Deep Vision, a leading company in machine learning, has been granted a patent (US11734006B2) for their development of a computer part that can essentially learn on its own. This system functions by holding, organizing, and changing image information simultaneously and at a blistering speed. It uses special built-in tools that aid in quicker calculations and data storage, resulting in improved machine understanding of images.
Currently, the foremost challenge in the widespread application of this technology in embedded devices is the steep computation cost of deep learning algorithms. While Graphics Processing Units (GPUs) are primarily used to implement deep learning solutions, they consume too much power for battery-operated embedded devices. Moreover, the cost of GPUs can be too steep for many industries.
Deep Vision's patent addresses this problem by introducing a fully programmable processor that can perform traditional image analysis approaches such as feature extraction, edge detection, filtering, or optical flow. This solution circumvents the risk of the chip becoming obsolete within a year due to the rapid pace of progress in the field of deep learning.
By reducing the power consumption and lowering the cost, this invention facilitates the more extensive application of machine learning technology. It is estimated that these processors can boost the efficiency of deep learning computation by 50 times as compared to GPUs, providing a more cost-effective solution.
In the foreseeable future, companies in sectors like automotive safety, security cameras, and drone systems will be the main beneficiaries of such innovation. For instance, car manufacturers can integrate these processors into vehicles to support advanced features such as driver alertness monitoring or collision avoidance systems. Similarly, in the security camera segment, manufacturers can leverage this technology to efficiently analyze and respond to visual data.
However, it's worth noting that although the patent has been granted, there is no certainty that this product will reach the market. Like any invention, it has to pass through numerous stages of production, testing, and modifications before becoming commercially available.
P.S. This news reflects a patent application and there is no surety whether this solution will ever hit the market or not. However, should it become available, it can revolutionize machine learning applications across different sectors.