Summary of Progsg: Cross-modality Representation Learning For Programs in Electronic Design Automation, by Yunsheng Bai et al.
ProgSG: Cross-Modality Representation Learning for Programs in Electronic Design Automation
by Yunsheng Bai, Atefeh Sohrabizadeh, Zongyue Qin, Ziniu Hu, Yizhou Sun, Jason Cong
First submitted to arxiv on: 18 May 2023
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Programming Languages (cs.PL)
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 paper proposes a new approach to designing domain-specific accelerators (DSAs) for accelerating various applications like deep learning and autonomous driving. To achieve this, it leverages high-level synthesis (HLS) tools that compile software code into low-level hardware description languages. However, existing HLS tools require manual microarchitecture decisions, which can be time-consuming. The authors aim to automate these decisions using deep learning techniques for predicting the quality of HLS designs. They introduce ProgSG, a novel method that combines sequence data (source code) and graph modalities (control data flow graphs) in a fine-grained manner. To alleviate the lack of labeled designs, the authors propose a pre-training method based on compiler’s data flow analysis tasks. Experimental results show that ProgSG outperforms baseline methods that only consider one modality or combine them without utilizing alignment information. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps us design special computers called domain-specific accelerators (DSAs) for things like artificial intelligence and self-driving cars. To do this, it uses tools that take software code and turn it into a language that computers can understand. The problem is that these tools need people to make decisions about how the computer should work, which can be slow. The authors want to use special AI techniques to help make those decisions faster and better. They created something called ProgSG that combines two different ways of looking at code: as a list of instructions (sequence data) and as a flowchart-like graph. This helps computers understand the code better. To get more accurate results, they also came up with a way to train the AI using special tasks from compilers. |
Keywords
* Artificial intelligence * Alignment * Deep learning
