About
I am interested in Machine Learning, especially in the field of Deep Learning. I have experience in Graph Machine Learning, Computer Vision (CV), and Natural Language Processing (NLP).
• During my undergrad days, I published a long paper at AAAI (top-tier ML conference) that has garnered over 100 citations
• Open source contributor to famous projects like Pytorch Geometric and TorchVision (Pytorch)
• Recipient of various prestigious academic fellowships as well as Microsoft AI for Earth grant
• Been part of teams that have won multiple national and international competitions
I completed my Bachelors from Indian Institute of Technology (IIT) Guwahati. Download Resume.
![image not loaded](assets/resume/publications/images/asap.png)
![image not loaded](assets/resume/publications/images/ae-cnn.png)
![image not loaded](assets/resume/publications/images/auto-scma.png)
![image not loaded](assets/resume/workexs/images/msft.png)
• Working on Knowledge Graph Search and Recommendation using Graph Machine Learning.
• Improved the relevance by 2x.
• Designed and deployed custom Message Passing Algorithms in low latency scenarios.
• Engineered dataset creation without explicit labeled data in an eyes-off and compliant setting.
• Developing the core relevance infrastructure for distributed supervised graph learning.
• Migrated Spark pipelines to Azure Machine Learning. Added CI/CD pipelines along with tests.
• Gave internal talks and wrote blogs for broader audience outside my team.
PyTorch Geometric (PyG) is a geometric deep learning extension library for PyTorch. It is a popular open source library for implementing Graph Neural Networks.
Following are some of my notable contributions to this library:-
• Added Dense Graph Convolution layer #445
• Added ASAP pooling and LEConv layers #1218
• Added Self-Attention Graph pooling #364
• Added Edge Weighted GraphConv #489
Contributors list: link
![image not loaded](assets/resume/workexs/images/pytorch-full.png)
TorchVision is the computer vision library maintained by the Pytorch team at Facebook.
Some of my notable contributions to this library are:-
• Added GPU-Memory Efficient Densenet #1003, #797
• Added Affine Transformation #793
• Refactored Image Segmentation, Detection and Classification models #1009, #1091, #889
Contributors list: link
![image not loaded](assets/resume/workexs/images/iisc.png)
Advisor: Dr. Partha Talukdar
• Proposed a hierarchical sparse pooling operator for Graph Neural Nets. It has better edge connectivity in the pooled graph compared to other sparse methods.
• Introduced a new self-attention framework that is better suited for pooling.
• Proposed a new graph convolution operator that can adaptively learn functions of local extremas in a graph substructure.
![image not loaded](assets/resume/workexs/images/IITKGP.png)
Advisor: Dr. Debdoot Sheet
• Worked on classification of thoracic diseases on NIH ChestX-Ray14 dataset.
• Proposed a model to reduce the loss of discriminative features which occurs during downsampling large images before feeding to classifiers.
• Achieved state of the art results on the official test set released by NIH.
- Sharechat@IndoML' 21: 2nd Rank on Multilingual Abusive comment identification challenge organized by Sharechat
- IndoRE@IndoML' 21: 2nd Rank on a relation extraction task for three low resource Indian Languages (Bengali, Telugu and Hindi)
- Sharechat@IEEE BigMM' 21: 4th Rank on Indic Multilingual Abusive comment identification challenge organized by Sharechat
- Shastri Research Student Fellowship 2019: Among 18 people from Canada and India to be offered the fellowship by Ministry of Human Resource and Developement (MHRD), Government of India.
- Charpak Lab Scholarship 2019: Among 30 people in India to be sponsored by Indian French Embassy
- Tsinghua Deep Learning Summer School 2019: Among top 40 students globally to be invited for the event
- Indian Academy of Sciences Fellowship 2018: Received the prestigious Indian Academy of Sciences Fellowship
- Joint Entrance Examination 2016: Among top 0.1% out of 1.2 million students
- Microsoft AI For Earth: Entitled $5000 Azure grant and inclusion to Microsoft AI for Earth community
- Microsoft code.fun.do++ 2018: Secured 2nd place at National Finals
- Microsoft Big Idea 2018: Among top 10 teams internationally and among top 2 teams in India
- American Express Analyze This 2018: Top 0.8% among 2700+ participating teams
- Microsoft Imagine Cup 2018: Finalist of Imagine Cup 2018 Nationals
- Microsoft code.fun.do 2017: Secured 1st postion in 'Junior Division' and 2nd in 'Senior Division'
- Inter-IIT Techmeet 2017: First in public leaderboard of Data science competition among all IITs
- Goldman Sachs GS Quantify 2017: Ranked 3rd in campus and 32nd in India at Machine Learning section
- American Express Analyze This 2017: Top 1.9% in 2600+ teams. Declared 'Outstanding Performer' by Amex