Research Interests

My research interests lie in the intersection of machine learning with natural language processing or computer vision. More recently, I became interested in causal NLP, where I am looking to tackle the following challenges:

causal representation learning
reasoning and planning for goal-based conversational agents
automated scientific discovery
AI for education
bias and fairness
theoretical understanding of neural networks

I believe it has promise in achieving interpretability, explainability, and generalization in modern deep learning methods.

Publications

Assertion Detection in Multi-Label Clinical Text using Scope Localization
Rajeev Bhatt Ambati, Ahmed Ada Hanifi, Ramya Vunikili, Puneet Sharma, and Oladimeji Farri, 2020
[arXiv]
Read, Highlight, and Summarize: A Hierarchical Neural Semantic Encoder-based approach
Rajeev Bhatt Ambati, Saptarashmi Bandyopadhyay, Prasenjit Mitra, 2019
[arXiv] [code]
Segmentation of Low-Level Temporal Plume Patterns from IR Video
Rajeev Bhatt, M. Gokhan Uzunbas, Thai Hoang, Ozge C. Whiting
Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2019
[pdf]

Patents

Method and system and apparatus for quantifying uncertainty for medical image assessment
Florin-cristian Ghesu, Awais Mansoor, Sasa Grbic, Ramya Vunikili, Sanjeev Kumar Karn, Rajeev Bhatt Ambati, Oladimeji Farri, Bogdan Georgescu, Dorin Comaniciu, 2023
[pdf]
Assertion detection in multi-labelled clinical text using scope localization
Rajeev Bhatt Ambati, Oladimeji Farri, Ramya Vunikili, 2023
[pdf]
Segmentation and prediction of low-level temporal plume patterns
Ozge Can Whiting, Rajeev Bhatt, Mustafa Gokhan Uzunbas, Thai Hoang, Vladimir Shapiro, John Passarelli, Weiwei Qian, John Hare, Taufiq Dhanani, Matthias Odisio, Edvardas Kairiukstis, 2022
[pdf]

Projects

Neural Tangent Kernel

Reproduced the results from the NTK paper, along with derivations.
Derived and plotted training dynamics for Momentum optimizer.

[pdf]

LLMs for Causal Inference

Used LLMs to solve problems in causal inference.
Added downstream verifiers to provide feedback for improvement.

[pdf]

Image Quality Assessment

Image Quality Assessment (IQA) problem is about assigning a quality score to an image.
IQA models can act as a differentiable proxy human to improve performance on several tasks such as Super Resolution.
Developed an unsupervised IQA method using autoencoders.

[pdf][code]

Human Motion Prediction

The task is to identify motion from videos such as in the Human 3.6M dataset.
Developed an inception layer to extract the motion parameters of an object.
The extracted parameters are further updated using a spatio-temporal neural network.

[pdf][code]

Image Compression

The task is to decrease the storage size of an image while maintaining similar quality
Used a convolutional autoencoder to achieve compression
Added an adversarial loss to further enhance the image quality

[pdf][code]

Natural Language Generation

WebNLG 2017 Challenge: The task is to generate text from RDF data
Used Delexicalisation, Aggregation, and Constituency Parsed Trees to improve the performance of seq2seq models.

[slides]

Page updated

Google Sites

Report abuse