Summary of Careful with That Scalpel: Improving Gradient Surgery with An Ema, by Yu-guan Hsieh et al.
Careful with that Scalpel: Improving Gradient Surgery with an EMA
by Yu-Guan Hsieh, James Thornton, Eugene Ndiaye, Michal Klein, Marco Cuturi, Pierre Ablin
First submitted to arxiv on: 5 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Machine Learning (stat.ML)
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 This paper proposes a novel approach for deep learning model estimation, which relies on an auxiliary objective to quantify desirable properties of the model. The authors cast the problem as a constrained minimization problem and develop a method called Bloop that combines the training loss gradient with the orthogonal projection of the auxiliary gradient. This approach, known as gradient surgery, has been shown to improve performance in previous works. By blending the gradients beyond a simple sum, Bloop can lead to better performances on NLP and vision experiments than other methods without an exponential moving average (EMA). |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about finding a better way to train deep learning models by using two kinds of information: what we want the model to learn from one dataset, and how well it agrees with what we already know. The new method, called Bloop, combines these two ideas in a clever way that helps the model work better. This can be useful for things like natural language processing and image recognition. By using this method, we might get better results without needing to use extra information or special tricks. |
Keywords
* Artificial intelligence * Deep learning * Natural language processing * Nlp
