Task-Adaptive Neural Network Search with Meta-Contrastive Learning
Authors: Wonyong Jeong, Hayeon Lee, Geon Park, Eunyoung Hyung, Jinheon Baek, Sung Ju Hwang
NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We validate the effectiveness and efficiency of our method on ten real-world datasets, against existing NAS/Auto ML baselines. The results show that our method instantly retrieves networks that outperform models obtained with the baselines with significantly fewer training steps to reach the target performance, thus minimizing the total cost of obtaining a task-optimal network. |
| Researcher Affiliation | Collaboration | Wonyong Jeong1,2 Hayeon Lee1,2 Geon Park1,2 Eunyoung Hyung1,2 Jinheon Baek1 Sung Ju Hwang1,2 KAIST1, AITRICS2, Seoul, South Korea |
| Pseudocode | Yes | For the full algorithm, please refer to Appendix A. |
| Open Source Code | Yes | Our code and the model-zoo are available at https://github.com/wyjeong/TANS. |
| Open Datasets | Yes | We collect 96 real-world image classification datasets from Kaggle*. Then we divide the datasets into two non-overlapping sets for 86 meta-training and 10 meta-test datasets. *https://www.kaggle.com/ |
| Dataset Splits | No | For each dataset, we use randomly sampled 80/20% instances as a training and test set. The paper specifies train/test splits for instances within each dataset (80/20%) and meta-training/meta-test splits for datasets, but does not explicitly state a separate validation split percentage or methodology for model training within each dataset. |
| Hardware Specification | No | The paper mentions 'GPU hours' but does not provide specific hardware details such as GPU models, CPU models, or memory specifications used for running the experiments. |
| Software Dependencies | No | The paper does not provide specific software dependency details, such as library names with version numbers (e.g., 'Python 3.8, PyTorch 1.9, and CUDA 11.1'), required to replicate the experiments. |
| Experiment Setup | Yes | We compare 50-epoch accuracy between TANS and the existing NAS methods on 10 novel real-world datasets. For a fair comparison, we train PC-DARTS [65] and Dr NAS [10] for 10 times more epochs (500)... For FBNet, OFA, and Meta D2A... we fine-tune them on the meta-test query datasets for 50 epochs. Please see Appendix B for further detailed descriptions of the experimental setup. |