Summary of Disaggregated Multi-tower: Topology-aware Modeling Technique For Efficient Large-scale Recommendation, by Liang Luo et al.
Disaggregated Multi-Tower: Topology-aware Modeling Technique for Efficient Large-Scale Recommendation
by Liang Luo, Buyun Zhang, Michael Tsang, Yinbin Ma, Ching-Hsiang Chu, Yuxin Chen, Shen Li, Yuchen Hao, Yanli Zhao, Guna Lakshminarayanan, Ellie Dingqiao Wen, Jongsoo Park, Dheevatsa Mudigere, Maxim Naumov
First submitted to arxiv on: 1 Mar 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Distributed, Parallel, and Cluster Computing (cs.DC); Information Retrieval (cs.IR)
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 Disaggregated Multi-Tower (DMT) technique addresses inefficiencies in deep learning recommendation models by decomposing the global embedding lookup process into disjoint towers that exploit data center locality. This is achieved through a novel training paradigm, Semantic-preserving Tower Transform (SPTT), which reduces model complexity and communication volume via hierarchical feature interaction. The DMT approach consists of three components: SPTT, Tower Module (TM), and Tower Partitioner (TP). Experimental results show that DMT can achieve up to 1.9x speedup compared to state-of-the-art baselines without sacrificing accuracy across multiple generations of hardware at large data center scales. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Deep learning recommendation models are used in many applications, but they can be slow and inefficient. To fix this, researchers propose a new way to train these models that takes advantage of the structure of data centers. They call it Disaggregated Multi-Tower (DMT). DMT works by breaking down the model into smaller parts that can run independently on different computers in the data center. This reduces the amount of information that needs to be shared between computers, making the whole process faster and more efficient. |
Keywords
* Artificial intelligence * Deep learning * Embedding
