This work was done in collaboration with Dr. Christin Glorioso (Epidemiologist, Physician, and Founder and CEO of the non-profit research organization, Academics for the Future of Sciences (AFS))

I was selected as a summer research student under the Openlab Program to collaborate with Dr. Gianluca Cerminara and Adrian Alan Pol of the Compact Muon Solenoid group at CERN

The project aims at applying recent progress in Machine Learning techniques to the automation of quality assessment allowing the check of large volumes of data in real-time and improving the ability to detect unexpected features. A test implementation using an unsupervised machine learning model (autoencoder) focused on the data of one of the CMS muon detectors (drift tubes) has been developed and bench-marked on real and fake data.

• Openlab Presentation of the project
• Report

For my undergraduate capstone project, I worked as an undergraduate research student at Institut de Recherche en Informatique de Toulouse under the guidance of Prof. Josiane Mothe.

The project aims at determining the flow of a particular news article on Twitter and on news channels. Further, based on the flow, we determined how the two mediums of flow affect each other in terms of the subject’s popularity.

During June-July 2015, I got an opportunity to work as undergraduate summer research at Institut de Recherche en Informatique de Toulouse under the guidance of Prof. Josiane Mothe with her team SIG.

During this period, I performed a comparative study intending to find a better method of indexing Clueweb12-B(shards and complete dataset) using terrier. After that, I devised a method for automated query reformulation using the concepts of graph theory such as indegree of nodes linked to the query terms and used them to query the ClueWeb12-B dataset. The algorithm showed positive results with a subset of the actual data, and the testing for the complete dataset is scheduled for December 2015.

In November 2014, I got an opportunity to work as a research apprentice to Prof M.C. Govil at Malaviya National Institute of Technology on fault tolerance in cloud systems.

During, this period I tried re-implementing Fault Management in Map-Reduce through Early Detection of Anomalous Nodes” by Selvi Kadirvel, Jeffrey Ho, José A. B. Fortes. I monitored the heartbeat of the Hadoop cluster using Ganglia. Further, I created a Model and performed Single label SVM to classify the healthy nodes and recognize the anomalous nodes.

The complete project report for the same is available here.

The working video can be found here; its source code is available at Github.

The result for the same is displayed below:

After successfully creating the above environment, I decided to use 5dt Data Gloves to navigate through the system and give an immersive 3D animation. I used the Vizard platform to perform the animation.

Geometric Modelling presented some major disadvantages, from being labor-intensive to time-consuming. Thus, I used the 3D warping technique. In this, several 2-D images were used, and after mapping the corresponding pixels in these images, Point Cloud Library was used to create their 3-D model in OpenCV using python. The result for the same is displayed below:

After the above implementation, I performed a comparative study of Geometric Modeling and Image based Rendering.Pre-print for which is at Arxiv.

The position of the sun continuously keeps changing throughout the day. Hence to make the solar devices more efficient, I created a device that changes direction with change in the light source position using Arduino, servo motors as the rotatory mechanism.