Summary of Towards Cheaper Inference in Deep Networks with Lower Bit-width Accumulators, by Yaniv Blumenfeld et al.
Towards Cheaper Inference in Deep Networks with Lower Bit-Width Accumulators
by Yaniv Blumenfeld, Itay Hubara, Daniel Soudry
First submitted to arxiv on: 25 Jan 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Hardware Architecture (cs.AR)
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 focuses on improving the efficiency of Deep Neural Networks (DNNs) without sacrificing their accuracy. The majority of existing work has focused on reducing the precision of tensors visible by high-level frameworks, but this approach still relies heavily on high-accuracy core operations like accumulating products. To address this limitation, the authors propose a simple method for training and fine-tuning DNNs to enable the use of cheaper, 12-bit accumulators without compromising performance. Additionally, the study demonstrates that further reducing accumulation precision can be achieved through the use of fine-grained gradient approximations, ultimately leading to improved DNN accuracy. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This research is all about making Deep Neural Networks (DNNs) more efficient and accurate. Right now, most people focus on making DNNs less precise, but that doesn’t help when computers still need high-precision calculations. The problem is that adding up lots of numbers gets very slow. This study shows how to train special kinds of DNNs to use cheaper addition methods without losing accuracy. It also explains how using better estimates for changes in the DNN can even improve its performance. |
Keywords
* Artificial intelligence * Fine tuning * Precision
