Efficient Model-Based Deep Reinforcement Learning with Variational State Tabulation

Authors: Dane Corneil, Wulfram Gerstner, Johanni Brea

ICML 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluated the Va ST agent on a series of navigation tasks implemented in the Viz Doom environment... We compared the performance of Va ST against two recently published sample efficient model free approaches: Neural Episodic Control (NEC) (Pritzel et al., 2017) and Prioritized Double DQN (Schaul et al., 2015).
Researcher Affiliation Academia Dane Corneil 1 Wulfram Gerstner 1 Johanni Brea 1 Laboratory of Computational Neuroscience (LCN), School of Computer and Communication Sciences and Brain Mind Institute, School of Life Sciences, Ecole Polytechnique F ed erale de Lausanne, Switzerland.
Pseudocode Yes The pseudocode of Va ST, and of our implementation of prioritized sweeping, are in the Supplementary Material.
Open Source Code Yes The full code for Va ST can be found at https://github. com/danecor/Va ST/.
Open Datasets Yes We evaluated the Va ST agent on a series of navigation tasks implemented in the Viz Doom environment (see Figure 3A, Kempka et al. (2016)).
Dataset Splits No The paper describes evaluation over "test epochs" but does not provide specific training/validation/test dataset splits with percentages or sample counts, which is typical for static datasets rather than online reinforcement learning environments.
Hardware Specification No The paper does not provide specific hardware details such as GPU or CPU models, processor types, or memory specifications used for running its experiments.
Software Dependencies No The paper mentions software environments like "Viz Doom environment" and "Arcade Learning Environment" but does not provide specific version numbers for these or any other ancillary software components or libraries.
Experiment Setup Yes We use a multilayer perceptron (3 layers for each possible action) for the transition model pθT... and ...temperatures taken from those suggested in (Maddison et al., 2016): λ1 = 2/3 for the posterior distribution and λ2 = 0.5 for evaluating the transition log probabilities. We used two replay memory sizes (N = 100 000 and N = 500 000).