Bio: I am a Senior Research Scientist at Adobe Research, Bangalore. My current research lies at the intersection of computer systems and machine learning, with a focus on efficiency and scalability. Previously, my work primarily addressed improving the performance and reliability of cloud and distributed systems and improving scalability of Big-Data processing and Recommender Systems. My research contributions include approximately 40 papers published in top computer systems conferences e.g. USENIX NSDI, EuroSys, USENIX ATC, SenSys, SIGMOD, VLDB, PLDI, and top artificial intelligence conferences e.g. AAAI, ICML, NeurIPS, ACL, and ECCV. I have co-invented 20+ US patents.
I received my Ph.D. in Electrical and Computer Engineering from Purdue University, West Lafayette, MS in Computer Engineering from University of Florida, Gainesville and BE in Electronics and Telecommunication Engineering from Jadavpur University, Kolkata.
Prior to Adobe Research, I had spent time as research interns at Microsoft Research – Redmond, AT&T Research – New Jersey and Lawrence Livermore National Labs - Livermore. Even prior to that, I worked in Software Engineering roles at Intel, Santa Clara and Atrenta (now Synopsys) on new product development on Electronic Design Automation.
Research Interests: AI-Systems Design, Efficient ML, Cloud Computing, Generative AI efficiency
I am currently interested in systems and pipelines for co-pilots and agentic frameworks, workflow optimization, efficiency of co-pilot/AI-assitant systems.
Note to prospective PhD interns: I am always looking for PhD students who is interested in interning at Adobe Research, Bangalore, on "Systems for ML" and "Optimizing Big-Data processing using ML". If you are a passionate PhD student, interested in the above topics "and" have co-authored at least one paper in top Systems (SIGMOD/VLDB/NSDI/OSDI/ATC etc. ) or ML/AI/NLP (ICML/NeurIPS/ACL/AAAI etc. ) conferences, please drop me an email.
NEWS
I recently gave a talk at IISc in the Network Seminar Series on "Redefining Caching for Generative AI"
Link: https://cni.iisc.ac.in/seminars/2025-02-11/
We have 2 papers accepted at SIGMOD-2025: “Cache-Craft: Managing Chunk-Caches for Efficient Retrieval-Augmented Generation” and
"Jade: Design Independence Via Physical Visualization Design"
Paper accepted at VLDB-2025: "FaDE: More Than a Million What-ifs Per Second"
Paper accepted at NAACL-HLT-2024: "Personalized Federated Learning for Text Classification with Gradient-Free Prompt Tuning"
Paper accepted at ECCV 2024: " ReCON: Training-Free Acceleration for Text-to-Image Synthesis with Retrieval of Concept Prompt Trajectories"
I was recently featured in a spotlight blog from Adobe Research:
Link: https://research.adobe.com/news/researcher-spotlight-subrata-mitra-uses-machine-learning-to-make-big-data-systems-more-efficient-and-reliable
Paper accepted at PAKDD 2024 (Oral): "ScaleViz: Scaling Visualization Recommendation Models on Large Data" [Adobe Research]
Paper accepted at NSDI 2024: "Approximate Caching for Efficiently Serving Text-to-Image Diffusion Models" [Adobe Research]
Paper accepted at SIGMOD 2024: "R2D2: Reducing Redundancy and Duplication in Data Lakes" [Adobe Research]
Research Snippets:
Text-to-image diffusion-models are slow. These go through a large number of iterative denoising steps, and are resource-intensive, requiring expensive GPUs and incurring considerable latency.
We introduce a novel approximate-caching technique that can reduce such iterative denoising steps by reusing intermediate noise states created during a prior image generation. Based on this idea, we present an end-to-end text-to-image generation system, NIRVANA, that uses approximate-caching with a novel cache management policy to provide 21% GPU compute savings, 19.8% end-to-end latency reduction, and 19% dollar savings on two real production workloads. We further present an extensive characterization of real production text-to-image prompts from the perspective of caching, popularity and reuse of intermediate states in a large production environment.
[NSDI-2024 "Approximate Caching for Efficiently Serving Text-to-Image Diffusion Models"
Enterprise data lakes often suffer from substantial amounts of duplicate and redundant data, with data volumes ranging from terabytes to petabytes. In this work, we focus on identifying and reducing redundancy in enterprise data lakes by addressing the problem of “dataset containment".
We propose R2D2: a three-step hierarchical pipeline that efficiently identifies almost all instances of containment by progressively reducing the search space. It builds (i) a schema containment graph, followed by (ii) statistical min-max pruning, and finally, (iii) content level pruning. We further propose minimizing the total storage and access costs by optimally identifying redundant datasets that can be deleted while respecting latency constraints.
