Planning with Sequence Models through Iterative Energy Minimization

Authors: Hongyi Chen, Yilun Du, Yiye Chen, Joshua B. Tenenbaum, Patricio A. Vela

ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we evaluate the planning performance of LEAP in Baby AI and Atari environments. We compare with a variety of different offline reinforcement learning approaches, and summarize the main results in Figure 4.
Researcher Affiliation Academia Georgia Institute of Technology Massachusetts Institute of Technology
Pseudocode Yes Algorithm 1 Iterative Planning through Energy Minimization (for discrete actions)
Open Source Code No The paper provides a 'Project website: https://hychen-naza.github.io/projects/LEAP.' in the abstract, but does not explicitly state that source code for the described methodology is released or available there.
Open Datasets Yes By iteratively generating actions through planning, we illustrate how our proposed framework outperforms prior methods in both Baby AI (Chevalier-Boisvert et al., 2019) and Atari (Bellemare et al., 2013) tasks. [...] Table 3: Quantitative Comparison on Atari. Gamernormalized scores for the 1% DQN-replay Atari dataset (Agarwal et al., 2020).
Dataset Splits No The paper mentions 'Models are trained with 500 optimal trajectory demos' and 'results are averaged over 5 random seeds,' but does not specify explicit training/validation/test dataset splits by percentage, absolute counts, or reference predefined splits with citations for reproducibility.
Hardware Specification No The paper does not provide specific hardware details such as GPU models, CPU models, or memory specifications used for running the experiments.
Software Dependencies No The paper mentions using specific models like Decision Transformer and Baby AI agent model's instruction encoder, and types of layers (GRU, Transformer attention), but does not provide specific version numbers for software dependencies (e.g., Python, PyTorch, TensorFlow, etc.).
Experiment Setup Yes The full list of hyperparameters can be found in Table 6 and Table 7, most of the hyperparameters are taken from Decision Transformer and Baby AI agent model. [...] We list out these parameters for LEAP and DT models in Table 8.