code and trained agents at agarwl.github.io/reincarnating_rl. |
| Open Datasets | Yes | We conduct experiments on ALE with sticky actions [57]. To reduce the computational cost of our experiments, we use a subset of 10 commonly-used Atari 2600 games: Asterix, Breakout, Space Invaders, Seaquest, Q Bert, Beam Rider, Enduro, Ms Pacman, Bowling and River Raid. |
| Dataset Splits | No | For the results in Section 4, we use 3 seeds per game on 10 games. |
| Hardware Specification | Yes | We obtain the teacher policy πT by running DQN [60] with Adam optimizer for 400 million environment frames, requiring 7 days of training per run with Dopamine [15] on P100 GPUs. |
| Software Dependencies | No | We use actor-critic agents in Acme [37]. |
| Experiment Setup | Yes | For the experiments in Section 4, we use learning rate of 1e-4, Adam optimizer, a batch size of 32, a discount factor of 0.99, a target update period of 2000, replay buffer size of 1M, and an epsilon schedule of 250k frames. |