And The Winners Are...

Kickoff

In October of 2025, a few dozen machine learning nerds, a few venture capitalists, some very generous team members from Weights & Biases, and the Perforated AI team gathered on the top floor of Frontier Tower in San Francisco during Open Source Week to kick off our second-ever hackathon during the PyTorch Conference. In the ensuing months, the scale of the event would explode, and by judgment day, we would see some of the best results we've ever seen from teams attacking a wide range of problem spaces and using our technology in exciting and novel ways. Here's a video of the results announcement. If you'd prefer to learn about it in blog form, read on.

The Premise

Our hackathons are set up a little differently than your average, run-of-the-mill one-weekend competition. ML engineers have free access to our patented dendritic optimization technology, Perforated Backpropagation^TM. They can then apply the tech to any model they like, optimizing for model compression, enhanced accuracy, or both. Participants cover a wide range of applications for their dendritic models, from medical imaging to robotics to exoplanet detection, or even to winning another hackathon.

Next, the participants implement the tech into their training pipeline (this part is very quick; many of the IRL participants successfully implemented it at the 3-hour kickoff event) and then spend the next couple of months training, fine-tuning, and generating results to be presented to the judges. This time around, the judging took place in mid-January 2026. Several thousand dollars were on the line, and a couple of one-year Pro Memberships to Weights&Biases were set to be awarded to the top projects. Along with traditional first, second, and third place prizes, five honorable mentions each won $250, and we also awarded $500 for "Hackathon Inception", an award for using our tech to win another hackathon!

When the dust settled, this hackathon had seen 693 participants, 71 submissions, 10 novel use cases, and 1 hackathon inception winner.

The Winners

Some hackathon blogs would save the best for last, but I can't wait, so here's the first-place submission:

First Place - $3,000

Team Neuron AI

Team Neuron AI used dendrites and applied YOLO to the Pascal VOC dataset. This ubiquitous object detection framework is used in edge devices around the world, but has been a particularly hard nut to crack for dendritic optimization. Successfully applying dendrites to YOLO had, so far, eluded previous hackathon participants and even Dr. Rorry Brenner (CEO and inventor of dendritic optimization). The team solved this problem and even achieved a reduced error rate of 5.2% on the project. This valiant effort resulted in a well-earned top prize of $3,000 and a one-year Weights&Biases Pro Membership.

Third Place - $1,000

Nicholas Mesa-Cucalon

The bronze medal went to Nicholas Mesa-Cucalon for Quadruped Robot Control. Using the DeepMind Control Suite, exercising continuous control with simulated physics, they were able to apply dendrites to a TD3 Reinforcement learning algorithm and showed a 13% improvement, rivaling the performance of other, much more complicated TD3 architectures. Nico was rewarded with $1,000 for the project.

Traditional neural network

Dendritic neural network

The Honorable Mentions

Five projects won a $250 honorable mention prize. Here they are!

Hackathon Inception

While the hackathon's specific goal was for participants to show the best application of dendritic optimization, the goal of dendritic optimization itself is to make neural networks smarter, smaller, and just better. This means that a well-implemented dendritic neural net should be able to outperform traditional networks in the real world, enabling participants to win other hackathons. This is the idea behind our $500 Hackathon Inception prize.

These Hackers Did Cool Stuff With Dendrites, And You Can Too!

Innovation is at the core of Perforated AI, and running hackathons like this is such a cool way to both get our tech into the hands of brilliant engineers and to show even more results that verify how powerful dendritic optimization really is. If you're a curious ML engineer, check us out on GitHub, and if you're interested in applying dendritic optimization to your model to make it smarter, smaller, and just better, get in touch!