Summary of Resilience Of Entropy Model in Distributed Neural Networks, by Milin Zhang et al.
Resilience of Entropy Model in Distributed Neural Networks
by Milin Zhang, Mohammad Abdi, Shahriar Rifat, Francesco Restuccia
First submitted to arxiv on: 1 Mar 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR)
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 technique to reduce communication overhead in edge computing systems by introducing entropy coding into distributed deep neural networks (DNNs). The authors train the DNN jointly with an entropy model, which is used during inference time to adaptively encode latent representations. However, the resilience of these entropy models to intentional and unintentional interference has not been investigated. This paper fills this gap by formulating and investigating the resilience of entropy models to adversarial attacks and environmental changes. The authors demonstrate that entropy attacks can increase communication overhead by up to 95%. To combat this, they propose a new defense mechanism that separates compression features in frequency and spatial domains, reducing transmission overhead by about 9% with only 2% accuracy loss. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about making computers communicate more efficiently. It’s like sending a message through the internet. The researchers came up with a new way to make this process faster and better. They did this by combining two ideas: deep learning, which is like teaching computers to learn from examples, and entropy coding, which is like finding the most efficient way to send information. The problem is that this new method can be affected by bad things happening while it’s sending the message, like someone trying to hack into it or a storm knocking out the internet connection. The researchers tested their idea and found that it works pretty well, but sometimes it can get slowed down quite a bit. To fix this, they came up with another idea: separating the compression features so that important information gets sent first. This makes the message get through faster, even if there are some problems along the way. |
Keywords
* Artificial intelligence * Deep learning * Inference
