Lou Schlessinger

Machine Learning Engineer & Researcher

About

I'm Lou, a machine learning engineer and researcher interested in mathematical reasoning, spatial cognition, and neuro-inspired AI.

I led the runner-up team in the 2023 Vesuvius Challenge, presented research at the NeurIPS Meta-Learning workshop, and have built large-scale multimodal ML systems in production.

Selected work

Two starting points in my work on computer vision and machine learning.

Vesuvius: ink detection

I led the runner-up team in the 2023 Vesuvius Challenge, detecting ink in micro-CT scans of carbonized papyri.

Explore the submission

Automated model search

Combining Bayesian optimization and genetic programming to search for machine learning models. Presented at the NeurIPS 2019 Meta-Learning workshop.

Read the paper

Research

Computer vision, geometry, and automated model search.

Comparison of ink labels and four patch aggregation methods

Vesuvius: patch aggregation

A report on patch aggregation methods in ink detection for the Vesuvius Challenge. Awarded August, 2024 Progress Prize.

View on GitHub

Vesuvius: ink detection

A method to detect ink in carbonized micro-CT scans of the Herculaneum Papyri. Awarded runner-up in the 2023 Grand Prize.

View on GitHub

Vesuvius: scroll geometry

A topological and geometric analysis comparing and contrasting scrolls and fragments from the Vesuvius Challenge to better understand domain differences.

View on GitHub

Automated model search

Presented at the 3rd Workshop on Meta-Learning at NeurIPS 2019, Vancouver, Canada.

View paper

Evolutionary kernel search

Master's Project, Washington University in St. Louis, May 2019.

View paper

Projects

Applications, visualizations, and experiments across my interests.

Occlusion and video prediction

Simple experiments for creating occlusions in PyBullet and testing video prediction models like PredRNN.

View on GitHub

Math problem classification

A model to classify the category of a math problem. A fine-tuned version of bert-base-uncased on a subset of the MATH dataset.

View on Hugging Face

USATT Rating Analyzer

A Gradio web app that helps analyze USA Table Tennis (USATT) tournament and league results by providing basic statistics and performance visualizations over time.

View Project

TraffickCam for Android

An app to help combat sex trafficking by allowing users to upload photos of hotel rooms they stay in. Developed for the Media and Machines Lab at Washington University in St. Louis.

View on Google Play Store

SnackHack

A web service and Android app that leverages Twitter's Geolocation API and WashU network data to provide real-time information on restaurant crowd levels on campus.

View Project

Chocolate ratings visualization

A D3 visualization of artisanal chocolate companies and ratings using expert review data from the Manhattan Chocolate Society.

View Project

Reddit community detection

A survey of community detection algorithms on Reddit data, based on the paper 'Detection and Analysis of Subreddit Communities'.

View on GitHub

Perceptron illustration: weighted inputs combined into an output

Algorithm implementations

A collection of algorithms implemented in various languages, covering topics from machine learning to multi-agent systems, including AdaBoost, neural networks, and the auction algorithm.

View on GitHub

Model accuracy comparison from the stock prediction report

Bayesian stock price prediction

An evaluation of stock price prediction algorithms using analyst ratings.

View on GitHub

U.S. election cartogram

A D3 cartogram of U.S. presidential elections ranging from 1940 to 2016.

View Project

Income prediction

An evaluation of several machine learning methods applied to the Adult Data Set to predict income.

View on GitHub

Global terrorism clustering with Spark

A visualization of k-means clustering on terrorist attack locations using PySpark. Dataset contains 170,000+ terrorist attacks worldwide from 1970 to 2016.

View on GitHub

Procedural terrain generation

A WebGL rendering implementation of a terrain generator based on a 3D-projected terrain map. Uses the Diamond Square Algorithm to procedurally generate terrain.

View Project

Image quilting

An implementation of 'Image Quilting for Texture Synthesis and Transfer' (Elfros & Freeman, 2001).

View on GitHub

Lou Schlessinger

About

Selected work

Vesuvius: ink detection

Automated model search

Research

Vesuvius: patch aggregation

Vesuvius: ink detection

Vesuvius: scroll geometry

Automated model search

Evolutionary kernel search

Projects

Occlusion and video prediction

Math problem classification

USATT Rating Analyzer

TraffickCam for Android

SnackHack

Chocolate ratings visualization

Reddit community detection

Algorithm implementations

Bayesian stock price prediction

U.S. election cartogram

Income prediction

Global terrorism clustering with Spark

Procedural terrain generation

Image quilting

Contact