Summary of Asmr: Activation-sharing Multi-resolution Coordinate Networks For Efficient Inference, by Jason Chun Lok Li et al.
ASMR: Activation-sharing Multi-resolution Coordinate Networks For Efficient Inference
by Jason Chun Lok Li, Steven Tin Sui Luo, Le Xu, Ngai Wong
First submitted to arxiv on: 20 May 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
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 paper proposes a novel method called Activation-Sharing Multi-Resolution (ASMR) coordinate network, which enhances the efficiency of implicit neural representations (INRs). INRs are compact neural networks that encode natural signals like images and videos. While previous methods focused on improving encoding capabilities, this work prioritizes inference efficiency, measured by multiply-accumulate (MAC) count. The ASMR model combines multi-resolution coordinate decomposition with hierarchical modulations, allowing it to share activations across grids of the data. This decouples its inference cost from its depth, rendering a near O(1) inference complexity regardless of the number of layers. Experimental results demonstrate that ASMR can reduce MAC by up to 500x compared to a vanilla SIREN model while achieving higher reconstruction quality. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Imagine having a special kind of computer program that can compress and reconstruct images or videos in a really fast and efficient way. This is what the paper is about: creating a new type of neural network called ASMR, which is much faster than existing methods. The key idea is to share information between different parts of the image or video, so that the program doesn’t have to process everything separately. This makes it much faster and more efficient. In experiments, this method was able to compress images and videos up to 500 times faster while still keeping them looking great. |
Keywords
» Artificial intelligence » Inference » Neural network
