Summary of The Implicit Bias Of Structured State Space Models Can Be Poisoned with Clean Labels, by Yonatan Slutzky et al.
The Implicit Bias of Structured State Space Models Can Be Poisoned With Clean Labels
by Yonatan Slutzky, Yotam Alexander, Noam Razin, Nadav Cohen
First submitted to arxiv on: 14 Oct 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 The paper investigates the implicit bias of structured state space models (SSMs), a type of neural network gaining popularity as an efficient alternative to transformers. It is known that SSMs exhibit generalization abilities, but prior work has not fully explored the phenomenon. The authors formally establish a previously undetected occurrence where special training examples can distort the implicit bias, leading to failed generalization despite having clean labels. They demonstrate this phenomenon empirically using SSMs trained independently and as part of non-linear neural networks. This finding is significant in the context of adversarial machine learning, particularly with the proliferation of SSMs in large language models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Structured state space models (SSMs) are a type of neural network that can generalize well to unseen data. Researchers have found that this generalization ability is due to an implicit bias in the training process. However, a new study shows that this implicit bias can be disrupted by special examples in the training set, causing the model to fail to generalize. This means that even with clean labels (where each example is correctly labeled), the model can still make mistakes when making predictions on new data. |
Keywords
» Artificial intelligence » Generalization » Machine learning » Neural network
