Learning Task Informed Abstractions
Authors: Xiang Fu, Ge Yang, Pulkit Agrawal, Tommi Jaakkola
ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirical evaluation shows that TIA leads to significant performance gains over state-of-the-art methods on many visual control tasks where natural and unconstrained visual distractions pose a formidable challenge. |
| Researcher Affiliation | Academia | *Equal contribution 1MIT CSAIL 2IAIFI. Correspondence to: Xiang Fu <xiangfu@csail.mit.edu>. |
| Pseudocode | No | The paper describes the learning process and model components textually and mathematically but does not include any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Our code is available at https://github.com/kyonofx/tia. |
| Open Datasets | Yes | Kinematic Control with Natural Video Distraction (Figure 4b) We consider the Deep Mind Control (DMC) suite with natural video background from the Kinetics dataset (Kay et al., 2017) used in prior work (Zhang et al., 2021)... Arcade Learning Environments (ALE) or Atari, (Figure 6) is a standard benchmark for vision-based control. |
| Dataset Splits | No | The paper mentions using specific environments (Many World, DMC, Atari) and reports results, but it does not provide specific training, validation, or test dataset splits (e.g., percentages or sample counts) for reproducibility. |
| Hardware Specification | Yes | Each seed takes ten days on a V100 Volta GPU. |
| Software Dependencies | No | The paper does not provide specific version numbers for any software dependencies or libraries used in the experiments. |
| Experiment Setup | No | The paper discusses the overall approach and mentions certain parameters like λRadv and λOs conceptually in Section 4.5 regarding their balance. However, it explicitly states, 'without any hyperparameter tuning (i.e., just a single value chosen based on the intuition described in Section 4.5)', indicating that specific numerical values for key hyperparameters (e.g., learning rate, batch size, number of epochs) are not provided in the main text for TIA's implementation. |