AI training often requires thousands of samples to become accurate, and it can be costly and time-consuming, for example, if you want to train a model to detect manufacturing defects you’d need to provide images with both defective samples and good samples.
Japanese AI experts at Hacarus have been working on a solution called Sparse Modeling which requires about 50 samples or even less for training, and worked with Congatec to provides an embedded AI computing kit leveraging the technology.
Sparse Modeling Technology
Hacarus does not go into great detail but explains Sparse Modeling technology is using a data modeling approach that focuses on identifying unique characteristics, in a way that humans recognize friends and family without having to look at everything from feet to head.
That means algorithms based on Sparse Modeling do not need as much data as traditional AI solutions, leading to much smaller AI footprint suitable for fanless low-power systems operating 24/7, and/or that may a limited power consumption margin to integrate AI or machine vision.
They have specific modules for both industrial and medical industries that work either in the cloud, or at the edge in embedded systems or PC running via an SDK, and/or chip IP for Xilinx FPGA and Arm chips, and soon for Intel Altera FPGA and RISC-V chips:
- Industrial
- Cloud, Embedded & Chip IP – Classification, Prediction, Anomaly Detection, Scoring (Analyze, categorize and rank user attributes and historical data), and Super Resolution (Generate high-resolution images from low-resolution images.)
- Embedded & Chip IP – Motion Detection, and Face Recognition
- Cloud & Embedded – Recommendation (Analyze user attributes and historical data and generate recommendation), and Inspection
- Medical
- Cloud only – Vascular Constriction, Treatment Suggestion, Disease Classification
- Cloud, Embedded & Chip IP – ECG
You can see a demo of the company’s SPECTRO Visual Inspection Solution below where they train a model with a few metal plate images without scratches, and then run the model which has been training in a few seconds to one minute against images of good plates and scratched plates.
Some source code, documentation, and samples are available on Github.
Hacarus embedded AI Computing Kit with Sparse Modeling technology
Hacarus can be installed in PCs, or integrated into your own embedded hardware, but to make things easier for customers the company also partnered with Congatec to provide a turnkey embedded AI computing kit.
Key features & specifications:
- SoC – Intel Atom or Celeron Apollo Lake processor
- Storage – 1x mSATA socket
- Connectivity – 2x GbE application ready for “GigE Vison”
- USB – 1x USB3.0/2.0, 4x USB2.0
- Serial – 1 x UART (RS-232)
- Extensions – 2x Mini-PCIe with USIM socket, 16x GPIO
- Power Supply – 9V-32V DC input.
- Dimensions – 173 x 88 x 21.7 mm
There may be a few more details on Congatec’s press release, but I could not find the product page for this new embedded computer kit in either Congatec nor Hacarus website.
Jean-Luc started CNX Software in 2010 as a part-time endeavor, before quitting his job as a software engineering manager, and starting to write daily news, and reviews full time later in 2011.
Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress
