Intel, a semiconductor giant, is working on a new software platform for artificial intelligence model training for computer vision. The launch is part of the company's business strategy to boost chip sales.
Simplifying AI model training for computer vision
Sonoma Creek is a new piece of software Intel is working on. According to The Register, citing internal company documents, it is expected to be available as a subscription or license starting in fall 2021. Data collection, labeling, model selection, training and optimization, and deployment of AI models would be dispersed across multiple departments in Sonoma Creek's unique workflow, The Register reports. The software will operate as follows:- First, Sonoma Creek will begin training computer vision models with just 10 to 20 photos or a video, using PyTorch and TensorFlow.
- Second, users will be able to use intelligent assistants to annotate the data, which will help Sonoma Creek learn and improve the accuracy of the model.
- Third, the software will automatically train the model and provide predictions once enough data has been annotated. Users will be able to approve or modify these to improve the accuracy of the model.
- Fourth, users will be able to either use the original model framework for inference or use Intel's OpenVINO toolkit, which optimizes the model for a variety of Intel hardware, including CPUs and GPUs.
- Finally, if a model is inaccurate, users will be able to update it by adding new photos and identifying them before re-exporting the model. "Activelearning" is the term for this process of constantly improving the model.