Summary of Erasure Coded Neural Network Inference Via Fisher Averaging, by Divyansh Jhunjhunwala et al.
Erasure Coded Neural Network Inference via Fisher Averaging
by Divyansh Jhunjhunwala, Neharika Jali, Gauri Joshi, Shiqiang Wang
First submitted to arxiv on: 2 Sep 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 Erasure-coded computing has been successfully applied in cloud systems to mitigate tail latency caused by straggling servers and heterogeneous traffic variations. However, current techniques are primarily designed for linear computations like matrix-vector multiplications and do not account for highly non-linear neural network functions. This paper proposes a method to code over neural networks, given two or more models, to create a coded model that outputs a linear combination of the original outputs. The problem is formulated as a KL barycenter problem, and the proposed algorithm COIN leverages diagonal Fisher information to create an approximately accurate coded model. Experimental results on real-world vision datasets demonstrate that COIN outperforms baselines in accuracy while being extremely compute-efficient. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Imagine trying to access data from multiple servers at once, but some of those servers are slow or unreliable. This is a problem in cloud computing, where many people need to access information quickly. Currently, there are ways to fix this by using special codes that help speed up the process. But these codes were designed for simple calculations and don’t work well with complex tasks like analyzing pictures. In this paper, scientists try to develop a new code that works better with more complicated tasks. They tested it on real-world images and found that their code was much faster and more accurate than previous methods. |
Keywords
» Artificial intelligence » Neural network
