Automatic Differentiation is a tool to calculate exact numerical derivative of a computer function. In this summer internship, I worked on the Reverse Mode Implementation of AD in the MAD package written by Prof Shaun A Forth. I implemented the editing and re-evaluation of global madtape to reduce the computation time efficiently by more than 5 times.
Report can be downloaded from download section.
Particle Swarm Optimization is one of new heurestic search methods for optimization. We tried to implement a new algorithm GEPSO which is a hybrid of PSO and gradient based approach. On each iteration, GEPSO tries to search new particle by using the gradient information of Global Best and Particle Best instead of their velocities.
The work also includes testing the algorithm on 7 Bench Mark functions. The implementation is done in MATLAB and includes a Particle class with global generation tapes and standard methods for this class. For gradients information, GEPSO implementation uses MAD class's forward mode operations.
Report can be downloaded from download section.
Unmanned air vehicles (UAV) are considered as technologies of future. Obstacle avoidance is one of the most challenging tasks which the UAV has to perform with high level of accuracy.
In this paper, obstacle is considered as a spherical threat zone and a geometric approach is used to develop the algorithm to avoid the collision. The algorithm uses particle swarm optimization (PSO) to evaluate the final parameters of the maneuver. The efficacy and gist of PSO are justified in the paper.
The results derived from PSO are verified with other existent mathematical tools to solve a non linear equation. The comparison is done with Muller's method, Newton-Raphson's method with Automatic Differentiation and Symbolic differentiation. The algorithm is simulated using MATLAB. The level of accuracy and simplicity of the algorithm is demonstrated.
This work is also got published in IEEE INDICON Annual Conference 2008. IEEE Paper can be downloaded from download section/ IEEEXplore website.
This work include generation of a spiral mesh for providing surface to a particle extraction simulation based on DEM method. The implementation is done in MATLAB. A GUI is also developed for user to select different parameters like radius and height along with number the turns in the spirals.
The algorithm first triangulate the 2D projection and then lift the surface to give a spiral surface. There were options to adapt the 2D triangulation according to height to give a more uniform 3D structure.
Report can be downloaded from download section.