Successor Feature Sets: Generalizing Successor Representations Across Policies

Authors: Kianté Brantley, Soroush Mehri, Geoff J. Gordon11774-11781

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

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct experiments to explore which of the potential barriers to scaling are most pressing. ... Experiments: Dynamic Programming We tried our dynamic programming method on several small domains: the classic mountain-car domain and a random 18 x 18 gridworld with full and partial observability. We evaluated both planning and feature matching; results for the former are discussed in this section, and an example of the latter is in Fig. 3.
Researcher Affiliation Collaboration Kiant e Brantley,1 Soroush Mehri, 2 Geoffrey J. Gordon 2 1 University of Maryland College Park 2 Microsoft Research
Pseudocode Yes Algorithm 1: Feature Matching Policy
Open Source Code No The paper does not provide an explicit statement or link for open-source code for the described methodology.
Open Datasets No The paper uses simulated environments ('mountain-car domain' and '18x18 gridworld') but does not specify or provide access to a distinct, publicly available dataset in the format required (e.g., a link, DOI, or formal citation for a pre-existing dataset).
Dataset Splits No The paper does not provide specific details on training, validation, or test dataset splits (e.g., exact percentages or sample counts), nor does it reference predefined splits with citations.
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, memory amounts, or cloud instance types) used for running its experiments.
Software Dependencies No The paper does not provide specific software dependencies or their version numbers necessary for replication.
Experiment Setup Yes In mountain-car, the agent has two actions: accelerate left and accelerate right. The state is (position, velocity), in [-1.2, 0.6] x [-0.07, 0.07]. We discretize to a 12x12 mesh with piecewise-constant approximation. Our one-step features are radial basis functions of the state, with values in [0, 1]. We use 9 RBF centers evenly spaced on a 3x3 grid. ... In the MDP gridworld, the agent has four deterministic actions: up, down, left, and right. The one-step features are (x, y) coordinates scaled to [-1, 1], similar to Fig. 3. In the POMDP gridworld, the actions are stochastic, and the agent only sees a noisy indicator of state. In all domains, the discount is γ = 0.9. ... We evaluate directions mi that we optimized for during backups, as well as new random directions. ... this persistent error is due to our limited-size representation of Φ. The error decreases as we increase the number of boundary points that we store.