Case Studies

Diagram showing the term 'Artificial Neuron' in the center, with lines extending outward indicating connections or components.
Diagram showing how a new artificial neuron connects to multiple dendrites.

As an optimizer, Perforated AI doesn’t build networks, we make yours better. This page is dedicated to the experiences of our users and what Perforated AI has been able to do for them.

Our latest customer story comes from a user who improved the predictive power and model size of BERT models across a wide spread architecture settings and two datasets. BERT, a model released by Google, is often considered one of the original LLMs.

Line graph titled 'BERT Deep Summing Network (DSN) - IMDB Dataset' showing accuracy on the y-axis and parameter count in millions on the x-axis. Different colored lines represent various width settings for PB + DSN and baseline models, with accuracy roughly between 0.74 and 0.88.
“Our experiments with Perforated AI’s Perforated Backpropagation™ technology showed consistent improvements across all model architectures we tested.
- - -
What impressed us most was how quickly we were able to implement and scale the technology. In just one week of experimentation, we had it working effectively”
— Evan Davis, CTO, Skim AI Technologies
Bar chart comparing accuracy and parameters for different models: Original AMP-BERT, Compact AMP-BERT, and Compact AMP-BERT with Dendrites. Accuracy is shown with gray and orange bars on the left y-axis, while parameters in millions are shown with a red line on the right y-axis.
“This kind of result was unimaginable with traditional methods!”
— Jingyao Chen, MSCB Student, Carnegie Mellon University
Bar and line graph showing the impact of different MobileNet model configurations on test accuracy and parameters. The x-axis lists three MobileNet models, and the y-axes show test accuracy on the left and parameters in millions on the right.
“For anyone building on a budget, or targeting mobile and embedded devices, this is a direction worth exploring”
— Rushi Chaudhari, Data Engineer, Deloitte
Bar chart titled 'Supply Chain Prediction' comparing average RMSE for 'Original' and 'PAI'; Original has a higher value, around 28, while PAI is about 17.
“In just a week we were able to improve ClaudIA. With Perforated AI now ClaudIA is giving better results achieving outstanding forecasting for demand planners in the supply chain industry.”
— Andres Enrique Marín Zambrano, Researcher at UTSA