Summary of Adjusting Pretrained Backbones For Performativity, by Berker Demirel et al.
Adjusting Pretrained Backbones for Performativity
by Berker Demirel, Lingjing Kong, Kun Zhang, Theofanis Karaletsos, Celestine Mendler-Dünner, Francesco Locatello
First submitted to arxiv on: 6 Oct 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 This paper proposes a novel technique to adjust pre-trained backbones for performativity, enabling better sample efficiency and reusing existing deep learning assets. The method focuses on performative label shift, training a shallow adapter module to correct the backbone’s logits given a sufficient statistic of the model to be deployed. This decouples input-specific feature embeddings from performativity mechanisms. The framework is evaluated under adversarial sampling for vision and language tasks, showing smaller loss along retraining trajectories and enabling effective selection among candidate models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps deep learning models work better in different situations by adjusting how they make predictions. It’s like having a special tool that lets you fine-tune an old model to do well in new scenarios. The method is tested with images and words, showing it can help choose the best model for a job. |
Keywords
» Artificial intelligence » Deep learning » Logits
