Blog posts

• Bookender: A Book Recommendation System
  

  December 06, 2020

  I built a book recommendation system over the past few months as my capstone project for a data science bootcamp, The Data Incubator. If you’d like to read more about my experience there, I’m planning to write about that as well and I’ll include a link here when I do. In this post, I’ll outline the structure of my recommendation system and write about my experience building it.

  ---

• A Generalized Elo System for Tennis Players, Graphics
  

  August 26, 2020

  Here is a graphic of the Big Four (Federer, Nadal, Djokovic, Murray) ratings according to my Elo system.

  ---

• A Generalized Elo System for Tennis Players, part 2
  

  August 13, 2020

  In this notebook, we will take our previous Elo system for tennis players and add playing surface as a parameter.

  ---

• A Visual Introduction to Interacting Particle Systems
  

  August 13, 2020

  An interacting particle system is a stochastic process in which particles randomly move around as time passes with certain exclusion rules. Here, we explain some problems and ideas in interacting particle systems literature.

  ---

• Testing Robustness of Neural ODEs against Adversarial Attacks
  

  July 22, 2020

  In my last post, I wrote an introduction to Neural ODEs. I made a simple network with one processing layer, one ODE layer, and one fully connected layer. In this post, I will compare the robustness of this model (which we call an “ODENet”) to regular ResNets. We see a statistically significant improvement against adversarial attacks when switching from ResNets to ODENets. However, the difference is quite small. This exploration was informed by the paper On Robustness of Neural Ordinary Differential Equations (Yan et al).

  ---

• A Simple Introduction to Neural ODEs
  

  July 17, 2020

  You might have heard some hype around neural ordinary differential equations (Neural ODEs). The idea of using a continuous process to model a discrete one is very fundamental in many parts of mathematics, so I was excited to learn about this stuff. Even though the code from the original paper is available online, I couldn’t find a simple high-level explanation + implementation of neural ODEs on a simple dataset. In this post, I’ll explain the idea behind and purported advantages of Neural ODEs and create a MNIST classifier using a Neural ODE.

  ---

• A Generalized Elo System for Tennis Players, part 1
  

  July 07, 2020

  Here, we will create an Elo rating system for male tennis players and do some analysis on the choice of parameters. We find that a random choice of parameters actually does quite well, and that a wide variety of K-factor weight functions do a good job predicting the outcomes of tennis matches. In future notebooks, we will further expand upon this model by adding various features.

  ---

• Simulated Annealing and Smitten Ice Cream
  

  June 29, 2020

  I live in Oakland, about a mile away from a Smitten Ice Cream store. Their selling point is their super-fast liquid nitrogen made-to-order ice cream. They claim that the ice cream, which is turned solid from liquid in 90 seconds, is creamier than regular ice cream. The validity of the scientific basis of this claim, I can’t answer, but I can make a simple mathematical model, derived from physical principles, to simulate the comparison between Smitten-made ice cream and regular ice cream.

  ---