Bishwaranjan Bhattacharjee
Senior Technical Staff Member – IBM T.J. Watson Research Center
Invited Talk: System Requirements for Deep Learning Foundational Models
Slides
Video
Abstract:
Foundational Models are creating a deep impact on many applications areas of Deep Learning. This includes language, security, biomedical, code etc. Training these models is very computation intensive and can consume even weeks on a large distributed system. Besides training, inference with these models can also be a challenge due to their high latency. In this talk I will describe the challenges posed in training and inferring with large foundational models with special emphasis on language models and what an efficient system would look like.
Biography:
Bhatta Bhattacharjee is a Senior Technical Staff Member (STSM) and IBM Master Inventor working at IBM T.J. Watson Research Center at Yorktown Heights, New York. His interests are in new research directions in deep learning, data management and its applications. In particular Bhatta is interested in scalable deep learning and database processing, data cleansing, compression, integration and exploitation of new hardware for both deep learning and database processing, clustering and indexing techniques, query processing and optimizations, access control and privacy protection.
Prior to joining IBM Research, he was associated with the Database Technology Group (DBT) at IBM Toronto Labs, Canada and the Advanced Numerical Research and Analysis Group (ANURAG) of the Ministry of Defence, Government of India, India.