The third neural network was a relatively simple and processed language using the vectorized expressions of those “Red Right” sentences. Finally, the fourth neural network functioned as an associative layer and predicted the previous three outputs in all time steps. “When we take action, we don’t need to make it a word, but at some point there is this language in our hearts,” says Vijayaraghavan. The AI built by him and his team intended to connect language, positive reception, action plan, and vision seamlessly.
When the robot’s brain was running, they began to teach some combinations of commands and movement sequences. But they didn’t tell them all.
Birth of composition
In 2016, Blenden Lake, a professor of psychology and data science, was issued. paper His team has a series of competitors’ machines that need to master and master to learn and think like humans. One of them was composition. The ability to configure or decompose the whole in the reused part. This reuse allows you to generalize your knowledge to new tasks and situations. “The configuration stage is when children learn to combine words to explain words. [initially] Learn the names of objects and actions, they are single words. When they learn the concept of this composition, they explain Vijayaraghavan, the ability to convey their types of explosions.
The AI built by his team was created for this accurate purpose. This is to make sure that it develops the composition. And that was done.
After the robot learned how a specific command and an action is connected, I learned to generalize that knowledge to execute commands that I have never heard before. Recognize the name of the unmuned action and execute them in a combination of blocks you have never seen. Vijayaraghavan’s AI understood the concept of moving something right or left or putting items on something. You can also combine words that name invisible actions, such as placing blue blocks on a red block.
