Frequently Asked Questions

Perforated is a Machine-Learning company that has created a technology to massively unlock performance for AI/ML models. Based on a recent neuroscience breakthrough, we act as the data-efficiency layer that helps teams build more accurate, parameter-efficient models. For enterprise users, this often means getting the same model performance with half the amount of training data, or recovering half of the remaining error of pruned, parameter-efficient models to achieve much higher levels of accuracy. Ultimately, this translates to saved costs, accelerated timelines, and more value from AI models.

Currently, Machine Learning engineers are inundated with overwhelming manual tuning chores, and they face critical gaps in available tooling when it comes to data efficiency and accuracy boosting. We solve this problem by providing a single, fundamentally new approach that fits within existing PyTorch workflows. We help users do more with less, empowering ML engineers to achieve better results without compromising across model size and accuracy.

The product's usability and interface are designed for machine learning engineers, but the ultimate outcomes—efficiency, cost savings, and performance—are built for business leaders.

The primary use cases are situations where parameter efficiency and model performance are critical and training data is hard to come by. This often shows up in edge applications or when engineers are dealing with constraints associated with hardware, compute cost or dataset sizes. These are often the highest-stakes models, where we see hard cutoffs in model performance requirements for regulatory approvals or safety (such as FDA approvals or autonomous vehicle safety).

It is actually both. Whether you are training a brand-new model from scratch or fine-tuning an existing model to improve its performance, our technology seamlessly plugs into your training and post-training process.

The benefits accrue heavily across both phases. During training, you benefit by requiring significantly less data and achieving a faster overall development cycle. During inference, you benefit from radically lower compute requirements if deploying on the cloud, or hardware efficiency and improvements in latency and memory if deploying on the edge.

Perforated comes in strictly during the training or fine-tuning phase. It modifies the computational structure of your network during training and leverages a set of unique learning rules to optimize the learning process itself. It then exports to any PyTorch processing step seamlessly, with no changes to existing processes.

It is something you use during training. Some teams train a model once, deploy it, and never have to fine-tune it again. Others periodically or consistently fine-tune their models as new data becomes available. Because models are constantly being minted and updated, Perforated is used whenever a model undergoes training.

Yes. We have commercial customers and are actively executing proofs of concept with partners spanning multiple use cases. Check out the case studies on our website to learn more about the results our customers and partners have achieved with Perforated.

Teams using the Perforated Suite typically see a 30% reduction in training data requirements, 20% error reduction, and 60% parameter reduction. From a business perspective, this means they are drastically lowering deployment and compute costs without sacrificing performance.

There are a number of other machine learning techniques out there (such as pruning, distillation, quantization, and Parameter-Efficient Fine-Tuning (PEFT) methods like LoRA) that engineers might leverage to optimize their models. Our technology is fully compatible with all of these techniques. In fact, we see the best results when Perforated is used in combination with those techniques. Perforated is the lightweight new layer that acts as a force multiplier, not a replacement, for existing techniques.

We identified an information arbitrage opportunity between neuroscience and computer science. While traditional AI has relied on an oversimplified artificial neuron, neuroscience recently had a breakthrough regarding the computational power of biological dendrites. We pulled that innovation into computer science, creating a new type of reinforcement learning enabled at the individual neuron level that makes artificial neural networks vastly more parameter-efficient.

Yes. Perforated was just accepted into the PyTorch Ecosystem Landscape, which is a major validation of our technology's maturity. Our solution can be integrated directly into training pipelines via a PyTorch extension using just 3 to 5 blocks of code, requiring no changes to your core architecture.

Perforated Backpropagation™ empowers the artificial neurons of deep neural networks to achieve better performance coding for the same features they coded for in the original architecture. After an initial network training phase, additional "Dendrite Nodes" are added to the network and separately trained with a different objective: to correlate their output with the remaining error of the original neurons. The trained Dendrite Nodes are then frozen, and the original neurons are further trained, now taking into account the additional error signals provided by the Dendrite Nodes. The cycle of training the original neurons and then adding and training Dendrite Nodes can be repeated several times until satisfactory performance is achieved.

We offer an open-source version (Apache 2.0) that features our core dendritic architecture trained with standard backpropagation for research and evaluation. We monetize through our commercial version across three customer tiers and enterprise partnerships. The commercial version includes our patented Perforated Backpropagation™ algorithm for maximum efficiency gains, as well as Perforated Studio, our GUI that supports parameter tuning, experimentation, and configuring runs.

We are a venture-backed AI infrastructure company currently navigating our capital strategy to rapidly scale our commercial and enterprise footprint.

We are proudly headquartered in Pittsburgh, Pennsylvania, because Pittsburgh quietly builds serious things. It understands hard problems and built steel, robotics, autonomous vehicles, and world-class AI before the rest of the market realized it mattered. Furthermore, Carnegie Mellon University produces people who care more about being right than being loud, making it the perfect home for our technology. We also have an office in San Francisco to maintain a presence in the Bay Area.

The team consists of experts spanning applied neuroscience, ML systems architecture, and enterprise deployment. Perforated is led by our CEO and neuroscientist Dr. Rorry Brenner, who pioneered the dendritic approach to next-generation AI infrastructure, alongside President Erin Yanacek and CTO Dr. Dean Alderucci.