DeepCubeA, a deep reinforcement learning algorithm created by UC Irvine’s research team, is said to be able to solve magic cubes in an instant without human help.
Magic Cube So, the Rubik’s Cube is a hexahedron made of 9 panels in 6 colors. Once you get used to it, you can match all the colors in about 50 times. Pierre Rudy, UC Irvine, attacked Magic Cube with a deep reinforcement learning algorithm called Deep Cube A. As a result of self-learning over two days without giving a solution in advance or helping humans, it is said that 100% could be solved, and 60.3% of these cases succeeded in obtaining the shortest solution. The time it took to unwind was also a moment.
According to the research team, it seems that Deep Cube A’s solution strategy is different from the human solution and the original reasoning method is different. As the ultimate goal of the project is to build a next-generation AI system, the research team explains that this is the first step toward a big goal.
This case can be said to be the point of reaching a solution through self-learning. Already, solving the magic cube using a computer quickly reached a considerable speed, and last year, an MIT engineering student succeeded in attacking in 0.38 seconds. Related information can be found here .