Summary of Sgd with Partial Hessian For Deep Neural Networks Optimization, by Ying Sun et al.
SGD with Partial Hessian for Deep Neural Networks Optimization
by Ying Sun, Hongwei Yong, Lei Zhang
First submitted to arxiv on: 5 Mar 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Optimization and Control (math.OC)
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 proposes a compound optimizer, called SGD with Partial Hessian (SGD-PH), for training deep neural networks (DNNs). By combining a second-order optimizer with a precise partial Hessian matrix and first-order stochastic gradient descent (SGD) for updating other parameters, the method inherits the advantages of both. The proposed optimizer precisely computes the diagonal associated Hessian matrices of channel-wise parameters from Hessian-free methods. This approach avoids approximating the Hessian information, which can result in unstable performance. Experimental results on image classification tasks demonstrate the effectiveness of SGD-PH. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper talks about a new way to help train deep neural networks (which are like super-powerful computers). The problem is that these networks are very hard to optimize, so we need a better way. The author proposes a new method called SGD-PH that combines two other methods together. This helps the network learn faster and more accurately. The results show that this new method works well on image classification tasks. |
Keywords
* Artificial intelligence * Image classification * Stochastic gradient descent
