Ari S
- Research Program Mentor
MS at University of Pennsylvania (UPenn)
Expertise
Machine Learning, Simulations, Statistical Betting Methods, Bayesian Analysis, Probability, Stochastic Processes, Algorithms, Math
Bio
Hi! My name is Ari, and I have just completed a Masters Degree in Applied Math and Computational Science at the University of Pennsylvania. What drives me is a desire to solve difficult technical problems in math, computer science, statistics, and engineering, and I make it my business to be familiar with all four of the above fields. I have done work in machine Learning, simulations, physics and graph Theory, and bayesian analysis. I am particularly interested in bridging the distance between mathematics and statistical theory with real world problems, in the hopes that said theory motivates novel solutions. I currently live in Philadelphia Pennsylvania, and am working as a researcher at Wharton. As an undergraduate I was a rugby player (and captain of my undergrad team). I currently run medium distances, hike, and play quite a bit of chess when I can find the time. I love participating with students and helping them reach a-ha moments, while at the same time helping them develop into creative, methodical thinkers.Project ideas
A bayesian basketball win prediction system
Bayes rule is crucial to modern statistics (as well as data science, machine learning). Using some results from Bayesian statistics one can model the distribution of basketball performance stat distributions. Using a bayesian model to predict the probability distribution of these statistics we can attempt to predict a teams win and loss rate versus another team by drawing samples from these distributions and computing correlation to win or loss. Project can be as simple or as complicated as the student would like, simple normal models, mixture models, Gibbs sampling, and hidden Markov models can all be featured, based on students interest and comfort. Student will spend some time 1) learning about bayes rule, and learning about some famous probability distributions 2) learning how to code a fairly simple simulation in R or python (choice is yours) 3)Learning how to interpret the significance of statistical results and adjust results over time based on success/failure of model over time.
The genetic algorithm
The genetic algorithm is a stochastic (randomized) algorithm which is not often used in computer science, but has sparked the curiosity of researchers in the past 40 years. The algorithm itself is both intuitive and appealing (one should look this up to see more), but has some natural drawbacks. Often randomized algorithms are used to provide heuristic solutions to NP-complete problems in computer science (i.e. the traveling salesmen problem). Using a genetic algorithm in such a problem would provide interesting results, and some insight into the strengths and advantages of this mysterious tool. We would begin with an overview of stochastic annealing, and some popular randomized algorithms. This would be followed with some search and review of the genetic algorithm, followed by some novel implementation in a general NP-complete CS problem.