I am an Assistant Professor at Georgia Institute of Technology in the School of Electrical and Computer Engineering and Computer Science. Prior to joining Georgia Tech, I was a Senior Researcher at Microsoft Azure’s Cloud Accelerated Systems & Technologies team since September 2019. In Microsoft, I led the research, design, and deployment of communication collectives for massive-scale distributed DNN training.
I received my Ph.D. from Georgia Institute of Technology in 2019 and obtained my Master’s from The University of Texas at Austin (2014). I obtained my Bachelor’s from Indian Institute of Technology Ropar where I was conferred the Presidents of India Gold Medal, the highest academic honor in IITs.
My research has been published in top-tier venues such as ISCA, HPCA, MICRO, ASPLOS, NeurIPS, and VLDB. My dissertation work has been recognized with the NCWIT Collegiate Award, 2017 and distinguished paper award at High Performance Computer Architecture (HPCA), 2016.
I am the founding director of the Systems Infrastructure and Architecture Research Lab. My research team is devising next-generation sustainable compute platforms targeting end-to-end data pipeline for large scale AI and machine learning. The work draws insights from a broad set of disciplines such as, computer architecture, systems, and databases. The work will take a radical approach towards distributed domain-specialized systems that will enable the next generation of massively-large models. We will achieve this by breaking away from the traditional CPU-based stack and integrating closely with data collection and management systems. I aim to offer solutions that are sustainable and can meet the technological demands of emerging economies.
I am always excited to meet students interested in research at the cross section of architecture and systems for data intensive applications. Please send your CV and we can setup sometime to chat.
News
Part of the MLSys 2024 and ISCA 2024 PC
Adnan's paper on Feature Interactions in Recommendation Networks was accepted to Machine Learning for Systems Workshop at NeurIPS 2023
Irene's paper on mitigating stragglers for federated learning (FLuID) has been accepted to NeurIPS 2023
Irene Wang and Seonho Lee have joined the team
Part of the ASPLOS 2024 and Eurosys 2024 PC
My post on Sigarch Blog about uArch Workshop, all undergrads please apply!
Starting September, I will be an assistant professor at Georgia Tech in School of Electrical and Computer Engineering and Computer Science, Go Jackets!!
PhD Students
Irene Wang (SCS)
Seonho Lee (ECE)
Mohammad Adnan (ECE, University of British Columbia) - primarily advised by Prashant Nair
Publications
Integrated Hardware Architecture and Device Placement Search, ICML 2024 (to appear)
Heterogeneous Acceleration Pipeline for Recommendation System Training, ISCA 2024 (to appear)
Accelerating String-key Learned Index Structures via Memoization-based Incremental Training, Very Large Databases (VLDB) 2024, (to appear)
NeuPIMs: NPU-PIM Heterogeneous Acceleration for Batched LLM Inferencing, ASPLOS 2024
Ad-Rec: Advanced Feature Interactions to Address Covariate-Shifts in Recommendation Networks, Machine Learning for Systems Workshop at NeurIPS 2023
FLuID: Mitigating Stragglers in Federated Learning using Invariant Dropout, NeurIPS 2023
Accelerating Recommendation System Training by Leveraging Popular Choices, VLDB 2022
Yin-Yang: Programming Abstractions for Cross-Domain Multi-Acceleration, IEEE Micro 2022
A Computational Stack for Cross-Domain Acceleration, HPCA 2021
Efficient Algorithms for Device Placement of DNN Graph Operators, Neurips 2020
In-RDBMS hardware acceleration of advanced analytics, VLDB 2018
Robox: an end-to-end solution to accelerate autonomous control in robotics, ISCA 2018
Scale-out acceleration for machine learning, MICRO 2017
AxBench: A Multiplatform Benchmark Suite for Approximate Computing, IEEE Design & Test of Computers 2017
From high-level deep neural models to FPGAs, MICRO 2016
Towards statistical guarantees in controlling quality tradeoffs for approximate acceleration, ISCA 2016
AxGames: Towards Crowdsourcing Quality Target Determination in Approximate Computing, ASPLOS 2016
TABLA: A unified template-based framework for accelerating statistical machine learning, HPCA 2016
Axilog: Abstractions for Approximate Hardware Design and Reuse, IEEE Micro 2015
Axilog: language support for approximate hardware design, DATE 2015
Contact
Email: divya [dot] mahajan [at] gatech [dot] edu
Klaus Advanced Computing Building, Office 2358, Atlanta, GA, 30332