Sudipta Sengupta
Ph.D., Electrical Engg. & Computer Science, MIT, Cambridge, USA.
M.S., Electrical Engg. & Computer Science, MIT, Cambridge, USA.
B.Tech., Computer Science & Engg., IIT-Kanpur, India.
ACM Fellow, IEEE Fellow (news article)
Vice President & Distinguished Scientist, Amazon AWS, Seattle, USA
LinkedIn Profile
Email: domain is gmail.com, username is sudipta.sengupta
Dr. Sudipta Sengupta is Vice President & Distinguished Scientist at Amazon AWS where he is leading new initiatives in Artificial Intelligence/ Machine Learning/ Deep Learning across AWS AI, Databases & Analytics, and EC2/ Annapurna Labs, including hardware (CPU/ GPU/ ASIC - Trainium/ Inferentia/ Elastic Inference), ML frameworks (Tensorflow/ PyTorch) and compilers (Neuron/ Neo), ML services (SageMaker), AI services (Vision/ Languages/ Speech), enterprise semantic search (Kendra), ML capabilities in databases (Aurora ML/ Athena ML/ Redshift ML), natural language interfaces for querying data (QuickSight Q), and making databases autonomous and self-driving (DevOps Guru for RDS).
Sudipta has a 25-year record of sustained contributions to research and development in Computer Science. He brings together a hybrid blend of expertise from industry, research labs, and academia. He has caught early trends in artificial intelligence, computer systems, data management, storage, and networking by starting multiple research projects in these areas that have gone beyond advancing the state-of-the-art in Computer Science. To this end, he has initiated partnerships with engineering groups (in the companies he has worked in) through deep, multi-year engagements and shipped his research in many industry leading, award winning products and services. By successfully formulating and executing this end-to-end approach to research, he has ultimately influenced industry thinking and practice in his areas of work.
Sudipta's work on oblivious routing of network traffic, which makes the network provide predictable guarantees in the face of highly variable and unpredictable traffic, received two major IEEE awards -- the IEEE Leonard G. Abraham Prize and the IEEE William R. Bennett Prize. Variable traffic appears in multiple network settings, including Internet backbone and cloud data centers. Continuing this line of work, Sudipta took on the challenges of networking at scale for the modern cloud data center and designed the network architecture and traffic oblivious routing algorithms for VL2, a new generation data center network that has been deployed in the Microsoft cloud and that introduced foundational ideas that have since been adopted across the cloud computing industry. The VL2 paper received the ACM SIGCOMM Test of Time Award.
Sudipta's research on data deduplication advanced the frontier from backup to primary data, persuaded Microsoft to develop deduplication technology in-house instead of acquiring it from outside, and was incorporated into the new primary data deduplication feature in Windows Server 2012 and I/O deduplication for virtualized storage in Windows Server 2012 R2. Primary data deduplication was rated among the top new features in Windows Server 2012 by customers, analysts, and digerati. This technology provided early thought leadership in the primary storage market where data deduplication is table stakes today, with major storage offerings building upon and extending ideas that Windows Server pioneered.
Sudipta advocated rethinking data storage and management for flash memory, championed the development of the first flash-optimized data store in Bing, and shipped multiple flash based key-value stores and indexing technology in Azure CosmosDB, Bing ObjectStore, and SQL Server Hekaton. Today, engineering groups across the industry see the clear need for flash in their products and services. Sudipta's work has helped develop broad understanding that the software stack needs to be optimized to exploit the benefits of flash and work around its peculiarities.
Previously, Sudipta spent five years at Bell Laboratories, the Research Division of Lucent Technologies, where he worked on Internet routing, optical switching, network security, wireless networks, and network coding. Before that, he was at Tellium, an optical networking pioneer, that grew from an early-stage startup to a public company during his tenure there. At both Lucent and Tellium, Sudipta conceived and led the development of new product features that were critical to customer contract wins. At both these companies, he was also responsible for defining and shaping the company's vision for next-generation Internet backbone architectures.
Sudipta received a Ph.D. and an M.S. in Electrical Engg. & Computer Science from Massachusetts Institute of Technology (MIT), USA, and a B.Tech. in Computer Science & Engg. from Indian Institute of Technology (IIT), Kanpur, India. He was awarded the President of India Gold Medal at IIT-Kanpur for graduating at the top of his class across all disciplines. He has published 100+ research papers in some of the top conferences, journals, and technical magazines. He has authored 100+ patents (granted or pending) in the area of computer systems, networking, storage, and data management. He has taught advanced courses at academic/research and industry conferences. His work has received widespread coverage in media/press and blogs.
Sudipta is ACM Fellow and IEEE Fellow. He has been recognized in the academic/research and industry community with the following awards/prizes/honors:
Sudipta has taught tutorials on data deduplication at USENIX FAST 2013 and on data center networks at ACM SIGCOMM 2013, IEEE Hot Interconnects 2012, IEEE GLOBECOM 2011, IEEE Hot Interconnects 2011, ACM SIGMETRICS 2011, IEEE ICC 2011, and ICCCN 2011, on peer-to-peer systems at ACM SIGMETRICS 2010, on oblivious routing of Internet traffic at IEEE ICC 2009 and ACM SIGMETRICS 2008, and on wireless network coding at ACM MOBIHOC 2008.
Sudipta is serving (or, has served)
The Domain Specialization Opportunity for Deep Learning Across Accelerators, Compilers, and Frameworks
Invited Keynote at Northwest Database Society (NWDS) Annual Meeting, January 2020.
The Heterogeneity Challenge & Opportunity with Deep Learning
Invited Talk at DARPA ISAT Workshop, January 2020.
Domain Specialization for Deep Learning across Accelerators, Frameworks, and Compilers
Invited Talk at UCSD Computer Science and Engineering, February 2020.
Integrating Deep Learning Accelerators with TensorFlow
Talk at the first O'Reilly TensorFlow World conference, October 2019.
The Bw-Tree Key-Value Store: From Research to Production | (video)
Invited Talk at UCSD Computer Science and Engineering, October 2015, and at Northwest Database Society (NWDS), hosted by UW Database Group, January 2016.
The Bw-Tree Key-Value Store and Its Applications to Server/Cloud Data Management in Production
Talk at UC Berkeley AMPLab, Berkeley, CA, USA and at Storage Developer Conference (SDC) 2015, Santa Clara, CA, USA, September 2015.
Evolution of Data Center Networking
Invited Keynote at IEEE LANMAN 2014, Reno, USA, May 2014.
Data Center Networking: What was not Working? What is Working? What needs Work?
Invited Panel Speaker at ICCCN 2013, Nassau, Bahamas, July 2013.
Primary Data Deduplication in Windows Server 2012
Talk at Storage Developer Conference (SDC) 2012, Santa Clara, CA, USA, September 2012.
Primary Data Deduplication: From Research to Windows Server 2012
Talk at Amazon.com, Inc., Seattle, WA, USA , August 2012. (Hosted by James Hamilton)
Smart Pricing: Parallels from the Cloud Computing World
Invited Talk at Smart Data Pricing Forum, Princeton University, Princeton, NJ, USA , July 2012.
App Aware Smart Pricing Enabled Cross-Provider Wireless Network Fabric
Invited Panel Talk at Smart Data Pricing Forum, Princeton University, Princeton, NJ, USA , July 2012.
Speeding Up Cloud/Server Applications Using Flash Memory
Talk at Storage Developer Conference (SDC) 2011, Santa Clara, CA, USA, September 2011.