Interpreting Operation Selection in Differentiable Architecture Search: A Perspective from Influence-Directed Explanations
Authors: Miao Zhang, Wei Huang, Bin Yang
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirical studies across different tasks on several spaces show that vanilla DARTS and its variants can avoid most failures by leveraging the proposed theory-driven operation selection criterion. |
| Researcher Affiliation | Academia | Miao Zhang1,2, Wei Huang3, Bin Yang4 1Harbin Institute of Technology (Shenzhen) 2Aalborg University 3RIKEN AIP 4East China Normal University |
| Pseudocode | No | The paper provides mathematical derivations and explanations but does not include any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper's author checklist indicates code is available (question 3a: 'Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [Yes]'), but the main body of the paper does not contain an explicit statement or link confirming the release of their source code for the methodology described. |
| Open Datasets | Yes | We consider two NAS benchmark dataset search space, NAS-Bench-201 [12] and NAS-Bench-1shot1 [46], and the most popular DARTS search space [27]. |
| Dataset Splits | No | The paper mentions 'validation loss' and evaluates models on datasets like NAS-Bench-201 and NAS-Bench-1shot1, which typically have predefined splits. However, it does not explicitly provide specific details on the train/validation/test dataset splits used for reproduction, such as percentages, sample counts, or explicit confirmation of using standard predefined splits from the benchmarks. |
| Hardware Specification | No | The paper refers to computational resources in general terms, such as 'GPU days' and 'additional GPU day', but it does not provide specific hardware details like GPU models, CPU types, or other detailed computer specifications used for running experiments. |
| Software Dependencies | No | The paper does not provide specific ancillary software details, such as library names with version numbers (e.g., PyTorch version, specific solver versions), needed to replicate the experiments. |
| Experiment Setup | No | The paper states that for reproducibility, they 'only replace the selection part in DARTS with IM and keep the remaining part identical,' and mentions 'different random seeds,' but it does not provide specific experimental setup details such as hyperparameter values (e.g., learning rate, batch size, number of epochs, optimizer settings) in the main text. |