00:00A team of researchers at Uber AI Labs in San Francisco has developed a set of reinforcement
00:10learning algorithms that outperform both human players and other AI systems in classic video
00:15games.
00:18These algorithms excel by learning from extensive data sets, recognizing patterns, and making
00:23educated guesses about new data.
00:26Traditionally, such algorithms struggle when faced with unfamiliar data, but this new development
00:32addresses that issue.
00:34The researchers introduced an innovative algorithm that effectively tracks all the paths previously
00:39taken by another algorithm while solving problems.
00:43When confronted with an anomalous data point, it refers back to its memory map to explore
00:48alternative routes.
00:51This method was tested by embedding rules from various video games with the objective
00:55of maximizing points and achieving higher scores over time.
01:00Their system was put to the test on 55 Atari games, where it outperformed other AI systems
01:0685.5% of the time.
01:10It particularly shone in Montezuma's Revenge, surpassing any previous AI performance, and
01:16even breaking a human record.
01:19The researchers are optimistic that their algorithm's success can extend beyond gaming,
01:24potentially revolutionizing applications in image and language processing by robots.
Comentarios