See All Categories

Learn More

현재 위치

  1. Home
  2. Product

What is the deep learning technology?

Deep learning is an artificial intelligence (AI) algorithm that identifies visual and voice signals using a special artificial neural network. Until recently, it was difficult for a computer to detect an object in an ordinary video. It could not tell the difference between a dog and cat. However, with the deep learning technology, it is now possible. In fact, deep learning is bringing a breakthrough not only in video recognition but also in almost all AI areas such as voice and language recognition.

Deep learning convolutional artificial neural network

In the deep learning technology, the recognition algorithm and domain knowledge are separated. The algorithm learns from a given set of data to identify any object (be it detecting dogs and cats or distinguishing cars from bicycles). One can say that the deep learning algorithm functions like a human brain, learning knowledge on various fields through data. (That is why a smart and well-trained brain is required for good performance). Then, how can you tell how smart a deep learning system is? Deep learning systems compete against each other in the ‘Imagenet Challenge,’ an annual artificial intelligence test held in the U.S. Famous deep learning algorithms such as AlexNet and Google Net won the contest. Many Korean teams have also participated in the contest every year and reaped good results. The smartest algorithm these days is ‘ResNet.’ Its intelligence is similar to that of a human. In the past, knowledge and algorithms were mingled together so different algorithms were required to identify different objects. However, with the deep learning technology, an algorithm with a high intelligence shows high performance across all areas. What is important is to first select a smart brain and train it well.

Separation of an algorithm and domain knowledge

I want to develop an object recognition service using
the deep learning technology. What do I need to do?

Deep learning goes through a process of training and continuous inference.

A high-performing GPU is used to train a deep learning system. Install a machine learning software program such as Caffe and TensorFlow, select an algorithm and train the system with images related to an object you want the system to recognize. Normally, a computer with a GPU is bulkier and consumes hundreds of more watt than a normal computer, but this is not a big problem because the training is conducted prior to providing your service to customers.

For object detection, you can use a computer with a GPU. A deep learning software program is also required. What is more, you need to develop an additional software environment connected with the existing service system.

This is where the problem lies. Numerous devices need to be deployed in different sites and communicate in real-time (think of using the technology to find pests like spotted lanternfly across the country). It would cost a lot to build such system, but even worse, it would be difficult to deploy it due to its high power consumption. Furthermore, you would need a commercial license to use a deep learning software program.

The Deep Runner offers an alternative. It is the size of a hand, consumes little electricity, and still supports the latest deep learning algorithms including ResNet at a fast speed to recognize an object. The Deep Leaner does not train the algorithms, but you can use a computer with GPU, like what you have done so far, or use our training software to training them through simple steps without profound understanding in deep learning.