Benjamin Bolte
New York based roboticist and machine learning engineer, with experience at Tesla AI, FAIR, Google Research and Amazon. Interested in self-supervised learning and fast neural network inference. Check out my blog, Github, Google Scholar, or shoot me an email.
Work History
Tesla Autonomous AI
- Machine learning engineer, March 2021 to May 2022
- Wrote ray casting CUDA kernel for producing ground truth data for a core geometry prediction model
- Got a language model to drive a car
- Designed and trained car-compatible autoregressive model architecture
- Rewrote network compiler stack to support subroutine capture and control flow
Facebook AI
- Seattle, May 2018 to February 2021
- Research engineer, worked on representation learning for speech. Coauthored multiple high-impact papers (available through my Google Scholar page)
- Designed and deployed model to increase accuracy for classifying terrorist posts by 54% over existing baseline, resulting in 500K more terrorist propaganda posts being deleted every day
- Shipped models which doubled previous accuracy for hate speech and tripled daily content actions across major markets
- Deployed BERT hate speech model to content across Facebook, building the core PyTorch NLP library used by our team and additional Hacklang system for logging and request handling, which decreased server usage by 95% over initial approach without reducing model accuracy
Facebook, Care Machine Learning (Intern)
- Seattle, Summer 2017
- Built tools for automatically evaluating and deploying objectionable content classifiers
- Tripled model refresh rate and eliminated several hours of manual work per refresh
Google Research, Handwriting Recognition (Intern)
- Mountain View, Fall 2016
- Improved mixed-script TensorFlow-based online handwriting recognition system for Chinese-English and Devanagari-English language pairs
- Trained a Generative Adversarial Network for producing handwriting using recurrent neural networks with attention components
Amazon, India Invoicing (Intern)
- Seattle, Summer 2016
- Build three APIs for interacting with the existing India invoicing service
- Build a web framework for helping product managers answer customer questions and solve bugs
- Diagnosed and fixed a Sev 2 which affected millions of dollars in transactions
Education
Emory University, BS in Mathematics and Computer Science
- Class of 2017
- 3.9 GPA
- Graduate-level coursework in deep learning, information retrieval, compilers, operating systems, computational neuroscience
- Additional coursework in economics, Chinese language, theater
- ACM ICPC
- 1st Place, Division 2 (2017)
- 2nd Place, Division 2 (2016)
- Dean’s Achievement Scholarship (2014 - 2017) Highest undergraduate merit award
- Computational Neuroscience Training Grant (2014 - 2016) Highly competitive NIH Blueprint Grant for studying computational neuroscience at Emory and Georgia Tech
Additional
- Personal Github projects and gists have accumulated over 1000 stars
- TSA PreCheck (2018 - Present) Highest civilian honor awarded by the United States Transportation Security Administration