Summary of Learning to (learn at Test Time): Rnns with Expressive Hidden States, by Yu Sun et al.
Learning to (Learn at Test Time): RNNs with Expressive Hidden States
by Yu Sun, Xinhao Li, Karan Dalal, Jiarui Xu, Arjun Vikram, Genghan Zhang, Yann Dubois, Xinlei Chen, Xiaolong Wang, Sanmi Koyejo, Tatsunori Hashimoto, Carlos Guestrin
First submitted to arxiv on: 5 Jul 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
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 new sequence modeling layers called Test-Time Training (TTT) layers that combine linear complexity with expressive power. The key innovation is treating the hidden state as a machine learning model itself, updated through self-supervised learning on test sequences. Two instantiations are presented: TTT-Linear and TTT-MLP, which match or exceed strong Transformer and Mamba baselines on large-scale datasets. Unlike RNNs, TTT layers can continue reducing perplexity by conditioning on more tokens. Preliminary results show promising speedup potential, with the linear variant already outperforming Transformer at 8k context. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about creating new ways to analyze and understand long sequences of data. Current methods have limitations when dealing with very long sequences. The researchers propose a new approach called Test-Time Training (TTT) that combines the best of both worlds: being able to handle very long sequences while still being efficient. Two versions are explored, each trying different techniques to make the analysis more effective. Early results show great promise, especially for handling very long sequences. |
Keywords
» Artificial intelligence » Machine learning » Perplexity » Self supervised » Transformer
