Resume

I am a Computer Science and Mathematics major with interest in data analytics, big data, image processing/analysis (of remote sensing, satellite images, medical images, etc), high performance computing (accelerators, MPI, OpenMP, Python), mathematical modeling, and numerical optimization.

I hold a Ph.D. from the Arizona State University in applied mathematics. My thesis focused on 'High Order Methods for Edge Detection and Applications'. I subsequently taught at Michigan State University and University of California, San Diego while pursuring research in molecular surface reconstruction as postdoctorate and visiting researcher repectively.

My current appointment is at Virginia Tech, Blacksburg. My research is in the area of land usage and trend detection from satellite images and is a collaboration between Forest Resources and Enviormental Conservation and the Department of Computer Science both at Virginia Tech.

Data analytics, statistics, machine learning

Big Data and High performance computing (Hadoop, GPU, MPI, OpenMP)

Signals, images, surfaces and other multivariate data: Edge detection, reconstruction, smoothing, denoising, visualization.

Remote sensing, satellite images, medical images

Spectral methods and their applications.

Mathematical modeling, numerical optimization, visualization

LULC-Net: Generalized convolution neural network classfier for detecting land use changes from satellite data

Preliminary investigation of a novel application of convolutional neural networks (CNNs), viz., locating regions that have undergone land use and land cover changes.

• Developed an algorithm, that strategically combines known algorithms and labels data to produce high quality labels describing land use and land change.
• The CNN currently consists of 3 convolutional and 3 fully connected layers.
• All training and validation was done on NVIDIA p100 GPU which features 3584 1GHz cuda cores with 12GB of memory.
• The work has broader significance on detecting changes from image stream from motion camera and webcam.

For a detailed report on the implementation and report see: lulc-net.pdf

Detection of trends and changes in land use using satellite images

Designed and developed an ensemble based approach for change detection from spatio-temporal stack of satellite images. The approach collates answers from principally different unsupervised and supervised techniques:

• dynamic programming based method
• harmonic regression based learning method
• random forest based classification tree method and
• convolution neural network based method (under development)

For a synposis of the results from this work see: here

Detecting users with dependency traits in Twitter

Developed technique to identify users exhibiting dependence (anomously high activity) to a particular lifestyle such as cosmetics, shopping, etc.

• Developed focussed crawler that, starting with a set of seed users, grows the potential pool of candidate users through biased exploration of the follower-followee graph.
• Analyzed approx. 10 million users, 10 billion tweets, 1 billion edges, and over 20 facets of the users.
• Analytics was written in python while data managment was done using database. This allowed for concurrent processing of various phases of the analytics pipeline

Rendering images from MRI sensors using inversion of circular radon transforms

The geometry and the mechanism for capturing signals inside a magnetic resonance imaging (MRI) scanner is complex and challenging -- the sensors are placed along circular arc where they capture the signal in frequency domain. This project proposed derivations and algorithms to reconstruct the signal in eucildean-time domain. The project

• represented non-planar MRI scans as recursive integrals
• developed methods that provide approximate answer for the integrals efficiently, as exact numerical solution turned out to be computationally demanding
• identified generic programming (C++ templates) as natural constructs to express recursive integrals and their computations
• developed strategy to parallelize the computation using linux threads
re

Image compression using generalized low rank approximation

Implemented generalized low rank approximation (GLRA) for matrices and applied it for image compression.

• obtained 30% data compression ratios, on an average
• GLRA has better time complexity (several orders of magnitude faster) than SVD.
• Reconstruction and classification done using GLRA-compressed images was better than that done using SVD compressed data.
• Codes were implemented in Python.

For a synposis of the results from this work see: here. Code, datasets, and the report is posted on Github

High Order Methods for Edge Detection and Applications

My dissertation mostly focused on the use of polynomials for the purpose of image processing, in general, and edge detection, in particular. For a synposis of the results from this work see: here

• Towards a Polyalgorithm for Land Use Change Detection, R. Saxena, L. T. Watson, R. H. Wynne, and V. A. Thomas, Computers and Geosciences, in preparation
• Scaling Constituent Algorithms of a Trend and Change Detection Polyalgorithm R. Saxena, L. T. Watson, V. A. Thomas, and R. H. Wynne, in High Performance Computing Symp. (HPC 2017), Proc. 2017 Spring Simulation Multiconference, Society for Modelling and Simulation International, Vista, CA, 12 pages, 2017
• Damage Localisation in Plate Like- Structures Using the Two-Dimensional Polynomial Annihilation Edge Detection Method
• Damage Localisation in Plate Like- Structures Using the Two-Dimensional Polynomial Annihilation Edge Detection Method C. Surace, R. Saxena, H. Darwich, and M. Gherlone, Journal of Sound and Vibration, 333 (21), pp. 5412-5426, 2014
• On the use of the polynomial annihilation edge detection for locating cracks in beam-like structures C. Surace, R. Archibald, and R. Saxena, Computers & Structures, 114-115, pp. 72-83, 2013
• A Multiscale Model for Virus Capsid Dynamics C. Chen, R. Saxena, and G.-W. Wei, International Journal of Biomedical Imaging, Article ID 308627, 2010}
• Discontinuity detection in multivariate space for stochastic simulations R. Archibald, A. Gelb, R. Saxena, and D. B. Xiu, Journal of Scientific Computation, 228, pp. 2676-2689, 2009
• High Order Method for Gradient Edge Detection R. Saxena, A. Gelb, and H. Mittelmann, Communications in Computational Physics, 5 (2-4), pp. 694-711, 2009