About
I'm a Principal AI Research Scientist at Autodesk AI Lab, where I focus on developing large foundation models for the 3D world, particularly in the architecture, engineering, and construction (AEC) domain. Previously, I was an Applied Scientist at AWS AI Labs working on Amazon Q Developer (formerly CodeWhisperer) and Bedrock large language modeling services.
I received my Ph.D. in Computer Science from the University of Minnesota, where I worked on deep generative models, approximate inference, and probabilistic models with applications to text and image data.
Research Interests
- Machine learning
- Deep generative models
- Large language models
- Multimodal foundation models
- Representation learning
Selected Publications
Multi-task pretraining with structured knowledge for text-to-SQL generation
R. Giaquinto, et al. | ACL 2023
We generate SQL code from natural language by using and encoder-decoder trained on a large pretraining dataset with multitask learning strategy. Our method leads to state-of-the-art performance with significant improvements on the Spider and CoSQL benchmarks.
Multi-lingual evaluation of code generation models
B. Athiwaratkun, et al. | ICLR 2023
We introduce MBXP, Multilingual HumanEval, and MathQA-X: new benchmarks for evaluating code generation models across over 10 programming languages. Our study reveals surprising capabilities in language models, including out-of-domain language generalization and zero-shot translation abilities, while demonstrating the advantages of multi-lingual models over mono-lingual approaches.
Gradient boosted normalizing flows
R. Giaquinto and A. Banerjee | NeurIPS 2020
We introduce Gradient Boosted Normalizing Flows (GBNF) to improve normalizing flows through gradient boosting rather than deeper architectures. GBNFs create a mixture model that grows in flexibility as components are added, offering better performance than traditional flows while maintaining simpler transformations.
Contact
Feel free to reach out if you want to discuss a research collaboration or are a PhD student interested in an internship.
Email: my last name dot ra at gmail.com