Perforated Information

Perforated makes AI models up to 10x cheaper to train and run. We upgraded the standard artificial neuron so models can reach the same accuracy at a fraction of the cost, within existing ML workflows. Our patented, neuroscience-inspired technology improves how individual neurons learn during training, enabling smaller, more data-efficient models without changing the deployment stack. Our published results across vision, language, and simulation models show up to 90% lower cost, 45% less training data, and 68% of remaining error eliminated. Perforated works with companies spanning gaming, engineering simulation, marketplaces, satellite analytics, and other compute-intensive applications to enable smarter, more efficient models. Our team brings together PhDs in neuroscience, computer science, and machine learning with experience commercializing technology at scale. Perforated was built out of the Carnegie Mellon ecosystem and is a member of the PyTorch Ecosystem.