[SIGMOD-2024] "R2D2: Reducing Redundancy and Duplication in Data Lakes"
We are the first to consider the impact of sampling in interactive data exploration settings as they introduce approximation errors. We ApproxEDA (AAAI 2023 paper) that uses a Deep Reinforcement Learning (DRL) based framework which can optimize the sample selection in order to keep the analysis and insight generation flow intact.
[AAAI-2023] "Reinforced Approximate Exploratory Data Analysis"
We have created a query processing systems called Electra (AAAI 2022 paper) that can provide low latency as well as high accuracy for big-data queries without incurring recurring backend compute cost on the cloud.
Electra is the first to utilize conditional generative ML models (a tailored version of conditional VAE) that generates very small samples on the fly, according to the predicates in the query. Then it executes the query only on those samples --- all at the client-side.
[AAAI-2022] "Conditional Generative Model based Predicate-Aware Query Approximation"
How to create a scheduler that can automatically learn to find best placement for diurnal workloads, i.e. the workloads that vary across different times of the day? We have proposed a system called TVW-RL (AAAI 2021 paper) that uses Reinforcement Learning techniques to find optimal placement of such workloads so that multiple workloads do not spike for the same resource at the same node at the same time.
[AAAI 2021] "Scheduling of Time-Varying Workloads using Reinforcement Learning"
Publications (Chronologically) :
A* conference (according to CSRankings.org) are in green.
SIGMOD-2025: “Cache-Craft: Managing Chunk-Caches for Efficient Retrieval-Augmented Generation”
Authors: Shubham Agarwal∗ (Adobe Research), Sai Sundaresan∗(Adobe Research), Subrata Mitra (Adobe Research), Debabrata Mahapatra (Adobe Research), Archit Gupta (IIT Bombay), Rounak Sharma (IIT Kanpur), Nirmal Joshua Kapu (IIT Kanpur), Tong Yu (Adobe Research), Shiv Kumar Saini (Adobe Research)
SIGMOD-2025: "Jade: Design Independence Via Physical Visualization Design"
Authors: Yiru Chen (Columbia University), Xupeng Li (Columbia University), Jeffrey Tao (University of Pennsylvania), Alana Ramjit (Cornell Tech), Ravi Netravali (Princeton University), Subrata Mitra (Adobe Research), Aditya Parameswaran (University of California, Berkeley), Javad Ghaderi (Columbia University), Dan Rubenstein (Columbia University), Eugene Wu (Columbia University)
VLDB-2025: "FaDE: More Than a Million What-ifs Per Second"
Authors: Haneen Mohammed (Columbia University), Alexander Yao (Columbia University), Charlie Summers (Columbia University), Hongbin Zhong (Georgia Institute of Technology), Gromit Yeuk-Yin Chan (Adobe Research), Subrata Mitra (Adobe Research), Lampros Flokas (Celonis Inc.), Eugene Wu (Columbia University)
NAACL-HLT-2024: "Personalized Federated Learning for Text Classification with Gradient-Free Prompt Tuning"
ECCV 2024: " ReCON: Training-Free Acceleration for Text-to-Image Synthesis with Retrieval of Concept Prompt Trajectories"
NSDI-2024 "Approximate Caching for Efficiently Serving Text-to-Image Diffusion Models"
PAKDD 2024 (Oral): "ScaleViz: Scaling Visualization Recommendation Models on Large Data"
SIGMOD-2024 "R2D2: Reducing Redundancy and Duplication in Data Lakes"
VLDB-2023 "SEIDEN: Revisiting Query Processing in Video Database Systems"
ICDCS-2023 "Stash: A Comprehensive Stall-Centric Characterization of Public Cloud VMs for Distributed Deep Learning"
ICML-2023 "Flash: Concept Drift Adaptation in Federated Learning"
ACL (Findings)-2023 "Federated Domain Adaptation for Named Entity Recognition via Distilling with Heterogeneous Tag Sets"
[SIGMOD-2023] (Demo) "Fast Natural Language Based Data Exploration with Sample". In: SIGMOD 2023 Demo Track #ML, #Approximate-computing
[AAAI-2023] "Reinforced Approximate Exploratory Data Analysis". In: 37th AAAI Conference on Artificial Intelligence (AAAI) 2022 (Acceptance rate: 19%) #ML-for-systems, #Approximate-computing
[Neurips-2022] "Root Cause Analysis of Failures in Microservices through Causal Discovery" #ML-for-systems
[SIGMOD-2022] (Demo) "Efficient Insights Discovery through Conditional Generative Model based Query Approximation". In: SIGMOD 2022 Demo Track #ML-for-systems, #Approximate-computing
