Deep Reinforcement Learning: A State-of-the-Art Walkthrough
Authors: Aristotelis Lazaridis, Anestis Fachantidis, Ioannis Vlahavas
JAIR 2020 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In order to give a compact and practical overview of their differences, we present comprehensive comparison figures and tables, produced by reported performances of the algorithms under two popular simulation platforms: the Atari Learning Environment and the Mu Jo Co physics simulation platform. We discuss the key differences of the various kinds of algorithms, indicate their potential and limitations, as well as provide insights to researchers regarding future directions of the field. |
| Researcher Affiliation | Collaboration | Aristotelis Lazaridis EMAIL Aristotle University of Thessaloniki, School of Informatics Thessaloniki, 54124, Greece Anestis Fachantidis EMAIL Medoid AI 130 Egnatia St, Thessaloniki, 54622, Greece Ioannis Vlahavas EMAIL Aristotle University of Thessaloniki, School of Informatics Thessaloniki, 54124, Greece |
| Pseudocode | No | The paper describes various algorithms and methods (e.g., Deep Q-Learning, Policy Gradient Methods, Model-based Algorithms) in detail but does not present any formal pseudocode or algorithm blocks. The descriptions are provided in paragraph form. |
| Open Source Code | No | The paper is a survey of state-of-the-art Deep Reinforcement Learning algorithms and does not present new original code or methods. It synthesizes and compares existing work, hence, there is no statement about providing open-source code for the methodology described in this paper. |
| Open Datasets | Yes | Then, we present higher-level comparison figures of their performance on 57 games included in the Arcade Learning Environment (ALE) (Bellemare et al., 2013) and the Mu Jo Co physics simulation platform (Todorov et al., 2012), where applicable, as they were originally given by the corresponding authors. The reason for choosing these two platforms is essential to this review, since they have become the gold standard for testing Deep RL methodologies. |
| Dataset Splits | No | The paper refers to 'no-op' and 'human-starts' regimes for evaluating algorithms on the ALE platform, as seen in Section 5.1 and Figure 1 & 2. However, these are evaluation protocols rather than explicit training/test/validation dataset splits with percentages or sample counts. The paper is a survey and compiles results from other works, so it does not detail specific data splitting for its own experiments. |
| Hardware Specification | No | This paper is a survey of state-of-the-art Deep Reinforcement Learning algorithms and presents performance figures extracted from original experiments conducted by other authors. It does not describe its own experimental hardware. |
| Software Dependencies | No | This paper is a comprehensive review of Deep Reinforcement Learning algorithms and does not describe its own experimental software dependencies with specific version numbers, as it synthesizes results from other published works. |
| Experiment Setup | No | This paper is a survey and comparative analysis of existing Deep Reinforcement Learning algorithms. It presents results reported by the original authors of those algorithms and does not detail its own experimental setup or hyperparameter configurations. |