Near Optimal Behavior via Approximate State Abstraction

Authors: David Abel, David Hershkowitz, Michael Littman

ICML 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We present theoretical guarantees of the quality of behaviors derived from four types of approximate abstractions. Additionally, we empirically demonstrate that approximate abstractions lead to reduction in task complexity and bounded loss of optimality of behavior in a variety of environments.
Researcher Affiliation Academia David Abel DAVID ABEL@BROWN.EDU D. Ellis Hershkowitz DAVID HERSHKOWITZ@BROWN.EDU Michael L. Littman MICHAEL LITTMAN@BROWN.EDU Brown University, 115 Waterman Street, Providence, RI 02906
Pseudocode No The paper contains mathematical derivations and proofs but no structured pseudocode or algorithm blocks.
Open Source Code Yes Our code base1 provides implementations for abstracting arbitrary MDPs as well as visualizing and evaluating the resulting abstract MDPs. 1github.com/david-abel/state_abstraction
Open Datasets No The paper describes environments like 'NChain', 'Taxi', 'Minefield', and 'Random MDP' which are used in experiments, but it does not provide concrete access information (specific link, DOI, repository name, or formal citation with authors/year) for publicly available datasets as 'training data' in the typical sense.
Dataset Splits No The paper mentions running trials and confidence intervals for results but does not provide specific dataset split information (exact percentages, sample counts, citations to predefined splits, or detailed splitting methodology) for training, validation, or test sets.
Hardware Specification No The paper does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper mentions software like Graph Stream and BURLAP but does not provide specific ancillary software details with version numbers (e.g., library or solver names with version numbers) needed to replicate the experiment.
Experiment Setup Yes For all experiments, we set γ to 0.95. ... In the Taxi and Random domains, 200 trials were run for each data point, whereas 20 trials were sufficient in Minefield and NChain.