Cofounder and Chief Data Scientist at Feature Labs.

I’m into Zen Buddhism, urban farming, data science & machine learning, pole vaulting, mobility/health/fitness, music, skiing, and generally making a fool of myself through art.

Work

Feature Labs is a venture-backed startup enabling enterprises to deploy predictive analytics solutions in a smart, fast, and disciplined way. We develop automation and UI tools so that business executives, subject-matter experts, data scientists, and software engineers can focus on solving business problems instead of computation and academic problems.

Spotlight

Should building AI products require you to “call in the cavalry” – No Read how to engineer data-driven AI products – “Machine Learning 2.0”

Delivering on the promise – first AI product built using our automated data science tools (collaboration with Accenture) MIT News | “The AI Project Manager” paper

Feature Engineering vs. Feature Selection

Applying Data Science Automation to Better Predict Credit Card Fraud

Projects we’ve done that we can talk about

  • Featuretools - the most popular open-source automated feature engineering framework on Github. Backed by DARPA in the D3M program
  • The AI Project Manager - a collaboration with Accenture that used ML 2.0 techniques that produced an augmentation tool for software project managers to accurately predict whether projects meet their goals.
  • Helping BBVA predict credit card fraud - automated feature engineering and problem-specific optimizations drastically reduce financial implications of fraud.

Past Research

My research at MIT focused on data science automation, and specifically, on automatically generated and solving meaningful prediction problems given arbitrary relational data. This work culminated in my master’s thesis as well as a paper in IEEE/ACM DSAA 2016 titled What would a data scientist ask? Automatically formulating and solving prediction problems.

I also helped develop an end-to-end data science foundry for MOOCs that enabled a systematic way of going from raw data to predictive models. This work was published in IEEE/ACM DSAA 2015 (link).

More fun projects include an implementation of near-optimal online matrix-completion algorithms (which had previously only existed in math form), multiplexed robotic scheduling for liquid-handling robots, recipe modification using recurrent neural networks, and scene detection using a combination of convolutional and recurrent neural networks.

Current Projects and Experiments

Pole vaulting ever higher

As always, I am trying to jump a little bit higher every day.

My first foray into urban farming

Everyone should be able to grow their own food. I’m learning how to do this myself, by building a Personal Food Computer and attempting to automate as much of the growing process as possible. My first crop is Purple Bok Choy!

Past Projects and Experiments

One day I will actually write descriptions for all of these!

Lil Neuron

Human-generated rap is so 20th century. Who’s to say Lil Neuron can’t be the illest rapper alive without being “alive”?

Lil Neuron is an AI rapper using recurrent neural networks. See Lil Neuron Takes His First Breaths for my first blog about it, and the source on Github. Please let me know if you want to contribute!

Predicting Body and Mental States

Attempting to predict how tired I’ll feel, how much energy I’ll have, how sick I’ll be, and the locations and times I do things using both passive phone and watch data as well as active recording of the food I eat both as it enters and exits my body and the way I feel throughout the day.

Online Matrix Prediction

RoboChef

Scenic Recursion

Learning Musical Style

Ben Schreck Shits Calendars

Gesture-Controlled Drone

Hacking Hackathons

Posts

subscribe via RSS