Hi, my name is Tejasvi Sharma
I'm an aspiring data scientist .

Know more

About me

Profile Image

I am Tejasvi Sharma , a grad student at IISc Bangalore pursuing Masters in Artificial Intelligence. I am interested in solving various data science problems. I am currently working on the problem of Machine Learning on the Edge(Federated Learning) and its application in the field of Mobile Keyword prediction. I have done a summer applied science internship at Amazon where I worked with Prime Video Team on unsupervised Quality estimation of subtitles.

Prior to joining IISc I have 1 year work experience as a database developer at a startup Alert Enterprise. My work revolved assisting various teams in design and optimization of SQL queries and other server related tasks like scheduled maintenance and archival and purging. Prior to that I have done my Bachelors in Computer Sciences from Panjab University, Chandigarh.

View Resume

Projects

Federated Learning for Next Word Prediction

Using decentralised learning on edge devices with low resource and private data with aim to learn a more personalized model per user. Major work revolves around design of various optimization algorithms for the non iid dataset and adapting it to the problem of keyword prediction on Mobile keyboards.

Source Code

Recommender System using Collaborative Filtering

The problem of recommender system is considered on the MovieLens dataset with aim to give personalized recommendations to the users based on their similarity with other users. Variational Autoencoders are used to model the problem and Recall Metric is used to evaluate the performance.

Source Code

MultiModal Neural Machine Translation from English to Hindi Lang.

The problem of image captioning is considered with aim to produce a translation in a target language given the caption in source language, the corresponding image. Attention Models are used to make use of the visual features to disambiguate the words.

Source Code

Quality Estimation in Machine Translation

The problem of unsupervised quality estimation of Translation is considered. Given a translation, it has to be evaluated on the quality aspects of semanticity and fluency. The fluency aspect was considered as part of this project and Perplexity scores were used to compare the fluency aspect of the various sentences and classify them as good or bad.

Source Code

SNLI classification

This project considers the problem on inference on the popular SNLI dataset where a pair of sentences are given and we have to classify them as entailment, neutral, negative.

Source Code

Image Classification on FashionMNIST

This project considers the problem of classification of 10 clothing items using deep models like MLP and CNN on the FashionMNIST dataset

Source Code

Contact

Tejasvi Sharma

Call to Action