Anuva Kulkarni

PhD, Carnegie Mellon University

About Me

I got my PhD in 2020 from the Department of Electrical and Computer Engineering at Carnegie Mellon University. I was advised by Dr. Franz Franchetti(advisor) and Dr. Jelena Kovačević(co-advisor).
My PhD thesis dealt with algorithm design for scalable convolutions on modern heterogeneous systems (CPU-GPU clusters, for example). More specifically, my work focused on scaling scientific simulation workloads such as differential equation solvers or spectral solvers that use Fast Fourier Transforms (FFTs) for convolution.
I am currently working as a Software Engineer (Machine Learning) at Google.
Previously, I was a Machine Learning Research Scientist at Motional, a Hyundai-Aptiv joint venture for autonomous driving.

Professional Service

Program Committee member, International Workshop on Accelerating Analytics and Data Management Systems Using Modern Processor and Storage Architectures (ADMS) 2022

in conjunction with VLDB 2022

Program Committee member, International Conference on Parallel Processing (ICPP) 2022

Program Committee member, International Parallel & Distributed Processing Symposium (IPDPS) 2022

Program Committee member, International Workshop on Accelerating Analytics and Data Management Systems Using Modern Processor and Storage Architectures (ADMS) 2020

in conjunction with VLDB 2020

Publications

FFTX and SpectralPack: A First Look

PFFT@IEEE HiPC 2018
Franz Franchetti, Daniele Spampinato, Anuva Kulkarni, Thom Popovici, Tze Meng Low, Michael Franusich, Peter McCorquodale, Brian Van Straalen, Philip Colella

Projects

Indexing for fast Nearest-Neighbor search in Artificial Intelligence (AI)-powered Databases

Developed an indexing algorithm to quickly identify candidates for querying an AI-powered database. This work was done during my summer internship at IBM T. J. Watson Research Center, Yorktown Heights, New York.

Porting Large-scale Fortran Simulations to CPU-GPU platforms

Developing algorithms to port legacy Fortran simulation code involving large scale Fast Fourier Transforms onto CPU-GPU systems. Some issues tackled include high memory requirement and communication overhead in parallel computing.

Model design and analysis automation for big mechanisms

Implemented frameworks for automated development of executable models using information extracted from literature for cancer cell networks

Unsupervised image segmentation using Winner-Take-All Hash and Random Walks algorithm

Developed an unsupervised segmentation algorithm for images, with a focus on biomedical images using the concept of comparative reasoning from machine learning followed by creating a probability map using Random Walks. Designed an automated seed selection method for Random Walks algorithm using Gibbs sampling for Dirichlet Process Mixture Models

Near-fall detection in seniors using head-mounted accelerometers and gyrometer sensor

Designed a system using multinomial logistic regression to predict the probability of instability in daily activities. The system is used for human motion database analysis for recognizing near-falls in senior citizens. It can be deployed on a smartphone to provide real-time feedback

Classification of Histological Teratoma Images

Used wavelet techniques and Gabor filters along with Support Vector Machines to find methods of efficient classification of different tissue sub-types in a histological image. Used Superpixels for test image partition and iterated over Superpixel coarseness to ensure fastest possible computation speed for a fixed desired accuracy

Bachelor Thesis: Algorithms for locating non-defective items in a large population.

Implemented three algorithms for the problem of Group Testing, which is a boolean form of Compressed Sensing. The algorithms were row-based, column-based and convex optimization (L1 - minimization) based. methods to find a small number of non-defectives from a large collection of items with very few measurements. This project was advised by Dr. Chandra Murthy,IISc Bangalore and Dr. Abhay Sharma

Wireless Automated Parking Lot using the LPC1769 ARM processor and a GSM Modem

Designed and implemented firmware for a wireless parking prototype which enables users to reserve space via SMS. Compiled results for analysis of spectrograms obtained using the Short Time Fourier Transform (STFT) for a Vehicle classifier, in MATLAB.

Internet based data acquisition and control of remote field devices using the MSP430 and the CS8900 Ethernet LAN controller

Designed a system to monitor data and send instructions to field devices by means of TCP/IP connection over a LAN. The support of Canopus Instruments, India is acknowledged.

Education

  • PhD in Electrical and Computer Engineering
    Carnegie Mellon University
    2015 - 2020
  • MS in Electrical and Computer Engineering
    Carnegie Mellon University
    2013 - 2014
  • BE(Hons.) in Electrical and Electronics Engineering
    Birla Institute of Technology and Science (BITS) Pilani
    2009 - 2013

Resume