Summary of Adakd: Dynamic Knowledge Distillation Of Asr Models Using Adaptive Loss Weighting, by Shreyan Ganguly et al.
AdaKD: Dynamic Knowledge Distillation of ASR models using Adaptive Loss Weighting
by Shreyan Ganguly, Roshan Nayak, Rakshith Rao, Ujan Deb, Prathosh AP
First submitted to arxiv on: 11 May 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 Adaptive Knowledge Distillation, a novel technique for model compression that adapts the weights assigned to task-specific and knowledge distillation losses at the instance level. The authors draw inspiration from curriculum learning, where sample difficulty increases with teacher loss, and design a plug-and-play method that can be applied on top of any task-specific and distillation objectives. Experimental results show that Adaptive Knowledge Distillation outperforms conventional knowledge distillation methods and existing instance-level loss functions. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about making AI models smaller and faster without losing their ability to perform well. Right now, there are ways to do this called knowledge distillation. But these methods give equal importance to two different goals: doing the main task and learning from a bigger teacher model. This can lead to bad results. The authors of this paper want to improve things by giving more weight to one goal or the other depending on how hard each individual example is. They call their new method Adaptive Knowledge Distillation, and it’s like a special filter that helps make AI models better. |
Keywords
» Artificial intelligence » Curriculum learning » Distillation » Knowledge distillation » Model compression » Teacher model
