Ashutosh Baheti

Undergraduate: Bachelor of Technology (Honors)
Department of Computer Science and Engineering
IIT Kharagpur


About Me

I am currently a Research Fellow at Microsoft Research India (MSR) working under the guidance of Dr. Monojit Choudhury and Dr. Kalika Bali. I am a programming and robotics enthusiast. My primary research interest lies in Natural Language Processing (NLP). At MSR I am working on computational models for code-switching and code-mixing. Code-switching or Code-mixing is a phenomenon where two or more languages are mixed in a single document, sentence, phrase or word.

I completed my undergraduate from Indian Institute of Technology, Kharagpur (IIT Kharagpur), one of the premier engineering intitute of India. I completed my Bachelor of Technology (Honors) Degree from the Department of Computer Science and Engineering with a CGPA of 8.91/10.

I was introduced to hobby robotics during my freshman year at IIT Kharagpur and found it very interesting. To futher explore robotics I joined Techonology Robotix Society (TRS). TRS organizes national scale robotics events every year. It also conducts robotics tutorial sessions and workshops. As a part of the team I have contributed various robotics tutorials which can be found tutorials and prezi. I have also worked in the embedded team of Kharagpur RoboSoccer Student's Group (KRSSG) which develops autonomous soccer playing robots. Apart from research and technical stuff I am also interested in community work. I have succesfully planted 21 trees (8 mango, 8 guava and 5 neem) in my hostel [image]. I, along with 3 other students of IIT Kharapur, have initiated a model school project. We plan to establish a model school in a village of Khagaria District of Bihar. We have already opened a Non-profit trust "Parivartan" (Transformation) and acquired 5 acre land. To overcome the dearth of qualified teachers in the villages we plan to employ e-learning model for the course curriculum. I wish to utilize my research experties to create a NLP algorithm which can take course material as input and give summarized presentations as output. This will give us huge amount of teaching content very quickly.
I have also worked in DigitalGreen as an intern. My technical experties was put in good use here. The details of my project here is described in the project section.


Barrier Coverage in Mobile Wireless Sensor Networks

Wireless sensor networks are used a lot these days for monitoring, exploration and surveillance activities across the world. Different applications require different coverage schemes. Among all the various coverage schemes, Barrier Coverage is gaining a lot of popularity these days. It is mainly used in intrusion detection in a belt region. In this project we assume sensors to be mobile and dynamically construct 1-barrier coverage after the random deployment of sensors.

Current work, "Non-linear Barrier Coverage using Mobile Wireless Sensors" is submitted to IEEE ICC 2017 Ad-Hoc and Sensor Networking Symposium in which I am the first author. The code will be released after submission

Personality Analysis of Quora Users

Problem: To extract BIG-5 personality traits of Quora users based on their answers.
I led a team of 5 students. Our work in the project is summarized below,

  • Identified linguistic patterns for personality traits from literature survery.
  • Implemented feature extractors for the identified patterns.
  • We used LIWC feature counts, a technique typically used for personality analysis. These LIWC word buckets were augmented using Word2Vec similarity on Quora corpus.
  • Finally for traning and testing data, we manually conducted BFI-10 personality survey of Quora users and then crawling their answers while retaining anonymity.
Best results achieved using Adaboost-Random Forest classifier which gave 75%-80% accuracy for all 5 personality traits on the curated features. It is desirable to test the method on larger annotated dataset.

PhysIO - A physiotherapist app

Aim: To design an android application targeted for wearable devices which can alone (without using cloud resources) assist and monitor physiotherapy exercises of patients.
Our team of 4 solved the problem as follows

  • Extracted IMU readings of an exercise run and created a feature vector.
  • Implemented automatic stop detection for hands-free user experience.
  • Trained a SVM based classifier which categorized new exercise run into predefined exercises.
  • Post classificaiton we filtered incorrect runs using dynamic time warping similarity

Distributed Chat and Conferencing

Developed a distributed chat and conferencing system in which messages are sent directly to the user without involvement of the central server unlike the conventional central server systems. The chat messages were ordered using Birman-Schiper-Stephenson protocol. The system could support dynamic addition and deletion of users from a group chat. The software was java based with clean gui and support of group chats

Stick Runner

A cool android game. Developed using unity3D game engine.

I invite everyone to give there suggestions and feedback for this site.