Mnemosyne: Learning to Train Transformers with Transformers
Jain, Deepali, Choromanski, Krzysztof Marcin, Dubey, Avinava, Singh, Sumeet, Sindhwani, Vikas, Zhang, Tingnan, Tan, Jie
–arXiv.org Artificial Intelligence
In this work, we propose a new class of learnable optimizers, called \textit{Mnemosyne}. It is based on the novel spatio-temporal low-rank implicit attention Transformers that can learn to train entire neural network architectures, including other Transformers, without any task-specific optimizer tuning. We show that Mnemosyne: (a) outperforms popular LSTM optimizers (also with new feature engineering to mitigate catastrophic forgetting of LSTMs), (b) can successfully train Transformers while using simple meta-training strategies that require minimal computational resources, (c) matches accuracy-wise SOTA hand-designed optimizers with carefully tuned hyper-parameters (often producing top performing models). Furthermore, Mnemosyne provides space complexity comparable to that of its hand-designed first-order counterparts, which allows it to scale to training larger sets of parameters. We conduct an extensive empirical evaluation of Mnemosyne on: (a) fine-tuning a wide range of Vision Transformers (ViTs) from medium-size architectures to massive ViT-Hs (36 layers, 16 heads), (b) pre-training BERT models and (c) soft prompt-tuning large 11B+ T5XXL models. We complement our results with a comprehensive theoretical analysis of the compact associative memory used by Mnemosyne which we believe was never done before.
arXiv.org Artificial Intelligence
Jun-16-2023
- Country:
- Europe > Switzerland
- North America > United States
- California > Los Angeles County
- Long Beach (0.14)
- Minnesota > Hennepin County
- Minneapolis (0.14)
- New York > New York County
- New York City (0.14)
- California > Los Angeles County
- Oceania (0.67)
- Genre:
- Research Report > New Finding (0.87)
- Technology: