Deep Decentralized Multi-task Multi-Agent Reinforcement Learning under Partial Observability

Authors: Shayegan Omidshafiei, Jason Pazis, Christopher Amato, Jonathan P. How, John Vian

ICML 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 6. Evaluation. We first evaluate single-task performance of the introduced Dec-HDRQN approach on a series of increasingly challenging domains. ... Performance is evaluated on both multi-agent single-target (MAST) and multi-agent multi-target (MAMT) capture domains...
Researcher Affiliation Collaboration 1Laboratory for Information and Decision Systems (LIDS), MIT, Cambridge, MA, USA 2College of Computer and Information Science (CCIS), Northeastern University, Boston, MA, USA 3Boeing Research & Technology, Seattle, WA, USA.
Pseudocode No No structured pseudocode or algorithm blocks were found in the paper.
Open Source Code No The paper does not provide an explicit statement or link for open-source code availability.
Open Datasets Yes Performance is evaluated on both multi-agent single-target (MAST) and multi-agent multi-target (MAMT) capture domains, variations of the existing meeting-in-a-grid Dec POMDP benchmark (Amato et al., 2009).
Dataset Splits No The paper evaluates performance based on 'randomly-initialized episodes' and 'training epochs' rather than explicit train/validation/test dataset splits with percentages or counts.
Hardware Specification No The paper does not provide specific details regarding the hardware used for running experiments, such as GPU/CPU models or processor types.
Software Dependencies No The paper mentions software components like 'Adam optimizer' and 'DRQNs' but does not provide specific version numbers for any libraries or frameworks used, such as 'PyTorch 1.9' or 'TensorFlow 2.0'.
Experiment Setup Yes All experiments use DRQNs with 2 multi-layer perceptron (MLP) layers, an LSTM layer (Hochreiter & Schmidhuber, 1997) with 64 memory cells, and another 2 MLP layers. MLPs have 32 hidden units each and rectified linear unit nonlinearities are used throughout, with the exception of the final (linear) layer. Experiments use γ = 0.95 and Adam optimizer (Kingma & Ba, 2014) with base learning rate 0.001. Dec-HDRQNs use hysteretic learning rate β = 0.2 to 0.4.