Summary of Gradient-based Automatic Mixed Precision Quantization For Neural Networks On-chip, by Chang Sun et al.
Gradient-based Automatic Mixed Precision Quantization for Neural Networks On-Chip
by Chang Sun, Thea K. Årrestad, Vladimir Loncar, Jennifer Ngadiuba, Maria Spiropulu
First submitted to arxiv on: 1 May 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Instrumentation and Detectors (physics.ins-det)
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 This research paper proposes High Granularity Quantization (HGQ), an innovative method for quantizing neural networks to reduce their size and inference speed at deployment time. By making per-weight and per-activation precision optimizable through gradient descent, HGQ enables ultra-low latency and low power neural networks on hardware capable of performing arithmetic operations with an arbitrary number of bits. The authors demonstrate that HGQ outperforms existing methods by a substantial margin, achieving up to 20 times reduction in resources and 5 times improvement in latency while preserving accuracy. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This research is about making computer programs called neural networks smaller and faster so they can be used on devices with limited power or speed. The problem is that these networks get bigger and slower when they’re sent to a device, which makes them harder to use. To solve this, the researchers developed a new way of “compressing” the network called High Granularity Quantization (HGQ). HGQ lets the computer choose how precise each part of the network should be, so it can make the network smaller and faster without losing its ability to work correctly. |
Keywords
» Artificial intelligence » Gradient descent » Inference » Precision » Quantization