[ACL-2022] "Few-Shot Class-Incremental Learning for Named Entity Recognition". #ML
[AAAI-2022] "Conditional Generative Model based Predicate-Aware Query Approximation". In: 36th AAAI Conference on Artificial Intelligence (AAAI) 2022 (Acceptance rate: 15%) #ML-for-systems, #Approximate-computing
[UCC 2021] "Scheduling ML Training on Unreliable Spot Instances". UCC 2021: 14th IEEE/ACM International Conference on Utility and Cloud Computing In: 14th IEEE/ACM International Conference on Utility and Cloud Computing (DML-ICC Workshop), 2021 #Cloud-computing
[ACM SenSys 2021 + ACM TOSN] "ApproxNet: Content and Contention-Aware Video Object Classification System for Embedded Clients" In: ACM Transactions on Sensor Networks (TOSN), pp. 1-27, 2021 #Approximate-computing
[USENIX ATC 2021] "SONIC: Application-aware data passing for chained serverless applications" In: USENIX Annual Technical Conference (ATC), 2021 #Cloud-computing
[AAAI 2021] "Scheduling of Time-Varying Workloads using Reinforcement Learning" In: 35th AAAI Conference on Artificial Intelligence (AAAI) 2021 (Acceptance rate: 21%) #ML-for-systems
[WSDM 2021] "Data-Sharing Economy: Value-Addition from Data meets Privacy" (Demo Paper) In: Proceedings of 14th ACM Conference on Web Search and Data Mining (WSDM) 2021
[ACM SenSys 2020] "ApproxDet: content and contention-aware approximate object detection for mobiles" In: Proceedings of the 18th ACM Conference on Embedded Networked Sensor Systems (SenSys) 2020 (Acceptance rate: 20.7%) #Approximate-computing
[USENIX ATC 2020] "OPTIMUSCLOUD: Heterogeneous Configuration Optimization for Distributed Databases in the Cloud" In: USENIX Annual Technical Conference (ATC), 2020 (Acceptance rate: 18.7%) #Cloud-computing
[ACM/SIGOPS APSys 2019] "DeepPlace: Learning to Place Applications in Multi-Tenant Clusters" In: Proceedings of the 10th ACM SIGOPS Asia-Pacific Workshop on Systems (APSys), 2019. [Paper] #ML-for-systems
[USENIX HotEdge 2019] "Edge-based Transcoding for Adaptive Live Video Streaming" In USENIX Workshop on Hot Topics in Edge Computing (HotEdge), 2019. [Paper]
[ USENIX ATC 2019 ] "SOPHIA: Online Reconfiguration of Clustered NoSQL Databases for Time-Varying Workload" In: USENIX Annual Technical Conference (ATC), 2019 (Acceptance rate: 19.9%) . [Paper]
[ Middleware 2018 ] "Pythia: Improving Datacenter Utilization via Precise Contention Prediction for Multiple Co-located Workloads" In: Middleware 18: The 2018 ACM/IFIP/USENIX International Middleware Conference (Middleware), pp. 1-14, Dec. 10-14, 2018, Rennes, France. ( Acceptance rate: 23.2% ) . [Paper] #Cloud-computing
[ USENIX ATC 2018 ] "VideoChef: Efficient Approximation for Streaming Video Processing Pipelines" In: USENIX Annual Technical Conference (ATC), 2018 ( Acceptance rate: 20.1% ). [Paper] #Approximate-computing
[ Middleware 2017 ] "Rafiki: a middleware for parameter tuning of NoSQL datastores for dynamic metagenomics workloads" In: Proceedings of the 18th ACM/IFIP/USENIX Middleware Conference (Middleware), December 2017 ( Acceptance rate: 23.5% ) #Cloud-computing
[ CGO 2017 ] "Phase-Aware Optimization in Approximate Computing" (with M. K.Gupta, S. Misailovic, S. Bagchi) In: International Symposium on Code Generation and Optimization (CGO), February 2017 ( Acceptance rate: 22.8% ) #Approximate-computing
[ EuroSys 2016 ] "Partial-parallel-repair (PPR): a distributed technique for repairing erasure coded storage" (with AT&T Research) [ pdf ] In: European Conference on Computer Systems (Eurosys), April 2016, (Acceptance rate: 21%) [ Acceptance rate: 21.1% ]
[ SRDS 2016 ] "Sirius: Neural network based probabilistic assertions for detecting silent data corruption in parallel programs" (with T. Thomas, A. J. Bhattad, S. Bagchi) In: International Symposium on Reliable Distributed Systems (SRDS), October 2016 ( Acceptance rate: 32% )
[ ISSRE 2016 ] “A Study of Failures in Community Clusters: The Case of Conte” (with S. Javagal, A. K. Maji, T. Gamblin, A. Moody, S. Harrell, S. Bagchi) In: International Symposium on Software Reliability Engineering (ISSRE), October 2016
[ DSN 2016 (Fast Abstract) ] "Cluster Workload Analytics Revisited" (with Purdue Research Computing and LLNL) In: International Conference on Dependable Systems and Networks (DSN), June 2016
