Summary of Orchestrate Latent Expertise: Advancing Online Continual Learning with Multi-level Supervision and Reverse Self-distillation, by Hongwei Yan et al.
Orchestrate Latent Expertise: Advancing Online Continual Learning with Multi-Level Supervision and Reverse Self-Distillation
by HongWei Yan, Liyuan Wang, Kaisheng Ma, Yi Zhong
First submitted to arxiv on: 30 Mar 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
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 Artificial intelligence systems must adapt to online content streams, a more challenging setting than traditional Continual Learning. Online Continual Learning (OCL) methods typically rely on memory replay of old training samples. However, OCL faces the overfitting-underfitting dilemma due to rehearsal buffers, making it harder to learn new tasks while preserving past knowledge. To address this, we propose Multi-level Online Sequential Experts (MOSE), which integrates multi-level supervision and reverse self-distillation. MOSE cultivates a model as stacked sub-experts, facilitating convergence of new tasks and mitigating performance decline of old tasks through knowledge distillation. Our approach demonstrates improved OCL performance on benchmarks like Split CIFAR-100 and Split Tiny-ImageNet, outperforming state-of-the-art baselines by up to 7.3% and 6.1%, respectively. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Imagine a computer program that can learn new things from the internet without forgetting what it already knows. This is hard for AI systems because they have to process information one piece at a time, like reading an article online. Researchers tried to solve this problem by replaying old information in their training, but it didn’t work well. To fix this, scientists created a new way called Multi-level Online Sequential Experts (MOSE). MOSE is like a team of experts working together to learn and remember new things while still keeping what they already know. This approach works much better than previous methods, with improvements of up to 7.3% on certain tests. |
Keywords
» Artificial intelligence » Continual learning » Distillation » Knowledge distillation » Overfitting » Underfitting
