AI will replace human jobs

The goal of AI is not to replace humans, but to work with them to improve the efficiency and quality of their work. AI can automate tedious tasks, freeing employees to focus on more creative and strategic work. This is an efficient and useful tool that can also help solve the problem of labor shortages.

As a result, the technology is gradually being used more widely in the manufacturing and logistics industries to address persistent labor shortages and other long-standing problems. The combination of AI and robots can achieve tasks such as object avoidance and ground mapping, so as to complete the distribution of goods in various facilities. The combination of AI and machine vision systems can take on essential repetitive quality assurance tasks, including missing detection and inspection of components.

D136-001-007

Industrial AI requires thousands of images and large data sets

Some people mistakenly believe that applying AI in industry requires huge data sets and thousands of images. In reality, there are many different types of AI technology, and some of these applications do require large data sets, but not all cases require such a large amount of data. For some applications, it is also possible to make effective predictions and decisions using limited data sets and experience.

The deep learning and edge learning technologies introduced by Cognex are representative of the above two situations:

Deep learning is known for its excellent ability to handle complex tasks. This technique is ideal for working with large image sets that contain a lot of detail and change significantly, but is also ideal for complex or highly customized applications. Because these applications involve so many detailed changes, a lot of image training and model execution is required upfront to automate complex tasks.

Edge learning is designed for ease of use. It uses a set of pre-trained algorithms to process on the device or at the “edge” of the data source. By embedding the knowledge of application requirements into the neural network connection in advance, the training method eliminates a lot of compute load, so the training can be deployed in minutes using only 5 to 10 images, without the need for Gpus, so that the application can be quickly scaled and easily adapted to change.