CV
Manav Nitin Kapadnis
Contact: +91-9322933311
Email: iammanavk@gmail.com
GitHub: https://github.com/manavkapadnis
LinkedIn: https://www.linkedin.com/in/manav-nitin-kapadnis/
Education
2019-24 | B.Tech + M.Tech (Dual Degree) | Indian Institute of Technology Kharagpur | Major - Electrical Engineering |9.18 CGPA
2018-19 | Class 12 | Shubham Raje Junior College, Thane |All India Senior School Certificate Examination | 91.24 %
2016-17 | Class 10 | D.A.V Public School, Thane |Central Board of Secondary Education |10 CGPA
Publications
Leveraging Pre-trained Language Models for Key Point Matching (EMNLP 2021)
International Conference on Empirical Methods in Natural Language Processing (EMNLP) 2021
- Achieved the best MAP score of 0.9096 and 2nd place on the global leaderboard, for the identification and ranking of relevant key points
- Designed and implemented a pre-trained language model-based architecture to predict the match score for key point and argument pairs
- Improved the MAP by adding syntactic features such as POS Tags, TFIDF scores, and data augmentation using external datasets
Compositional Committees of Tiny Networks (ICONIP 2021)
28th International Conference on Neural Information Processing
Advisor - Prof. Tomas Maul
- Designed and implemented a new form of compositional ensembles that speeds up computations in deep neural networks
- Achieved an increase in speed by25.7times in a standard CPU setup as compared to 6-layered CNN without any reduction in accuracy
- SurpassedMLP-Mixer(SOTA) in terms of speed by exploiting the composition of the hidden representations of classifiers
Fine-grained Intent Classification in the Legal Domain (AAAI 2022)
36th AAAI Conference on Artificial Intelligence
- Proposed a new dataset in the Legal Domain for the task of fine-grained as well as coarse-grained intent classification
- Achieved an accuracy score of0.74for the task of document classification by leveraging Pre-trained Language Models such asDeBERTa
- Established a concrete baseline with an accuracy score of0.902for fine-grained intent classification usingJointBERT
Internship and Research Experience
Institute for Data, Systems, and Society
| Massachusetts Institute of Technology (MIT) | Racism and Violent Incidents Detection in Historical Archives | Advisor - Kiran Garimella | (Oct ’21 - Dec ’21) |
- Developed a complete framework to enhance the analytical study of violent incidents throughout the country over a period of 100 years
- Applied BART, a pre-trained Transformer based Language Model as a classifier to classify articles into violent and non-violent incidents
- Achieved the best accuracy score of94.5%, a recall score of 92%and a precision score of88%
- Currently working with a team from ISB Hyderabad and MIT to study the causation of the violent incidents and riots in the country
Financial Computing and Analytics Group
| University College London (UCL) | Governance Pulse: AI for real-time governance | Advisor - Prof. Philip Treleaven | (Dec ’20 - April ’21) |
- Formulated a Governance based analytical project providing real-time monitoring information to the Board members of the company
- Developed a Time - Series model for forecasting stock price movements using Long Short-Term Memory on LSE data
- Implemented a RoBERTa based text-mining tool for the discovery of hidden semantic structures in a text body
- Achieved a ROUGE-1 scoreof0.259for the task of summarizing the contents of financial reports of the company
Complex Networks Research Group
| Indian Institute of Technology Kharagpur | A New Intent-Based Metric for Legal Document Summarization | Advisor - Prof. Niloy Ganguly | (May ’21 - June ’21) |
- Developed a new summarization metric, which takes the help of intent of the document for the evaluation of generated summary
- Achieved better relevance and overall human satisfaction scores as compared to other automated metrics such asBLEUandROUGE-L
- ImplementedJointBERTfor the task of automated intent phrase extraction, as well as document classification(submitted in NAACL’22)
Relevant Skills/Coursework
CS/ML: Probability and Statistics | Algorithms | Programming and Data Structures | Machine Learning | Natural Language Processing
Libraries / Frameworks: Tensorflow | PyTorch | HuggingFace | Keras | Spacy | Numpy | Pandas | Regex | Matplotlib | Seaborn | Plotly
Languages: C | C++ | Python | MATLAB
Positions of Responsibility
Technology Robotix Society, IIT Kharagpur
- Lead a 3-tier team to execute the annual Robotix fest & conducting workshops with over 1000+ participants from across the country
- In charge of Maker Space - an open-source lab for robotics enthusiasts seeking guidance & component
- Mentored 100+ students in the IEEE Winter School of AI & Robotics workshops on Machine Learning Image & Processing
Kharagpur Data Analytics Group, IIT Kharagpur
- Successfully organized the first Kharagpur Data Science Hackathon, featuring 1200+ teams and a total prize pool of INR 50000
- Organized an interactive month-long data Science Workshop, designed for freshmen with over 300 participants from across the country
Extra-Curricular Activities
- Mentoring five freshmen students under the student Welfare Program, under the aegis of the Dean of Student Affairs, IIT Kharagpur
- Selected as an Indian Delegate to represent IIT Kharagpur at Harvard US-India Initiative ’21 amongst more than 2000 applicants