Announcing the Self-Learning Design Methodology in CES 2023

Being an AI SW/HW design service house, eNeural Technologies, Inc. is striving for delivering the best quality of embedded AI models one has to offer. To do so we developed an in-house toolchain to automate the AI process flow from labeling, modeling, training, augmentation, pruning, and quantization.

This automated process allows us to produce quality and lightweight inference models running on AI SoCs with 8-bit or smaller integer Neural Processing Units (NPU.) This process is now enhanced with a Self-Learning Design Methodology. It first uses a small number of labeled data to train a baseline target inference model. The toolchain then feeds the model with unlabeled data and quickly converges into a highly accurate one. We have applied this methodology to several user applications and obtained more accurate models in 6 times faster time-to-market.