Projects

BioniVisionXR

• An Open-Source Virtual Reality Toolbox for Bionic Vision
• Featured at Unite Conference 2024
• Uses Pytorch and Unity Sentis to create neurophysiologically inspired and psychophysically validated models to simulate the visual experiences that could be generated by future bionic eye implants.

Neural Manifold Analyses of Working Memory with a Cortical Visual Prosthesis Spring 2024

• Problem: Analyze the neural dynamics and enhance decoding accuracies associated with a visual working memory task.
• Uses dimensionality reduction techniques like GPFA and tSNE to improve task decoding accuracy on the neural recordings.
• Findings suggest task-specific neural differences, and contributes to understanding neural underpinnings of working memory in visual prosthetic users

Graph Approaches for Adaptive AR

• Course Project for CS 292F: Machine Learning on Graphs (Spring 2022)
• Problem: assigning objects of a virtual scene to new positions in an arbitrary real-world location.
• Compared how purely geometrical algorithms and Graph ML techniques perform on this problem

Evaluation of Subretinal Prosthetic Implant (PRIMA) Simulation for Visual Acuity

• Course Project for CS 291A: Bionic Vision (Fall 2021)
• Built a gaze-contingent VR experiment to simulate PRIMA prosthesis
• To evaluate the visual acuity that can be acheived by PRIMA, sighted participants can participate in this experiment and perform Landolt-C acuity test
• Developed using HTC Vive Pro Headset, Unity and pulse2percept

noGram

• Course Project for CS 291A: Scalable Internet Services (Fall 2021)
• Implemented a basic Social Network Web app using Ruby on Rails with a PostgreSQL database
• Performed different tSung Load tests to analyze bottlenecks and scalability to better handle traffic
• Performed Cost Analysis on AWS to determine best hardware configuration for a user arrival rate

Flick Gesture Interaction in Augmented Reality: AR Carrom

• Course Project for CS 291A: Future User Interfaces (Spring 2021)
• Built an AR app with a flick gesture mechanic that utilizes hand gestures and doesn’t rely on any external hardware
• Developed using Unity, ARCore and Manomotion

Performance Analysis and Optimization on Edge AI devices

• Course Project for CS 263: Runtime Systems (Spring 2021)
• Aim: To study Optimization Techniques to increase inference performance on Edge devices
• Studied Nvidia’s Nsight profiling tool and compared performance of basic AI models using TensorRT on Jetson Nano

Facial Emotion Recognition

• Course Project for CS 265: Machine Intelligence (Winter 2021)
• Aim: Given an image, identify the emotion expressed within the image
• Compared the performance of different classifiers (SVM, CNN, GAN) and dimensionality reduction techniques (PCA, AutoEncoders(AE), Convolutional AE) on FER 2013 dataset
• Built a Super Resolution model using AEs to study the effects of super resolution images on classification task

Cancer subtype detection using Human Gene Expression data

• Bachelor’s Project, Advisor: Dr. Clint P. George
• Aim: Clustering and detection of different cancer types using the Gene Expression data through Unsupervised Clustering & dimensionality reduction techniques to help in diagnostic and to reveal possible previously unknown cancer subtypes.
• Studied Autoencoder model and implemented it on Breast Cancer Data using Keras API to get a reduced representation of the gene data which retains the important information.
• Formulated possible optimizing ideas to get better reduced representation so that the model can generalize to all cancer types.

Bird Speech Recognition

• Course Project for Introduction to Machine Learning (Spring 2019)
• Built a Hidden Markov Model (HMM) model to classify Bird species using their sound recording.
• Implemented Spectrogram, Short Fast Fourier transform (FFT) & Gaussian HMM using Python.