Summary of Efficient Sample-specific Encoder Perturbations, by Yassir Fathullah et al.
Efficient Sample-Specific Encoder Perturbations
by Yassir Fathullah, Mark J. F. Gales
First submitted to arxiv on: 1 May 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Machine Learning (cs.LG)
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 Medium Difficulty summary: This paper introduces a novel approach to control the behavior of encoder-decoder foundation models by modifying their outputs based on specific attributes. The proposed method uses a small proxy network to find sample-by-sample perturbations that improve decoder performance. The authors demonstrate this framework’s effectiveness on machine translation and speech recognition tasks, achieving state-of-the-art results using COMET and WER evaluation metrics. Specifically, they modify Flan-T5 for Machine Translation and Whisper foundation models for Speech Recognition, showcasing consistent improvements in performance. The proxies are also shown to be robust across different domains. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Low Difficulty summary: This research paper shows how to make machine learning models better at specific tasks by changing their behavior based on certain characteristics. The authors created a new method that uses a small network to find tiny adjustments that improve the model’s performance. They tested this approach on two tasks: translating language and recognizing speech. The results are impressive, with the modified models achieving top scores. This method can be used for different types of data and is robust. |
Keywords
» Artificial intelligence » Decoder » Encoder decoder » Machine learning » T5 » Translation
