prashantrivedi Goto Github PK

graph-coloring-problem

Given an undirected graph, the graph coloring problem is as follows: color the nodes such that adjacent nodes have different color. The aim is to find the minimum number of colors needed. The integer program for this problem is implemented.

interview-question-data-science-

lp-vs-milp

The simplest application of MILP is when the variables model discrete objects that can only take whole-number values (e.g. number of aeroplanes to manufacture). A commander in Indian Airforce needs to purchase aircraft for transporting goods and people. She wants to have maximum possible carrying capacity. Five different types of aircraft are available. Their capacities are 2; 1; 2; 1; 2 units per aircraft, respectively. The budget is limited to 13 units (say, crores). The costs per aircraft are 3; 3; 2; 2; 3 units, respectively. In addition, parking space is limited to 11 units. The space requirements per aircraft are 2; 1; 3; 1; 2 units, respectively. We model it using AMPL to findthe number of aircraft of each type that can be bought in order to maximize the goal. Sensitivity analysis is also carried out.

maccm

Contains the codes for our AISTATS 2023 paper titled `Multi-Agent Congestion Cost Minimization with Linear Function Approximations'

multiclass-perceptron

In the online setting, learning takes place in a sequential order and at each step . At each step, the learner makes a prediction or take some action, each of which can potentially result in some loss, and the goal is to update the prediction/decision model at the end of each step so as to minimize the total loss over a sequence of trials. So far, many online learning algorithms have been discussed. Perceptron algorithm is one of them. A lot of literature is available related to 2-class Perceptron algorithm but only few focus on Multi- Class Perceptron algorithm which is one of the natural extension of 2-class. In the present study, we focus on Multi-Class Perceptron algorithm and propose a new update rule and compare the results with the existing rules based on stability of norm of weights and the number of mistakes made by the algorithm.

online-convex-optimization

Consider the problem of prediction with expert advice with d = 10. Assume that the losses assigned to each expert are generated according to independent Bernoulli distributions. The adversary/environment generates loss for experts 1 to 8 according to Ber(.5) in each round. For the 9th expert, loss is generated according to Ber(.5 − ∆) in each round. The losses for the 10th expert are generated according to different Bernoulli random variable in each round for the first T /2 rounds, they are generated according to Ber(0.5 + ∆) and the remaining T /2 rounds they are generated according to Bernoulli random variable Ber(0.5 − 2∆). ∆ = 0.1 and T = 10^5. It is implemeted for FLT and FTRL.

prashantrivedi.github.io

reinforcement-learning

Implementation of Reinforcement Learning Algorithms. Python, OpenAI Gym, Tensorflow. Exercises and Solutions to accompany Sutton's Book and David Silver's course.

reinforcement_learning_course_materials

Lecture notes, tutorial tasks including solutions as well as online videos for the reinforcement learning course hosted by Paderborn University

rl-workshop-iimbangalore

simulating-passengers-risk

Security is one of the major and fundamental need of any country. Everyone wants their system such that there is no fear of any kind of threat but there is always a possibility of breakthrough in the system. What many would think of doing is to employ all the possible technology and always trying to be on the top of the game, but is it really worth it to invest such huge money if the improvement is not quantiable? In particular, if we talk about aviation security, then it's major goal is to prevent harm to aircraft, passengers, and crew, as well as to support national security and counter-terrorism policy1. While lots of efforts have been done to improve the reliability and safety from the engineering point of view, new security threats are emerging day by day. A very basic question that arises now is: Do all the new security measures really make the aircraft more secure, and if yes, can we quantify it? The security objective then becomes that of ensuring the integrity of the process. The answer to such a question is not at all easy. In general the efficiency of the entire airport is checked with the help of Red Teams by Department of civil aviation. This method is time-consuming, very expensive and ad hoc. We, therefore, use the tools of simulation to answer some of such highly non-trivial questions that can arise in the airport security system. Mainly we did the analysis for Bombay airport, but similar analysis for any other airport or group of airports can be done. This is joint work with my collegue khushboo Agarwal at IIT Bombay.

traveling-salesman-problem

Traveling Salesman Problem , often abbreviated as TSP. The integer programing problem formulaiton is implemented to solve TSP. Integer programming formulation is well known MTZ method.

walkins-scheduling

This is part of my course project called "IEOR for health care". A new algorithm for scheduling the walkin patients in a hospital is implemented. Various preformance evaluation criterias are also applied.