[ PACT 2015 ] “Dealing with the Unknown: Resilience to Prediction Errors” (with G. Bronevetsky, S. Javagal, S. Bagchi) [ pdf ] In: International Conference on Parallel Architectures and Compilation Techniques (PACT), October 2015, (Acceptance rate: 21.2%)
[ ICAC 2015 ] “ICE: An Integrated Configuration Engine for Interference Mitigation in Cloud Services” (with A. Maji, S. Bagchi) [ pdf ] In:International Conference on Autonomic Computing (ICAC), July 2015, (Acceptance rate: 20.3%) #Cloud-computing
[ WCNC 2015 ] “VIDalizer: An Energy Efficient Video Streamer” (with A. Raha, V. Raghunathan, S. Rao) [ pdf ] In: IEEE Wireless Communications and Networking Conference (WCNC), March 2015
[ Middleware 2014 ] "Mitigating Interference in Cloud Services by Middleware Reconfiguration" (with A. Maji, B. Zhou, S. Bagchi, A. Verma) [ pdf ] In:International Middleware Conference (Middleware), December 2014, (Acceptance rate: 18.8%) #Cloud-computing
[ PLDI 2014 ] "Accurate application progress analysis for large-scale parallel debugging" (with I. Laguna, D. H. Ahn, S. Bagchi, M. Schulz, T. Gamblin) [ pdf ] In: Programming Language Design and Implementation (PLDI), June 2014, (Acceptance rate: 18.1%)
[ SC 2013 ] "Scalable Parallel Debugging via Loop-‐Aware Progress Dependence Analysis" (with I. Laguna, D. H. Ahn, M. Schulz, T. Gamblin, S. Bagchi) [ pdf ] In:Supercomputing Conference (SC), November 2013
[ SRDS 2013 ] "Automatic Problem Localization via Multi-dimensional Metric Profiling" (with I. Laguna, F. Arshad, N. Theera-Ampornpunt, Z. Zhu, S. Bagchi, S. P. Midkiff, M. Kistler, A. Gheith) [ pdf ] In: International Symposium on Reliable Distributed Systems (SRDS), October 2013, (Acceptance rate: 32.8%)
News Features:
[2018] We successfully deployed a scalable, diverse and fair recommendation engine for Behance (an Adobe social network product for creators) that hosts several millions of creative projects and serves several millions creative professionals and users. [Links to NEWS coverages: link1, link2, link3]
[2016] Our research (Eurosys'16) on distributed storage with AT&T Research received VURI award from AT&T and is in Purdue spotlight [ Link ]
[2015] Purdue news features our research on cluster workload analytics! [ Link ]
[2014] Our work is highlighted by LLNL Science & Technology Review magazine with the title “Supercomputing Tools Speed Simulations”. [ Link ]
Patents (Not updated):
Parallel partial repair of storage
Integrated configuration engine for interference mitigation in cloud computing
Self-learning Scheduler for Application orchestration on shared compute cluster
Tenant-Side Detection, Classification, and Mitigation of Noisy-Neighbor-Induced Performance Degradation
Cooperative Platform for Generating, Securing, and Verifying Device- Graphs and Contributions to Device Graphs
I am really fortunate to get to closely mentor these awesome interns, graduate students and research associates. (Not regularly updated)
Alind Khare (Research Intern) => Senior Researcher, Microsoft, Bangalore
Shweta Pande (Research Intern) -- IISc
Sarthak Chakraborty (Research Associate at Adobe ) => PhD student at UIUC
Raunak Shah (Research Associate at Adobe ) => Graduate student at UIUC
Nikhil Sheoran (Research Associate at Adobe ) => Graduate student at UIUC => Databricks
Aashaka Shah (undergrad at IIT-Roorkee) => Adobe Research Intern => PhD student at UT Austin => Senior Researcher, Microsoft Research, Redmond
Shanka Subhra Mondal (undergrad at IIT-Kharagpur) => Adobe Research Intern => PhD student at Princeton
Pradeep Dogga (undergrad at IIT-Kharagpur) => Adobe Research Intern => PhD student at UCLA
Ran Xu (PhD student at Purdue University, advised by Prof. Saurabh Bagchi) => Adobe Research Intern => Senior Deep Learning Software Enginner, NVIDIA
Ashraf Mahgoub (PhD student at Purdue University, advised by Prof. Saurabh Bagchi and Prof. Somali Chaterji)
Piyush Bagad (undergrad at IIT-Kanpur) => Adobe Research Intern => Wadhwani AI => DPhil Student, University of Oxford
Sheng Yang (PhD student at University of Maryland, College Park, advised by Prof. Samir Khuller) => Adobe Research Intern
Ayush Chauhan (undergrad at IIT-Roorkee) => Adobe Research Intern => Research Associate, Adobe Research => Microsoft