Select Page

Artificial Intelligence (AI) is often considered the most revolutionary innovation of the modern age. It truly has the potential to reshape innumerable aspects of daily living in the modern world.

There are several exciting areas emerging in AI research. Kevin Frans spoke at a 2018 TED Talk, where he discussed the most prominent areas of AI research for Quora. Some innovations that he finds most impactful include:

  • Generative Models
  • Image Translation
  • CycleGAN
  • Reinforcement Learning

Generative Models and Images
AI applications for image technology owe much to the convolutional network. Before this breakthrough, even the generation of written digits was challenging. Amazingly, the convolutional network is a very young technology. Clearly, AI innovation is still in its infancy.

The convolutional network made images much easier to manipulate and incorporate into useful applications. The generative adversarial network would help AI take its next step in the practical use of images. Some aspects of image translation were thought to be impossible. Generative models recently proved otherwise. AI technology can now generate near perfect faces with a very high degree of accuracy. This is so baffling that Frans compares this image translation to magic.

In the past, AI translated images by using training pairs. Limited training data has been a very big obstacle in advancing AI. CycleGan succeeded in circumventing this limitation. It can translate images without the need for training pairs. This is another advancement that shocked AI developers. It caused them to look at what is really possible and gives a lot of hope for the future of AI technology.

Reinforcement Learning
The advantages of reinforced learning (RL) may become the most influential of all AI innovations. For this reason, it has been likened to a holy grail for AI technology. Why is RL considered more important than any other AI technology? Simply, RL has the potential to solve problems devoid of supervision and data. The reality is that RL is not currently predictable. Also, when it does work, there are often tricks involved to complete an operation. In recent years, solid progress has been made in RL research. True techniques are improving its application. Scientists are confident that RL innovation will continue to progress steadily.