Dynamic allocation of limited memory resources in reinforcement learning

Authors: Nisheet Patel, Luigi Acerbi, Alexandre Pouget

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We derive from first principles an algorithm, Dynamic Resource Allocator (DRA), which we apply to two standard tasks in reinforcement learning and a model-based planning task, and find that it allocates more resources to items in memory that have a higher impact on cumulative rewards.
Researcher Affiliation Academia Nisheet Patel Department of Basic Neurosciences University of Geneva nisheet.patel@unige.ch Luigi Acerbi Department of Computer Science University of Helsinki luigi.acerbi@helsinki.fi Alexandre Pouget Department of Basic Neurosciences University of Geneva alexandre.pouget@unige.ch
Pseudocode Yes Algorithm 1: Dynamic Resource Allocator (DRA)
Open Source Code Yes Code to run DRA and reproduce our results is available at https://github.com/nisheetpatel/ Dynamic Resource Allocator.
Open Datasets Yes First, we consider the grid-world adapted from Mattar and Daw [16] and depicted in Fig. 1a. Next, we test DRA on the mountain car problem [39]
Dataset Splits No The paper does not explicitly state training, validation, and test splits by percentages or sample counts.
Hardware Specification No The paper does not specify any hardware details such as CPU, GPU, or memory used for the experiments.
Software Dependencies No The paper does not list any specific software dependencies with version numbers.
Experiment Setup Yes Set hyper-parameters θ = (α1, α2, β, γ, λ) Initialize q, σ, table of memories = s, a, r, s , N( qsa, σ2 sa I) |S A|