Here are some of the projects which I have done in the past

Improving the Efficiency of IP Traceback at the DoS Victim (Report in pdf)
    In this work, an improvement to the IP Marking approach for traceback was proposed which reduced the convergence time before a victim can begin traceback during a DoS attack. Also, a Variable Marking Probability (VMP) scheme was proposed where, instead of marking IP packets with a constant marking probability, the probability of a router marking a packet would depend on the router's distance from the victim. This was done to even out the probability of receiving marked packets from all routers along the attack path. Simulations performed showed that the convergence time was reduced considerably
A Protocol Architecture for Wireless Sensor Networks (Report in pdf)
    This project involved the design of a communication protocol architecture for Wireless Sensor Networks (WSNs). At present there is no interoperability between different sensor network components. In this work, a standard communication architecture was introduced which consisted of a control plane abstraction for reliable delivery of control data. This work introduced about the concept semi-reliability in sensor networks and talked about ways to achieve them
Information Extraction from the MUC-3 terrorism corpus (Report in pdf)
    This project involved extraction of various kinds of information (such as targets, victims, weapons used, perpetrator, organization involved) from news articles relating to terrorist incidents belonging to the MUC-3 corpus. Caseframes which identify sentence structures were automatically generated from a tagged training corpus and applied for information extraction on the test corpus. A very robust partial parser for English was developed which included functionalities like clause segmentation, coreference resolution and subject inference for subject-less clauses.
Face Identification using Neural Networks
    The goal of this project was to match a human face in a photo (with varying orientation, shading, noise etc) with one from a database of faces. A 3-layered perceptron neural network with backpropagation for error correction was developed and used for this purpose. With minor changes, the same algorithm was also used to box out all the human faces in a group photograph.
Parallelized Face Recognition using Principal Component Analysis (PCA)
    The PCA algorithm for face recognition was implemented. Eigen vectors of a face photograph was used to create eigen-faces (which are basically representations of the face in a lower dimension). There were then used for the purpose of comparison with a database of faces. Furthermore, this algorithm was parallelized and implemented on the PARAM supercomputer. Implementation was done with the help of the MPI library on Solaris platform.
Design and Implementation of a Peer-to-peer file sharing system
    A study of structured and unstructured peer-to-peer networks was carried out with focus on resource location in the GNUtella P2P network. A peer-to-peer resource locator was designed and implemented in Java.