There also used to be a long section here, stating to the recruiters about what my skills are and what are my programming languages. But I had, slightly passed the prime age/qualifications of getting an entry positions hence I won't talk about those things. I know programming more than your average entry level programmers, and explosively good at algorithms and computing in Julia/python. I grew tired of programming cool stuff and failed to get a job, while all these time, failed to realize mathematics were my true sanctuary.
Publications soon.
Made with for the purpose of practicing GRE verbal reasoning tests and web-development
Page LinkThis is a Conway's Game of life written in javascript. It runs fasts and it's really cool.
Page Link
I wrote in details about the Knapsack problem, with mathematical proofs and Psudo codes and link to repo with implementations, topics include:
The Branch and bound algorithm for the extended knapsack, (or branch and bound algorithm for LP in general) experiences numerical instability due to the IEEE 754 representation of number. Which is a common standard for most programming language. Here we have a discussion on the following:
The K-Minimum Spanning tree's solution could be interpreted as a "Clusters on Non-Gaussian points". In this project, we are concerned with some of the variations of the Kruskal Algorithm and how it can be applied for Classfications proboem, more specifically:
This project failed.
We are comparing the robustness of L1, L2 Norm as loss function when it comes to sensitivity to outliers in the data.
Implemented in python, uses matplotlib, numpy and CVXOPT for solving the L1 Optimization problem.
Page Link
We are going to look into the conjugate gradient method in details and then apply it to the task of image deblurring.
the whole algorithm is implemented in python using numpy, uses matplotlib for visuals and scipy signal for speedy image convolutions.
Page Link Source Code Link
Conjugate gradient method can be derived using the idea of minimizing the energy norm of the error vector over the Krylov Subspace.
No code is involved this is purely math.
Page Link
The frustrations with ADMM based Conic Solver and Julia Optimization Frameworks sparks my own efficient implemention of accelerated proximal gradient descend for the lasso path problem.
We have math and code and images.
Page Link Source Codes Link
Lanczos Algorithm and Conjugate Gradient
We review results from the literature on the conjugate gradient algorithm for solv- ing symmetric positive definite linear systems and the related Lanczos algorithm. We derive the conjugate gradient algorithm from the more general conjugate direction method, using projectors. We establish error bounds using exact arithmetic theory and also discuss what can happen when floating-point arithmetic is used. We present numerical experiments to illustrate this behavior.
Download Thesis Source Codes Link