Anirud Thyagharajan

About Me

Hi! I'm a Research Scientist at Intel Labs (Processor Architecture Research Labs, India), where I work on AI algorithms and architecture research. My research areas lie broadly in computer vision, machine learning and computer architecture. At Intel Labs, I work on AI algorithm research along with exploring novel accelerator architectures for emerging workloads.

Being a Dual Degree graduate from the Department of Electrical Engineering, IIT Kharagpur, I received an Institute Silver Medal for being the top scorer in the EE Department, along with an award for the Best Masters Thesis Project.

I also play the tabla, a Hindustani classical percussion instrument. You may also find me rambling about here in my blog, or dabbling with ping pong.

Latest Project

Towards Noise Resilient SLAM

Proposing blocks to improve robustness against depth noise

Sparse-indirect SLAM systems have been dominantly popular due to their computational efficiency and photometric invariance properties. Depth sensors are critical to SLAM frameworks for providing scale information to the 3D world, yet known to be plagued by a wide variety of noise sources, possessing lateral and axial components. In this work, we demonstrate the detrimental impact of these depth noise components on the performance of the state-of-the-art sparse-indirect SLAM system (ORB-SLAM2).

We propose (i) Map-Point Consensus based Outlier Rejection (MC-OR) to counter lateral noise, and (ii) Adaptive Virtual Camera (AVC) to combat axial noise accurately. MC-OR utilizes consensus information between multiple sightings of the same landmark to disambiguate noisy depth and filter it out before pose optimization. In AVC, we introduce an error vector as an accurate representation of the axial depth error. We additionally propose an adaptive algorithm to find the virtual camera location for projecting the error used in the objective function of the pose optimization. Our techniques work equally well for stereo image pairs and RGB-D input directly used by sparse-indirect SLAM systems.

Our methods were tested on the TUM (RGB-D) and EuRoC (stereo) datasets and we show that they outperform existing state-of-the-art ORB-SLAM2 by 2-3x, especially in sequences critically affected by depth noise.

Other Projects

Metric Learning for Face Recognition

Development of a face recognition system using multiple metric learning algorithms and using custom ensemble techniques for increasing the veracity of the person thus recognised from the set of algorithms.

Design of an Attendance System for Large Classrooms

Development of a face detection system for automating attendance procedures with high number of attendees, using image processing algorithms such as the Viola Jones face detection algorithm. Construction of a real time registration procedure for students based on a remote server in the class for validation, and a prototype successfully developed on a Raspberry Pi. Developed as part of my Bachelor Thesis Project under the guidance of Prof. Aurobinda Routray at IIT Kharagpur.

Automated Essay Scoring

Designed an automated essay scoring system using machine learning and natural language processing techniques as part of a Term Project in a course, "Language Processing for eLearning" under Prof. Plaban Kumar Bhowmick. Updated new features and estimated their relevance on the Kaggle dataset.

Handwritten Digit Recognition using SVMs

Employed SVM kernels and feature selection wrappers to classify digits from the famous MNIST database as part of a Term Project in an additional course, "Machine Intelligence and Expert Systems" under Prof. Sudipta Mukhopadhyay

Network Security WarGames

Worked around with a number of network security wargames based on SSH, Unix/GDB hacks, bufferoverflows and other related exploits. A few of the games include Bandit, Leviathan, MicroCorruption, Krypton, etc.

Image Restoration

It revolves around the making of an Android Application in Java, and a deconvolution process, which will de-blur the preview and amplify the image size using which people can read small font scripts by placing their mobiles over the text and using our application. Under the guidance of Prof. Sudipta Mukhopadhyay, the project is aimed at deblurring of the mobile camera sensor to enable better quality output to people who have deterred eyesight.

Work Experience

Research Scientist - Intel Labs (May 2017 - Present)

Work on AI algorithm and archiecture research at Processor Architecture Research Labs.

Interim Engineering Intern - Qualcomm India (May - July 2016)

Being part of the Linux Data team at Qualcomm, the project involved implementation of an architecture redesign for optimisation during the SSR scenario, and succeeded in reducing the system recovery time by 50%. Apart from suggesting a suitable hybrid solution, studied and outlined various call-flows of the Linux Data Architecture.

Systems Engineering Intern - Minio Technologies (December 2015)

Ideating in the IoT domain, the work involved the areas of Network Security in their communication module and Software Engineering for custom package management in OpenWrt, on a RaspberryPi.

Interim Engineering Intern - Qualcomm India (May - July 2015)

Being part of the Linux Data team at Qualcomm, the work involved an off-target test framework, fixing several bugs and creating new modules in the framework. The project revolved around implementing inter-process communication between test cases and interfacing with several components of the Linux Data Architecture.

Data Science Contributor - HackerRank (December 2015)

Contributed to machine learning related content on HackerRank, involving tutorials and programming exercises, aimed at starters and intermediate learners.

Summer Research Intern - IIT Bombay (May - July 2014)

Assembled and designed a student friendly economical hobbyist electronic kit consisting of a signal generator, frequency counter, PC oscilloscope and AC & DC voltmeters as part of the summer project under Prof. Girish Kumar, EE Department, IIT Bombay.

Competitions

Ideaquest (Runners Up) - Qualcomm June 2016

Successfully completed a working prototype of Placefully, a location based services application, consisting of a myriad of features viz. locating friends, adding reviews, and novel ways to view reviews and emergency protocols. Engineered an end to end communication system between an Android application and a multithreaded HTTPS server.

Ideaquest (Winners) - Qualcomm June 2015

Integrated audio processing with machine learning frameworks to construct an Android application to detect diseases like Parkinson’s disease, Alzheimer’s disease, etc. using voice samples. Composed of a Python backend with Yaafe & PyBrain for feature extraction and neural networks, with an Android frontend.