Ramakrishna Vedantam |
Hi! I am currently a visiting researcher at the NYU Center for Data Science (CDS) working with Prof. Julia Kempe and Prof. Andrew Gordon Wilson. Previously, I was a Research Scientist at Facebook AI Research (FAIR) in New York. I enjoy working on problems in core machine learning that make AI models more general, robust and human-like. In pursuit of these goals, my current focus is on multimodal machine learning, robustness and data efficient learning.
At FAIR, I proposed the Decodable Information Bottleneck for Representation Learning which provides a theoretical and practical foundation for worst case risk mitigation for deep representation learning.
Previously, I graduated with a Ph.D. in Computer Science from the School of Interactive Computing at Georgia Tech. My PhD advisor was Devi Parikh. My thesis was on āInterpretation, Grounding and Imagination for Machine Intelligenceā. Among other things, during my Ph.D. I proposed the CIDEr metric commonly used for evaluating image captioning models (3000+ citations) and contributed to the Grad-CAM method popularly used for interpreting neural network predictions (13000+ citations).
For more details on my research see my Google Scholar page or checkout my publications. My PhD research was supported by the highly competitive Google Ph.D. fellowship in Machine Perception, Speech Technology and Computer Vision.
I have been fortunate to work with some great mentors and collaborators during grad school, including Larry Zitnick, Dhruv Batra, Kevin Murphy, Gal Chechik, and Samy Bengio.
In a previous life, I was an undergrad in ECE at IIIT-Hyderabad where I worked with K. Madhava Krishna in Robotics. Here is a link to my old website.
News
- [February, 2023] Paper on Nullspace characterization for Out of Distribution (OOD) generalization accepted at ICLR, 2023!
- [February, 2023] Paper on Robustness of VQA models to distribution shifts accepted at CVPR, 2023.
- [May, 2022] Paper on Measuring Equivariance of Objectness models accepted at ICML, 2022.
Code
- MSCOCO Caption Evaluation code
- Codes from MSCOCO Caption Evaluation for metrics (BLEU, ROUGE, CIDEr-D and METEOR), independent of the COCO annotations
- Code for our CVPR'16 paper on Learning Visually Grounded Word Embeddings
- Code for our ICLR'18 paper on Generative Models of Visually Grounded Imagination
If you like this layout/page, see this to build your own using github+jekyll