Adaptable Agent Populations via a Generative Model of Policies

Authors: Kenneth Derek, Phillip Isola

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We test our generative model s capabilities in a variety of environments, including an open-ended grid-world and a two-player soccer environment.
Researcher Affiliation Academia Kenneth Derek MIT CSAIL kderek@alum.mit.edu Phillip Isola MIT CSAIL phillipi@mit.edu
Pseudocode Yes See Algorithm 1 in the supplement for additional details. [...] Details can be found in Algorithm 2 of the supplement.
Open Source Code Yes Code, visualizations, and additional experiments can be found at https://kennyderek.github.io/adap/.
Open Datasets No The paper describes using simulated environments ('Farmworld' and 'Markov Soccer') which generate data dynamically. It does not provide access information for a static, publicly available dataset.
Dataset Splits No The paper refers to training and testing in ablated environments but does not provide specific details on dataset splits (e.g., percentages or sample counts) for training, validation, or test sets in the traditional sense of static datasets.
Hardware Specification No The paper does not provide specific hardware details such as GPU/CPU models or processor types used for running experiments.
Software Dependencies No The paper mentions frameworks like PPO and DIAYN but does not list specific software dependencies with version numbers (e.g., Python 3.8, PyTorch 1.9).
Experiment Setup No The paper states, 'Training and Hyperparameters We train each method for the same number of timesteps (30 million), and generally keep hyperparameters constant across methods. These are described in the supplement.' Since the details are deferred to the supplement, they are not explicitly in the main text.