Course project: Storage Systems, Carnegie Mellon University Fall 2017
Design of Flash Translation Layer and Hybrid File Systems
Instructor: Prof. Greg Ganger

Course project: Optimizing Compilers for Modern Architectures, Carnegie Mellon University
Spring 2017

ArchRecommender: Mappring Programs to Platforms
Instructor: Prof. Phillip Gibbons

Course project: Computer Systems & Architecture, Carnegie Mellon University Spring 2017

Dynamic Scheduling in Heterogeneous Systems
Instructor: Prof. Brandon Lucia

M.Tech Thesis, Technische Universität München, Germany Sept. 2014- Jun. 2015
Reconfigurable Communication Middleware for FlexRay-based Distributed Embedded Systems
Advisors: Prof. Purandar Bhaduri, IITG & Prof. Samarjit Chakraborty, TUM
Summary: In this report we consider the case of a network of Electronic Control Units (ECUs) connected through a FlexRay bus in the automotive domain. Multiple distributed applications can run on this underlying architecture, each partitioned into tasks that are mapped on different ECUs. These applications can often be executed in different functional modes with different requirements on the communication resources in terms of data size and sampling period. Moreover, new applications can be deployed on to the ECUs at run-time. To efficiently utilize the communication resources and accommodate new applications, a certain flexibility in reallocation of the resource is necessary. However, the FlexRay bus requires static configuration of schedules and data mapping in order to guarantee a more deterministic system behavior, allowing little room for flexibility. In order to address this problem, we propose a reconfigurable communication middleware that lies between the application layer and the communication controller layer, which maps messages onto FlexRay schedules, and can be reconfigured at runtime. The configuration is synthesized and deployed online, allowing a reallocation of communication resource to applications. In this report, we describe the design of such a reconfigurable communication middleware and demonstrate its function with an implementation using industry-strength FlexRay design tools.

Publication: RTCSA 2015
Award: IEEE Technological Innovation Award 2015 for Best Thesis in Eastern Zone of India

Term-Paper: Computational Systems Biology, IIT Guwahati, India Spring 2014
Detection of Novel Drug Targets from Protein-Protein-Interaction Network
Instructor: Dr. Ashish Anand, IIT Guwahati
Summary: Although machine learning techniques proved to be a good predictor of additional drug-targets (here is a review), some claimed that they may not be useful to extend the current drug target inventory to surprisingly novel hits. Now, applying K-Means clustering method on the feature set, we found a small set of drug targets, say, x from the set of known drug targets, having completely different characteristics from the rest. Hence, the moment x was discovered for the first time, it must have been a ‘novel hit’. Now, we find out some unverified proteins that are closest to x in terms of properties, and claim that they must be drug targets. Cytoscape and Matlab were used for all experiments.

Term-Paper: Learning with Kernels, IIT Guwahati, India Spring 2014
Detection of Forest Fire using One-Class SVM
Instructor: Dr. Vijaya Saradhi, IIT Guwahati
Summary: The attributes of the forest-fire dataset were spatial coordinates of the origin of the fire, climatic conditions just before the outbreak and the date & time of the fire. By using this training data, a one-class SVM classifier was created. If newly encountered data is similar, according to some measurement, from this model, the program throws a forest-fire-warning. If newly encountered data is too different, it is labeled as out-of-class (forest is safe) and the program continues monitoring the conditions. 

Term-Paper: Speech Processing, IIT Guwahati, India Autumn 2013
Implementation: Recognition of Vowels in Speech Signal
Instructor: Prof. P. K. Das, IIT Guwahati
Summary: The cepstral coefficients of the vowels were computed and stored in a database. When a new speech signal containing a vowel arrives, linear predictive analysis followed by cepstral coefficient computations were done on the new signal. The vowel whose coefficients are closest to that of the new one (using Tokhura’s Distance measure) is considered to be contained in the signal. Entire implementation was done in C.

Term-Paper: Speech Processing, IIT Guwahati, India Autumn 2013
Implementation: Recognition of Digits in Speech Signal
Instructor: Prof. P. K. Das, IIT Guwahati
Summary: For each digit, we build a Hidden Markov Model based on the training set. Once a new speech signal arrives, we compute the set of ten probabilities that the observation sequence derived from the signal was produced from each of the ten models. We find out the maximum among these probabilities, and the corresponding digit is declared as the digit contained in the signal.

Term-Paper: Computer Systems, IIT Guwahati, India Autumn 2013
Establishment of dedicated path in virtual circuit packet switching network
Instructor: Dr. Pinaki Mitra, IIT Guwahati
Summary: I designed and estimated the complexity of an algorithm based on the concept of flooding to establish a dedicated path between source to destination in the network. The algorithm tries to minimize time of delivery of packets which essentially reduces the path length and simultaneously increases the minimum link capacity of the path.    

B.Tech Thesis, West Bengal University of Technology, India Jun. 2012- Jun. 2013
Detection & Diagnosis of Rust Disease in Wheat Leaves: An Integrated Digital Image Processing Approach
Advisors: Dr. A. Chakraborty, Dr. D. K. Kole (WBUT) & Prof. Dwijesh D. Majumder (Indian Statistical Institute, Kolkata)
Summary: An integrated Image Processing System was built using Matlab and the application was subsequently loaded on a server machine which enables end users to utilize the system even if his/her machine does not have Matlab installed in it. This intranet module allows users to submit a wheat leaf image and the server can return information about the presence of rust disease in the leaf. Through the use of Artificial Neural Networks, detection of the presence of disease in wheat leaves has been successful in 97% of the cases. Also, the type of disease, if present, in wheat leaf has been successfully identified in 85% of the cases.