Earlier, the machine learning programs have cracked the secret of surviving and playing games like Mortal Combat or chess. But, some of the computer scientists have taken it to other levels by creating a program which allows the machine to solve Rubik’s cube, which is a shocking change.
The researcher in the abstract of their report published on Arvix said, “Our algorithm is capable of solving the Rubik’s cube for all the possible randomly scrambled cubes in just 30 moves, which is at par with the solvers who utilizes human domain knowledge.
DeepCube, the algorithm designed to solve Rubik’s cube, utilizes the method which is known as Autodidactic iteration, which is a type of machine learning developed by the researchers. The big difficulty while designing the algorithm was to allow the machine to look out for its own rewards at the time of solving the puzzle, a goal they can accomplish.
However, it is difficult to achieve, with earlier methods of solving the Rubik’s Cube, which is based on the principle of scrambling the cubes more, as machines are designed to find out the right way and do not utilize any reverse psychology.
Basically, DeepCube begins with the concept of a completed cube and then more backward to find the ideal solutions. The researchers said in the paper, every single iteration as the input to the neural network was developed by beginning from the goal state and taking random actions.
Machine learning algorithms are capable of learning the things which their programmers want them to learn. Games are among the ideal beginning point for any new technology especially in case of machine learning because of the clear concepts and rules of getting success or failure. So, it is never too late to learn to solve the puzzle, in case you are envious that a machine can do this and you cannot.