Hello, I am Bernhard (Ben) Firner. My most notable work was at NVIDIA, where I worked on autonomous driving until the summer of 2023. I joined the AV team at Bell Works in Holmdel, NJ in May of 2015. The goal of our project was to train an end-to-end neural network to drive a car in a human-like manner, thus demonstrating that end-to-end was viable approach to autonomy. NVIDIA's CEO talked about our project at GTC 2016. There is a nice write-up of the project's history, from 2015 to 2021, in an NVIDIA blog post. At that link you can also find a video of a drive from North Carolina to New Jersey. That was one of several long test drives that were required once the distances between driving failures required multi-state distances for proper testing.
Besides my work in autonomous vehicles, I've worked on a few other interesting projects. I completed my undergraduate and masters degrees from Stevens Institute of Technology in Hoboken, NJ, along with minors in Electrical Engineering and Philosophy. After graduating, I worked at ITT avionics for a brief time before returning to school to earn my PhD at WINLAB, a wireless networking laboratory at Rutgers University in New Jersey. While I worked on my thesis, I learned far too many skills to list under the mentorship of Dr. Yanyhong Zhang and Dr. Rich Howard. The final result of my thesis was a low-power wireless protocol and some highly optimized code that allowed small wireless sensors to run for 20 years on a CR2032 coin cell battery (the current consumption is actually below the discharge rate of the batteries, so effectively the lifetime was as good as the battery chemistry would allow). For more on my PhD work, see my research page.
We took that technology and made a small startup company for environmental monitoring systems in the laboratory animal care market. At the same time, I also taught several courses in the Rutgers computer science and computer engineering departments. After that, I joined NVIDIA and stayed there until 2023 where I worked on every part of a machine learning pipeline: data collection, data labelling, data loading, data augmentation, training, simulation, testing, evaluation, and optimizations of all of those steps. My projects page has more information and links to some of our arxiv papers.
My current interests are machine learning, good software engineering practices, and teaching those two topics. Machine learning isn't easy -- the field has moved too quickly for any best practices to take hold, and, even if they had, the diversity of applications seem to require bespoke solutions for each problem. "Seem to" is the key phrase: there are some good first principles that can guide any machine learning project, as long as we are willing to put in the effort. In the long term, quality doesn't cost; it pays. I've written some of the thoughts that have crossed my mind in my web log.
As a friend and mentor likes to say, specialization is for insects. That does sell insects short (Especially social insects; "The Mind of a Bee" by Lars Chittka is an enlightening read, especially if for any machine learning or robotics researcher who would enjoy being humbled.), but it is true to an extent. My hobbies are reading, obviously, but I also enjoy running and playing the Irish fiddle. An active body supports a healthy mind, so you've got to get out there and punch every day in the face.