Summary of Improve Generalization Ability Of Deep Wide Residual Network with a Suitable Scaling Factor, by Songtao Tian et al.
Improve Generalization Ability of Deep Wide Residual Network with A Suitable Scaling Factor
by Songtao Tian, Zixiong Yu
First submitted to arxiv on: 7 Mar 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Statistics Theory (math.ST)
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 explores ways to improve the generalization ability of Deep Residual Neural Networks (ResNets) by identifying an optimal scaling factor for the residual branch. Researchers show that if this factor remains constant, the network’s induced functions become increasingly difficult to learn as the depth increases. Surprisingly, even allowing the factor to decrease with depth doesn’t always prevent this issue. However, when the factor decreases rapidly, kernel regression with early stopping can achieve optimal performance for certain target functions. The study uses synthetic data and real-world classification tasks like MNIST, CIFAR10, and CIFAR100 to support its findings. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Deep Residual Neural Networks are super smart at doing lots of jobs. But sometimes they don’t do as well on new tasks because they get too good at memorizing old ones. This paper figures out how to make them better by finding the right “secret ingredient” for their hidden layers. It shows that if you use this secret ingredient, ResNets can learn really quickly and accurately. The study also finds some surprising things about how these networks work and what makes them successful. |
Keywords
* Artificial intelligence * Classification * Early stopping * Generalization * Regression * Synthetic data
