Summary of Decomposing the Neurons: Activation Sparsity Via Mixture Of Experts For Continual Test Time Adaptation, by Rongyu Zhang et al.
Decomposing the Neurons: Activation Sparsity via Mixture of Experts for Continual Test Time Adaptation
by Rongyu Zhang, Aosong Cheng, Yulin Luo, Gaole Dai, Huanrui Yang, Jiaming Liu, Ran Xu, Li Du, Yuan Du, Yanbing Jiang, Shanghang Zhang
First submitted to arxiv on: 26 May 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI)
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 The proposed Continual Test-Time Adaptation (CTTA) method aims to adapt pre-trained vision models to ever-evolving target domains, overcoming the issue of catastrophic forgetting. The integration of Mixture-of-Activation-Sparsity-Experts (MoASE) as an adapter for CTTA is explored, drawing inspiration from the human visual system’s ability to process shape and texture according to the Trichromatic Theory. MoASE decomposes neural activation into high-activation and low-activation components using Spatial Differentiate Dropout (SDD), and a multi-gate structure comprising Domain-Aware Gate (DAG) and Activation Sparsity Gate (ASG) is devised for adaptive combination of experts and feature selection threshold assignment. A Homeostatic-Proximal (HP) loss is introduced to bypass error accumulation during continuous adaptation. The proposed methodology achieves state-of-the-art performance in both classification and segmentation CTTA tasks on four prominent benchmarks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper presents a new approach to adapting pre-trained vision models to changing environments. It uses a special type of adapter called MoASE, which helps the model forget old things but remember new things. The idea is inspired by how our eyes work, where we can focus on different details like shape or texture depending on what we’re looking at. The researchers also came up with a way to make sure the model doesn’t get confused when it’s learning new things. They tested their method on several big datasets and found that it works better than other methods. |
Keywords
» Artificial intelligence » Classification » Dropout » Feature selection
