Akash Nagaraj

Akash Nagaraj

Research Scholar at Brown Univesity

Posts by Collection

portfolio

Multiview Masked Autoencoders for 3D Vision

Developing robust, interpretable, and grounded representations using multi-view spatiotemporal NeRF data of 3D models and sequences from camera-captured videos using 3D Gaussian Splatting. A novel spatiotemporal vision model with a causal vision modeling objective inspired by human visual development.

Closed-loop optogenetic system for deep behavioral phenotyping

Closed-Loop Optogenetic System for Evoked Responses for automated sensory stimulation integrated with comprehensive behavioral analyses incorporating: i) real-time markerless pose tracking that guides precise laser stimulation of freely moving mice, enabling fully automated, standardized activation of genetically tagged neuronal subtypes innervating the skin; and ii) a computer vision pipeline for automated behavior classification.

ProjectCARE: Suicide Prevention from Tissue Damage Images

An NIH-funded clinical decision-support tool to automatically to locate and characterize non-suicidal self-injury tissue damage in images for suicide risk prediction, with the potential to lower suicide rates using custom objective functions and training methods to learn from our noisy, real-world data.
In collaboration with Harvard University and Massachusetts General Hospital.

Mouse Behavior Tracking

An objective framework to characterize the influence of external cues, genes and neural activity on behavioral modules from videos, using Generative Modelling, Transformers & Representation Learning.
In collaboration with National Institutes of Health.

Sensorium: Visual Cortex Modeling

Accurate predictive models of 28,000 neurons from primary visual cortex responses (captured using calcium imaging) to thousands of natural stimuli. Achieved a single trial correlation of 0.41 using an optimized HMAX model with neural circuits & recurrent connections.

Real-time Action Recognition

Action Recognition System for automated real-time hand sanitization auditing, working with real-world data from PES Institute of Medical Science and Research, primarily aimed to reduce bedside infections in hospitals, using a new approach to Action Recognition in videos.

Real-time Automated Answer Scoring

A system that evaluates descriptive answers in real-time, automatically, built using word embedding models and Ensemble Learning. Used to track a student's progress and scoring trends for an answer.

Rahat: AI-based Disaster Management Platform

Multilingual, end-to-end, AI-based disaster management platform using a custom protocol over GSM (no internet required). This project was my entry to Microsoft code.fun.do++, and ranked 4th amongst 6000+ entries.

Unvoiced: Sign-languge to Speech

A real-time application, using a camera for a live video feed, and in the live stream, detect alphabets signed by the sign language speaker and convert that into text and further into speech, to aid in communication with people who have trouble with spoken language due to a disability or condition.

Digital Image Forensics

An algorithm that identifies which camera was used to capture an image using traces of information left intrinsically in the image, using filters, followed by a deep neural network on these filters. Detection of the source camera from a given image using noise, Fast Fourier Transformations, and other parameters, to train Deep Learning Models in an ensemble.

Photogrammetry 3D Printer

A 3D printer built from scratch which takes a 2D image as input and prints out the image in 3 Dimensions, using a combination of Photogrammetry.

CDAE4InfoExtraction

A characteristic extraction algorithm and the Multi-domain Image Characteristics Dataset of characteristic-tagged images to simulate the way a human brain classifies cross-domain information and generates insight. The intent was to identify prominent characteristics in data and use this identification mechanism to auto-generate insight from data in other unseen domains.

litFS: A lightweight File System

A FUSE file system built using the Bazil framework in Go. Apart from providing normal I/O operations on files and directories, it also achieves persistence across machine reboots by emulating a single Unix file as a disk for the file system.

NGO Protal

This project aims at being a complete, end-to-end solution for the functioning, and management of the school, which in our case is the Non-Profit Organization Sri Vidya Nikethan School, Bangalore.

NotePal

A simple, intuitive text editor for Linux, with drag and drop functionality, and a GTK based Front End.

r2ic: Rust to Intermediate Code

A simple front end compiler for converting basic constructs of the Rust programming language (if, if-else, for, while, loop) to a form of intermediate code (quadruples). It also supports generation of an abstract syntax tree (AST). The code generated has three kinds of optimizations - constant folding, constant propagation, and loop unrolling.

Learning Algorithms to detect Vulnerabilities in Source Code

A project to detect key security vulnerabilities (buffer overflows, SQLi, etc.) in web application source code using a combination of static fuzzing and machine learning. The project is sponsored by Hindustan Computers Limited (a Indian-based multinational IT services company).

sesh: Simple Elegant Shell

A simple, elegant Unix shell environment. It supports piping, redirection, history, aliasing, background processes, and customizable functions.

watchIt

watchIt is a Smartwatch built entirely from scratch, using an ARM based micro-controller and has multiple sensors and modules to do more than the average Smartwatch at the present time can.

publications

teaching

Teaching Assistant: CS-314 Big Data

Department of Computer Science, PES University, 2018

Taught and mentored students for Dr. Dinkar Sitaram’s course CS-314: Big Data, with 60 students.

Teaching Assistant: CS-341 Cloud Computing

Department of Computer Science, PES University, 2018

Taught and mentored students for Dr. KV Subramaniam’s course CS-341: Cloud Computing, with 44 students.

Course Design and TA: CLPS-1291 Computational Methods for Mind, Brain and Behavior

Department of Cognitive, Linguistic and Psychological Sciences, Brown University, 2022

Designed and created assignments for the Dr. Thomas Serre’s course CLPS-1291: Computational Methods for Mind, Brain and Behavior, with ~50 students.

Course Design and TA: CLPS-1291 Computational Methods for Mind, Brain and Behavior

Department of Cognitive, Linguistic and Psychological Sciences, Brown University, 2023

Designed and created assignments for the Dr. Thomas Serre’s course CLPS-1291: Computational Methods for Mind, Brain and Behavior, with ~60 students.