Summary of Packd: Pattern-clustered Knowledge Distillation For Compressing Memory Access Prediction Models, by Neelesh Gupta et al.
PaCKD: Pattern-Clustered Knowledge Distillation for Compressing Memory Access Prediction Models
by Neelesh Gupta, Pengmiao Zhang, Rajgopal Kannan, Viktor Prasanna
First submitted to arxiv on: 21 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: 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 The proposed PaCKD approach compresses Memory Access Prediction (MAP) models while maintaining prediction performance, addressing the limitations of existing DNN-based MAP models. By clustering memory access sequences into distinct partitions and training pattern-specific teacher models for each partition, the approach achieves a 552x model size compression with an F1-score of 0.4538 on LSTM, MLP-Mixer, and ResNet models in graph applications. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary PaCKD is a new way to make computer memory work better by using patterns in data to predict what’s needed next. This helps reduce delays in accessing memory, which is important for many tasks like image classification. The approach uses three steps: grouping similar patterns together, training special teacher models for each group, and then training a smaller student model that can do the same job as well as the teachers. Tests show that PaCKD works better than other methods, with a significant reduction in the number of calculations needed while still being accurate. |
Keywords
* Artificial intelligence * Clustering * F1 score * Image classification * Lstm * Resnet * Student model
