Manohar Karki

Email: ac0jg6@r.postjobfree.com Cell: 469-***-**** URL: http://csc.lsu.edu/~mkarki/

I am a Ph. D. candidate at Louisiana State University looking for Research/Data Scientist positions. My research has been focused on the field of Machine Learning, Computer Vision and Artificial Intelligence. I have been working on developing a framework with the help of Deep Learning to solve Computer Vision tasks such as image reconstruction, image colorization, object classification and segmentation from images at pixel level. Previously, I have also worked on Video Analytics, Satellite Image Recognition, noisy character Recognition, analysis of hyper-parameters of deep networks etc.

Recent Achievements:

-Conceptualized and built a framework for pixel level classification (or regression) of images using transfer learning with a help of a pre-trained Convolutional Neural Network (CNN)

-Worked on a vehicle and human tracker and built the activity recognition system from videos for the Department of Defense (DOD) sponsored ARO/DARPA project

-Initiated and helped to build the Satellite Imagery Segmentation Project sponsored by NASA that classifies high resolution satellite images

-Created Noisy datasets of characters and used a system that segments the noise out and uses Quad trees to learn features for character recognition

Education:

Louisiana State University, Baton Rouge, LA

Pursuing PhD in Computer Science GPA: 3.91

University of Texas at Arlington

Bachelor of Science in Software Engineering GPA: 3.62

Relevant Experience and Projects:

Graduate Research Assistant, Robotics Lab, LSU (August 2011- Present)

I was involved in various research projects related to Computer Vision and Machine Learning:

Core-sampling Framework for pixel-wise classification: I designed and implemented a segmentation by pixel-wise prediction framework that uses GPUs to increase training efficiency. The system was designed to detect objects in SAR (Synthetic Aperture Radar) images but has various applications. The framework has comparable performance to state of the art segmentation algorithms such as SegNet. Other applications that I have used it for are Automatic Colorization of Grayscale and Near-Infrared Images on large data with no input features. The framework was designed using Python, Convolution Neural Networks (CNN), Deep Belief Networks (DBN), Theano, caffe etc. [more information]

Video Motion Classification for DARPA : I conceptualized and designed the symbolic framework for recognizing activities in videos for singular and composite activity recognitions such as walking, running, digging, robbery, vandalism etc. and was involved in implementing parts of the tracking algorithms for vehicles etc. Used MATLAB, C/C++ with OpenCV and SVM, Neural Networks, Random Forest as classifiers. [video activity recognition demo links]

Character Recognition by using Quadtrees and Pixel-level Segmentation: I created noisy datasets of Bangla Numeric and Alphabet character images of at different levels of white noise, motion blur and contrast variations then used pre-trained CNN, Probabilistic Quad trees and DBN to aid Classification. I have been solely working on the implementation of this project.

High Resolution Segmentation Project: I was involved in designing the initial phases of the segmentation project where the system was able to classify Satellite Imagery as forests, dirt, grass, buildings, roads etc. I used texture features, Vegetation Indices, and other image processing to segment different classes appropriately using Machine Learning (one vs all classifiers). [article]

Student Assistant for Research, UT Arlington (September 2008 – December 2008)

I assisted PhD students on experiments (strength of cement blocks consisting of variable ratio of steel, cement, sand and other parameters) for the Civil Engineering Department.

Intern Developer, T-Mobile Summer 2008

I developed a web-based application using Microsoft ASP .NET, C# in Visual Studio and used it to visualize total run time, errors, frequency of use of serialization processes etc. I also debugged, made Test Cases and deployed the application.

Publications:

1.Learning Sparse Feature Representations using Probabilistic Quadtrees and Deep Belief Nets, Neural Processing Letters Journal 2016

2.Semiautomated Probabilistic Framework for Tree-Cover Delineation From 1-m NAIP Imagery Using a High-Performance Computing Architecture, in Geoscience and Remote Sensing, IEEE Transactions on, vol.53, no.10, pp.5690-5708, Oct. 2015 doi: 10.1109/TGRS.2015.2428197.

3.A Symbolic Framework for Recognizing Activities in Full Motion Surveillance Videos, IEEE Symposium Series on Computational Intelligence, IEEE SSCI 2016

4.A Probabilistic Real Time Tracking Framework by Integrating Motion, Appearance and Position Models. In Proceedings of the 10th International Conference on Computer Vision Theory and Applications 2015

5.An Agile Framework for Real-time Motion Tracking, IEEE COMPSAC Workshop 2015

6.DeepSat - A Learning framework for Satellite Imagery, In Proceedings of ACM SIGSPATIAL 2015.

7.A Theoretical Analysis of Deep Neural Networks for Texture Classification, International Joint Conference on Neural Networks, IJCNN 2016.

8.Learning Sparse Feature Representations using Probabilistic Quadtrees and Deep Belief Nets, European Symposium on Artificial Neural Networks, ESANN 2015

Patent (pending):

COMPUTER IMPLEMENTED SYSTEM AND METHOD FOR HIGH PERFORMANCE VISUAL TRACKING

Computer implemented systems and methods for tracking the motion of objects in video, detecting patterns in the motion of objects, and reasoning about tracked motion and patterns to identify activities or events.

Contact this candidate