Summary of Revisiting Inverse Hessian Vector Products For Calculating Influence Functions, by Yegor Klochkov and Yang Liu
Revisiting inverse Hessian vector products for calculating influence functions
by Yegor Klochkov, Yang Liu
First submitted to arxiv on: 25 Sep 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
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 proposed method improves the calculation of influence functions for large models by optimizing three hyperparameters: scaling factor, batch size, and number of steps. By leveraging spectral properties of the Hessian, such as its trace and largest eigenvalue, these hyperparameters can be chosen effectively. The approach is compared to Proximal Bregman Retraining Functions (PBRF) and shows promising results. This work has implications for understanding the role of inverse Hessians in calculating influence functions. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary A team of researchers developed a new way to calculate how training data affects a model’s predictions. They found that by adjusting three settings, they could make this calculation more efficient for big models. This is important because it helps us understand how our models are making decisions and can improve their performance. The method was tested against another approach called Proximal Bregman Retraining Functions (PBRF) and showed similar results. |
