When Does Bounded-Optimal Metareasoning Favor Few Cognitive Systems?

Authors: Smitha Milli, Falk Lieder, Thomas Griffiths

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

Reproducibility Variable Result LLM Response
Research Type Experimental We investigate these questions in two settings: a one-shot decision between two alternatives, and planning under uncertainty in a Markov decision process. Our experiments show that the optimal number of systems increases with the variability of the environment but decreases with the costliness of metareasoning.
Researcher Affiliation Academia Smitha Milli EECS Department University of California Berkeley, CA 94720 smilli@berkeley.edu Falk Lieder Helen Wills Neuroscience Institute University of California Berkeley, CA 94720 falk.lieder@berkeley.edu Thomas L. Griffiths Department of Psychology University of California Berkeley, CA 94720 tom griffiths@berkeley.edu
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide concrete access information (link, explicit statement) to the source code for the methodology described.
Open Datasets No The paper describes how data was generated through sampling from distributions (e.g., Bernoulli, uniform, Gamma) for simulations but does not refer to a publicly available or open dataset with concrete access information.
Dataset Splits No The paper describes experimental setup but does not provide specific dataset split information (exact percentages, sample counts, or citations to predefined splits) for training, validation, or testing.
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, memory amounts) used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details (e.g., library or solver names with version numbers) needed to replicate the experiment.
Experiment Setup Yes For the 2AFC setting: 'We model this by sampling θ from a uniform distribution. Without loss of generality, we assume that θ Pθ = Unif(0.5, 1)... re 1 Γ(α, β) was sampled from a Gamma distribution so that acting is always as least as costly as sampling (re 1).' For Planning under Uncertainty: 'a simple 20 20 grid world where the agent s goal is to get from the lower left corner to the upper right corner with as little cost as possible. The horizon was set to 500, the maximum number of rollouts at any thinking stage to 10, and the depth of each BRTDP rollout to 10. BRTDP was initialized with a constant value function of 0 for the lower bound and a constant value function of 106 for the upper bound.'