Nano Dimension Announces Another Deep Learning AI Patent Granted to its DeepCube Technology

The Patent Relates to Model Training and Optimization on a Decentralized Networks

The Milestone Represents a Critical Innovation to Enable the Company’s Commercialization of Its Industrial AI Solution

Waltham, Mass., March 19, 2024 (GLOBE NEWSWIRE) --  Nano Dimension Ltd. (Nasdaq: NNDM) (“Nano Dimension” or the “Company”), a leading supplier of Additively Manufactured Electronics (“AME”) and multi-dimensional polymer, metal & ceramic Additive Manufacturing (“AM”) 3D printing solutions, today announced that a patent was granted for technology developed by its industrial artificial intelligence (“AI”) group, DeepCube, that enables better training and optimization of decentralized deep learning-based AI models.

The U.S. patent, formally titled System and method for mimicking a neural network without access to the original training dataset or the target model, (the “Neural Network Mimicking Patent”) addresses one of the core challenges of deploying AI models in the real-world, specifically continuously training models on new data when that data belongs to a customer. In the industry at-large, dealing with new customer data has often been a limitation due to sensitivity and confidentiality concerns that limit data shareability. The new patent addresses this challenge by ultimately training and improving the AI models on customers’ premises, without Nano Dimension having direct access to the new data or model.

This patent is another key component in Nano Dimension’s efforts to transform DeepCube from a solely in-house AI group to a leading industrial AI solution provider. Nano Dimension has already made progress in this initiative, having announced agreements and memorandums of understanding (“MOUs”) with several parties. DeepCube is currently developing an end-to-end AI platform for industrial usage that is not only limited to additive manufacturing. The software platform is intended to run autonomously on customers’ premises, and continuously improve itself, such that the more it is used, the higher the accuracy will become.

Importantly, the Neural Network Mimicking Patent is not just about software, but also hardware. The training infrastructure at the core of this innovation exclusively uses Nvidia graphics processing units (“GPUs”), while the deployed inference models are currently optimized for Nvidia GPUs, along with Intel and AMD central processing units (“CPUs”).