Amazon and IIT Bombay announce inaugural award recipients
Amazon IIT–Bombay AI-ML Initiative seeks to advance artificial intelligence and machine learning research within speech, language, and multimodal-AI domains.
Amazon and IIT Bombay (IIT-B) today announced the inaugural award recipients of the Amazon IIT–Bombay AI-ML Initiative. The awards recognize researchers whose work fulfills the goals of the initiative: to advance artificial intelligence and machine learning research within the speech, language, and multimodal-AI domains.
The Amazon-funded collaboration, launched in March 2023 and housed in the IIT Bombay Department of Computer Science and Engineering, aims to promote partnership among faculty and leading scholars and foster a diverse and sustainable pipeline of research talent.
In line with the goals of the initiative, the award winners are conducting research in multiple domains including large language models (LLMs), machine learning, federated learning, and natural-language processing (NLP).
"We are pleased to witness the unfolding of this relationship and the significant potential it embodies,” said Sachin Patwardhan, dean of research and development at IIT Bombay. “The selection of nine project proposals under the Amazon IIT-B AI-ML Initiative further accentuates our shared commitment to advancing knowledge and innovation."
The research award provides selected professors at IIT Bombay with up to one full year of funding to pursue independent research projects. The nine research projects selected will be run by IIT Bombay faculty and researchers.
“We are excited about the progress we have made with our partnership with IIT-B,” said Snehal Meshram, senior manager of product management technology with Alexa. “Through the diverse project proposals, we have ensured investments in key research areas that will mutually benefit Amazon and IIT-B. This is just the beginning of what we anticipate to be a long and fruitful collaboration.”
The winners of the awards are as follows:
Bhattacharyya and Jyothi’s research revolves around speech-to-speech machine translation with a focus on Indian languages. They intend to develop a multilingual and multimodal pretrained model for Indian languages and specific linguistic phenomena seen in India (such as disfluency and code mixing). The goal is a model that demonstrates effectiveness in a speech-to-speech translation system, while the project will also release curated training data specific to the Indian use case.
De and Chakrabarti’s research focuses on multimodal representation, retrieval, and transformation using graph-structured objects. The proposal aims to develop methods for information retrieval on a large corpus of graphs, which should allow for multimodal retrieval (text and images).
Ghosh and Nath’s proposal aims to leverage game-theoretic constructs to incentivize contributions from users in federated learning, where a large number of users submit locally computed results that are merged in a centralized server.
Preethi Jyothi, associate professor of computer science and engineering
Jyothi’s proposal aims to build techniques that will make cross-lingual transfer more efficient and effective, using both automatic-speech-recognition (ASR) and NLP pretrained models.
Ankur A. Kulkarni, Kelkar Family Chair associate professor of systems control and engineering
Kulkarni’s proposal aims to develop game-theoretic mechanisms for strategic classification, such as when inputs to a machine learning model have been purposefully altered during testing time.
Ajit Rajwade, associate professor of computer science and engineering
Rajwade’s research will examine nearest-neighbor search in large datasets via group testing. By combining multiple samples in a particular way, the research will determine whether fewer tests are needed.
Ramakrishnan and Jadhav’s research will apply LLMs in the healthcare domain and explore both explainability and verifiability. Their work will incorporate domain-specific training, retrieval augmentation, and alignment to feedback from healthcare experts.
Sunita Sarawagi, associate professor, Center for Machine Intelligence and Data Science
Sarawigi’s research will explore the integration of LLMs with structured databases. Her proposal aims to investigate a series of problems, including converting text to SQL, the continuous improvement of LLMs, and relating trends in structured data to real-world events, and to provide these capabilities in custom LLMs that are developed over private data.
Virendra Singh, professor of computer science and electrical engineering
Singh’s research will use natural language to align reinforcement learning (RL) agents toward human-like behavior. His proposal aims to develop techniques to make RL more sample efficient, using auxiliary NLP tasks such as providing textual descriptions of the actions taken by RL agents.