Linear Feature Encoding for Reinforcement Learning

Authors: Zhao Song, Ronald E. Parr, Xuejun Liao, Lawrence Carin

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

Reproducibility Variable Result LLM Response
Research Type Experimental 5 Experiments The goal of our experiments is to show that the model of and algorithms for feature encoding presented above are practical and effective. Figure 1 shows the results with means and 95% confidence intervals, given different numbers of training episodes, where Encoder τ corresponds to the version of Algorithm 1 with τ changes in the encoder.
Researcher Affiliation Academia Zhao Song, Ronald Parr , Xuejun Liao, Lawrence Carin Department of Electrical and Computer Engineering Department of Computer Science Duke University, Durham, NC 27708, USA
Pseudocode Yes Algorithm 1 Iterative Learning of Encoder and Policy. Algorithm 2 Linear Feature Discovery.
Open Source Code No All code is written in MATLAB and tested on a machine with 3.1GHz CPU and 8GB RAM. No explicit statement of public release or link to code repository.
Open Datasets Yes To represent raw states for the encoder, we use three concatenated sampled MNIST digits and hence a raw state is a 28 × 28 × 3 = 2352 dimensional vector.
Dataset Splits Yes For the encoder, the number of features k is selected over the validation set to achieve the best performance.
Hardware Specification Yes All code is written in MATLAB and tested on a machine with 3.1GHz CPU and 8GB RAM.
Software Dependencies No All code is written in MATLAB and tested on a machine with 3.1GHz CPU and 8GB RAM. No specific MATLAB version or other software dependencies with versions are listed.
Experiment Setup Yes The discount factor is set to be 0.95. ... We used k = 50 features for both linear encoder and random projection. ... We set k = 203 features for the linear encoder.