profile image

MANSI KHAMKAR

International Institute of Information Technology, Hyderabad
M.Tech in CSE

Download Resume

About Me

I am a technology enthusiast having several years of experience in software development. Currently, I am more inclined towards the field of Artificial Intelligence and wish to work in this field in the future. I started learning AI in the final year of B.E. and involved myself in AI projects. Recently, I worked as a Research Assistant on a project based on Computer Vision at CVIT Lab and have co-authored two successful publications.

The field of AI has always inspired me to brainstorm ideas for the automation of laborious tasks. I am intrigued about this field, as it makes me ponder why existing machine learning models behave the way they do and how can their inner working mechanisms be made more interpretable. My passion for diving deep into concepts and involving myself in AI-related projects aimed at solving real-world problems has driven me to opt for research in this field. Becoming an active researcher, I wish to work on improving deep learning models and come up with useful applications.

Publications

Wisdom of (Binned) Crowds: A Bayesian Stratification Paradigm for Crowd Counting

Sravya Vardhani Shivapuja, Mansi Pradeep Khamkar, Divij Bajaj, Ganesh Ramakrishnan*, Ravi Kiran Sarvadevabhatla*; In proceedings of the 2021 ACM Conference on Multimedia [* - Mentor]

  • Highlighted the issues with training and evaluation of crowd counting deep networks till date, which report only MAE but ignore the standard deviation which is large even in the case of SOTA networks.
  • Proposed a modified crowd counting pipeline that involves partitioning the count range into bins optimal for uniform sampling and minibatch creation, by employing Bayesian stratification approach to alleviate the effect of heavy tailed dataset distribution.
  • Proposed a novel bin sensitive loss function to incorporate decrease of variance at each training step and suggested a new evaluation metric.
View Paper View Website

Stutter Diagnosis and Therapy System based on Speech Processing and Deep Learning

Gresha Bhatia*, Binoy Saha, Mansi Khamkar, Ashish Chandwani, Reshma Khot; In 13th INDIACom-2019, 6th International Conference on Computing for Sustainable Global Development [* - Mentor]

  • Attempted to detect and classify stutters in the input audio while existing works focused only on detection of stutters. Trained a Gated Recurrent CNN on MFCC audio features for stutter detection and classification.
  • Proposed an SVM-based system that can suggest therapies based on the type and severity of the stuttering.
  • Enhanced the dataset by manually extracting and segregating audio chunks into various stutter categories along with stutter severity.
View Paper

Key Projects

Machine Learning

project name
Find out more

Wisdom of (Binned) Crowds : Crowd Counting

Bayesian Stratification Deep CNNs Error Variance Reduction PyTorch

Worked under the guidance of Prof. Ravi Kiran Sarvadevabhatla and Prof. Ganesh Ramakrishnan on devising and evaluating approaches for crowd counting in images. Implemented the proposed variance reduction crowd counting pipeline along with the novel bin-sensitive loss function and new data augmentation techniques; and conducted experiments on the SOTA counting networks.

project name
Find out more

Stutter Diagnosis & Therapy System

Speech Processing CNN GRU SVM Keras Android Firebase

Developed an Android application for stutterers under the guidance of a speech therapist, that diagnoses the type and severity of stutter and recommends speech therapies. The app exploits CNN followed by GRUs to detect disfluencies in recorded speech and uses SVM to suggest the best therapies based on the user’s monitored performance.

project name
Find out more

ChefGAN : Meal Image Generation from Ingredients

Multi-modal Learning StackGAN-v2 BiLSTM ResNet Pytorch

Trained an attention based cross-modal network consisting of Bidirectional LSTM and ResNet to match an ingredient list and its corresponding image in a joint latent space using triplet loss. Trained a modified StackGAN-v2 conditioned on encoded ingredients for generating meal images from noise by using cycle consistency constraint

Information Extraction and Retrieval

project name
Find out more

Wikipedia Search Engine

Parsing Stemming Indexing Searching Ranking

Built a scalable and efficient search engine on large Wikipedia dump of size ~80GB. Parsed the XML dump, processed words and created inverted index by recording word counts in each page. Implemented multi-field query search and page ranking mechanism using Tf-Idf metric. Optimized the system for index creation time, index size and query search time.

project name
Find out more

Social Champion Identification

Data Search Ranking Facebook Graph API Twitter REST API

Developed a web application for the NGOs to obtain a list of prospective donors and influencers on social networks who can help in promoting NGOs. Implemented a program to retrieve Facebook and Twitter feeds regarding social awareness by using the respective APIs to analyse the users associated with them, considering various aspects, and ranked users based on their calculated prospectiveness score.

System Applications

project name
Find out more

Peer-to-Peer Group Based Torrent

File Tracker Piece-selection Seeders/Leechers Socket Programming C++

Built a terminal based fault-tolerant file sharing system where users can login, join groups and share or download files. Divided the files into logical pieces and created a tracker to maintain which user has which part of the file. Applied piece selection algorithm to download chunks parallelly from multiple seeders and leechers.

project name
Find out more

Linux Terminal File Explorer

System Calls Interrupt Handlers C++ Shell

Built a terminal based file explorer for Linux that allows user to navigate through the entire file system using special keys to scroll, select or open files and move forward or backward in the directory structure. Displayed file properties and provided a command mode interface using system calls where user can type and execute basic file commands.

Web Applications

project name
Find out more

Virtual Lab for Requirements Engineering

Interactive Learning React-Js GoJs

Developed a React-JS web application to create an interactive environment for the user to learn Requirements Engineering in a practical way by simulating various game-like activities. Created a canvas using GoJS to let users draw use-case diagrams for the given problem statements, provided relevant hints whenever required and evaluated the users based on their performance in all activities.

project name
Find out more

Restaurant Management Website

CMS Payment Gateway SEO Lazy Loading PHP / AJAX

Developed a full-fledged responsive website for online food ordering having content management system and kitchen inventory management system. Designed database schema, implemented App shell architecture and lazy loading for performance optimization and performed on-site Search Engine Optimization.

project name
Find out more

What's Happenning in My City

Android Firebase Geolocation API Real-time Push Notifications

Developed an Android application that connects local businesses to the citizens or visitors and sends realtime notifications regarding all the startups, sales, offers, exhibitions and other events happening in their city. Used Firebase to store data and send real-time notifications.

project name
Find out more

Exhibition Planning System

Oracle Distributed Database MVC Architecture Python - Flask

Developed a web-portal using Python (Flask) and MVC architecture to let exhibitors host exhibitions, request and manage participations from several businesses and manage visitor registrations. Designed the database schema and managed user searches and registrations using distributed database architecture.

project name
Find out more

College Events Portal

Event Registrations Discussion Forum Laravel PHP / AJAX

Developed a college portal for displaying notices, managing college events and registrations. Built a discussion forum for college students to connect with event managers or staff for queries.

More on GitHub