in

DeepMind's artificial intelligence can apply learned knowledge to complete new tasks

DeepMind's artificial intelligence can apply learned knowledge to complete new tasks

Google's roofing company Alphabet's artificial intelligence platform deepmindcan learn to generalize beyond his immediate experiences. In a study conducted jointly with Stanford and London University College, DeepMind scientists investigated whether systems can apply the knowledge they have learned in one task to other tasks. Teams conducting the study reported that DeepMind's artificial intelligence correctly uses the nature of a language to interpret instructions never seen before.

Scientists say that although these systems of artificial intelligence trained in ideal or reduced situations fail to exhibit a compositional or systematic understanding of their experience, this competence can have access to many examples of versatile observations can easily emerge. The team found that the ability to entertain a thought implies the ability to entertain thoughts with semantically related content; and to what extent they can give a cognitive concept, a systematic artificial intelligence model. For example; when it comes to systematics “John loves MaryBir a person who understands the sentence “Mary loves John.“is likely to think.

Given the two objects that are randomly positioned and trained using rewards to strengthen the desired behaviors, the intermediaries can be used to apply to objects that they have not seen before.lifting, upgrading“In addition, in the next task requiring them to place objects on beds or trays according to the instructions, 90 percent placement accuracy is provided.

In a separate test, the team explored tasks that could be solved, with or without language. They positioned the agent in a virtual grid of eight randomly positioned objects. Object types were correctly identified and the agents received a reward for collecting such objects.

As a result, researchers, proved to be critical in all tests. The number of input and input provided by the number of words and objects experienced during the training, the first-person perspective and the agent's perspective over time.

Leave a Reply

Your email address will not be published. Required fields are marked *

CS: Global Offensive: Valve testet neues Schlupfloch für Beuteboxen

CS: Global offensive: Valve tests new loophole for loot boxes

Nintendo Switch: EA verkauft FIFA 19 als FIFA 20

Nintendo Switch: EA sells FIFA 19 as FIFA 20