Summary of Longhorn: State Space Models Are Amortized Online Learners, by Bo Liu et al.
Longhorn: State Space Models are Amortized Online Learners
by Bo Liu, Rui Wang, Lemeng Wu, Yihao Feng, Peter Stone, Qiang Liu
First submitted to arxiv on: 19 Jul 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 This paper presents a novel approach to building state-space models (SSMs) for sequence modeling, which offers linear decoding efficiency while maintaining parallelism during training. The authors draw inspiration from online learning and formulate precise objectives that drive the design of SSMs as meta-modules for specific problems. This insight leads to the introduction of Longhorn, a deep SSM architecture that updates its state using closed-form solutions derived from solving online associative recall problems. Experimental results show that Longhorn outperforms state-of-the-art SSMs, including Mamba, on standard sequence modeling benchmarks, language modeling, and vision tasks, achieving a 1.8x improvement in sample efficiency compared to Mamba. This work has significant implications for the development of large-scale language models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about creating new models that can quickly process long sequences of information. The current way of doing this uses an architecture called Transformer, but it’s not very efficient. The authors came up with a different approach using something called state-space models. They showed that these new models are better than the old ones and can handle longer sequences. This is important because we need to be able to process more information in less time. |
Keywords
* Artificial intelligence * Online learning * Recall * Transformer
