Possibility Before Utility: Learning And Using Hierarchical Affordances

Authors: Robby Costales, Shariq Iqbal, Fei Sha

ICLR 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In our experiments we aim to answer the following questions: (1) Does HAL improve learning in tasks with complex dependencies? (2) Is HAL robust to milestone selection and context stochasticity? (3) Can HAL more effectively learn a diverse set of skills when trained task-agnostically?
Researcher Affiliation Collaboration Robby Costales1 Shariq Iqbal1 Fei Sha2 1University of Southern California 2Google Research
Pseudocode Yes Algorithm 1 Learning procedure for HAL
Open Source Code Yes All code for environments, HAL, and other relevant baseline algorithms is provided at the following link: https://github.com/robbycostales/HAL.
Open Datasets No The paper describes custom-built environments (CRAFTING and TREASURE) that are procedurally generated, implying the data is created during runtime rather than from a pre-existing public dataset. It does not provide access information (link, DOI, citation) to a specific public dataset used for training.
Dataset Splits No The paper operates within a reinforcement learning framework where data is generated dynamically through environment interaction, rather than from a fixed dataset with explicit training/validation/test splits. Therefore, no explicit dataset split information is provided.
Hardware Specification No The paper does not specify the hardware used for running the experiments (e.g., CPU, GPU models, memory, or cloud instances).
Software Dependencies No The paper mentions using the 'Rainbow Deep Q-Learning algorithm' and specific algorithmic components like 'Noisy Nets' or 'C51', but it does not provide specific version numbers for software dependencies like programming languages, libraries, or frameworks (e.g., Python 3.8, PyTorch 1.9).
Experiment Setup Yes Hyperparameters are shown in Table 2.