Summary of Heterogeneous Learning Rate Scheduling For Neural Architecture Search on Long-tailed Datasets, by Chenxia Tang
Heterogeneous Learning Rate Scheduling for Neural Architecture Search on Long-Tailed Datasets
by Chenxia Tang
First submitted to arxiv on: 11 Jun 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper addresses the challenge of applying Neural Architecture Search (NAS) algorithms, specifically Differentiable Architecture Search (DARTS), to long-tailed datasets where class distribution is highly imbalanced. Traditional re-sampling and re-weighting techniques used for standard classification tasks lead to performance degradation when combined with DARTS. To mitigate this, the authors propose a novel adaptive learning rate scheduling strategy tailored for DARTS architecture parameters integrated with Bilateral Branch Network (BBN) for handling imbalanced datasets. The method dynamically adjusts the learning rate of architecture parameters based on training epoch, preventing disruption of well-trained representations in later stages of training. The impact of branch mixing factors on algorithm’s performance is also explored. Through extensive experiments on CIFAR-10 dataset with artificially induced long-tailed distribution, the authors demonstrate that their method achieves comparable accuracy to using DARTS alone. Experiment results suggest that re-sampling methods inherently harm the performance of DARTS algorithm. The findings highlight the importance of careful data augmentation when applying DNAS to imbalanced learning scenarios. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper tries to make Neural Architecture Search (NAS) work better on datasets where one class has way more examples than others. Right now, NAS doesn’t do well with these kinds of datasets because it uses tricks that help standard classification tasks but actually hurt its performance. To fix this, the authors came up with a new way to adjust how fast the algorithm learns as it’s training. This helps prevent it from messing up things it already figured out. They also tested different ways to mix together parts of the network. The results show that their method does just as well as using NAS alone, but only if you don’t use those old tricks. It’s important to be careful when trying to make NAS work better on imbalanced datasets. |
Keywords
» Artificial intelligence » Classification » Data augmentation
