Identifying Selections for Unsupervised Subtask Discovery

Authors: Yiwen Qiu, Yujia Zheng, Kun Zhang

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our empirical results on a challenging Kitchen environment demonstrate that the learned subtasks effectively enhance the generalization to new tasks in multi-task imitation learning scenarios.
Researcher Affiliation Academia Yiwen Qiu Carnegie Mellon University Pittsburgh, PA 15213 yiwenq@andrew.cmu.edu Yujia Zheng Carnegie Mellon University Pittsburgh, PA 15213 yujiazh2@andrew.cmu.edu Kun Zhang Carnegie Mellon University, MBZUAI Pittsburgh, PA 15213 kunz1@andrew.cmu.edu
Pseudocode Yes Algorithm 1 Seq NMF for learning subtasks
Open Source Code Yes The codes are provided at this link.
Open Datasets Yes We use the demonstrations provided by (Gupta et al. [2019]) for reproducibility, which only contain state and action pairs but not reward.
Dataset Splits No The paper only explicitly mentions 'training' and 'testing' sets, but does not provide details for a distinct 'validation' set.
Hardware Specification Yes All experiments were conducted on either NVIDIA L40, or Ge Force RTX 3080 Ti, or a Mac M1 chip with 16GB of RAM.
Software Dependencies No The paper does not provide specific version numbers for software dependencies or libraries used in the experiments.
Experiment Setup Yes For the experiments in seq-NMF, we use the following hyperparameters in Tab. 4. ... For the experiments in transfering to new tasks, we use the following hyperparameters in Tab. 5.