Reawakening knowledge: Anticipatory recovery from catastrophic interference via structured training
Fuente:
arXiv
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866910711387521024 |
|---|---|
| author | Yang, Yanlai Jones, Matt Mozer, Michael C. Ren, Mengye |
| author_facet | Yang, Yanlai Jones, Matt Mozer, Michael C. Ren, Mengye |
| contents | We explore the training dynamics of neural networks in a structured non-IID setting where documents are presented cyclically in a fixed, repeated sequence. Typically, networks suffer from catastrophic interference when training on a sequence of documents; however, we discover a curious and remarkable property of LLMs finetuned sequentially in this setting: they exhibit anticipatory behavior, recovering from the forgetting on documents before encountering them again. This behavior occurs even though the documents are never presented in context together. The behavior emerges and becomes more robust as the architecture scales up its number of parameters. Through comprehensive experiments and visualizations, we demonstrate a new mechanism by which over-parametrized neural networks can recover from catastrophic interference and uncover new insights into training over-parameterized networks in cyclically structured environments. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2403_09613 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Reawakening knowledge: Anticipatory recovery from catastrophic interference via structured training Yang, Yanlai Jones, Matt Mozer, Michael C. Ren, Mengye Machine Learning Computation and Language We explore the training dynamics of neural networks in a structured non-IID setting where documents are presented cyclically in a fixed, repeated sequence. Typically, networks suffer from catastrophic interference when training on a sequence of documents; however, we discover a curious and remarkable property of LLMs finetuned sequentially in this setting: they exhibit anticipatory behavior, recovering from the forgetting on documents before encountering them again. This behavior occurs even though the documents are never presented in context together. The behavior emerges and becomes more robust as the architecture scales up its number of parameters. Through comprehensive experiments and visualizations, we demonstrate a new mechanism by which over-parametrized neural networks can recover from catastrophic interference and uncover new insights into training over-parameterized networks in cyclically structured environments. |
| title | Reawakening knowledge: Anticipatory recovery from catastrophic interference via structured training |
| topic | Machine Learning Computation and Language |
| url | https://arxiv.org/abs/2403.09613 |