Rubik cube, yes that square object having six faces covered by nine stickers, each of one of six solid colours. How much time will you take to solve this hell complex puzzle ? Might me 15 minutes, 10 minutes or 5 minutes, you might be superfast but not as fast as of our new AI.
Meet DeepCube, an artificially intelligent system that is as good at playing the Rubik’s Cube as the best human master solvers. DeepCube is a new kind of deep-learning machine. It has mastered solving the Rubik’s Cube puzzle in just 44 hours without any human help. Deepmind takes only 4.22 seconds to fully solved the Rubik’s cube.
DeepCube is created by a research team from the University of California, Irvine. The team was comprised of PhD candidates Stephen McAleer, Forest Agostinelli and Alexander Shmakov, as well as UC Irvine computer science professor Pierre Baldi.
The paper has published titled ‘Solving the “Rubik’s Cube Without Human Knowledge’.
“A generally intelligent agent must be able to teach itself how to solve problems in complex domains with minimal human supervision,” write the paper’s authors. “Indeed, if we’re ever going to achieve a general, human-like machine intelligence, we’ll have to develop systems that can learn and then apply those learnings to real-world applications.”
Creating such kind of AI was not a easily task for the researchers. The biggest challenge for the team was teaching DeepCube how to learn. To solve the learning problem the research team from the University of California, Irvine, uses reinforcement learning technique.
Reinforcement learning is the most use technique. It is widely use to teach AI how to play games like chess or Go. Games like chess are those games which have a definitive way to tell whether a move is ‘good’ or ‘bad.’ It makes learning easy for AI. However, Rubik’s Cubes don’t have good or bad moves in the conventional sense.
To solve this problem team created a new technique called autodidactic iteration. It allows system to figure out which faces to turn and when. By using this technique, DeepCube trained itself without any human help.
In Autodidactic iteration technique, the system starts with the finished cube and works backwards to find a configuration that is similar to the proposed move. In this process it figures out whether a move would be an improvement over the current scrambled configuration.
This system help DeepCube to evaluate the overall success of the move and proficiency of the move. When DeepCube decides on a move it jumps all the way forward to the completed cube then all the way back to its proposed adjustment.
DeepCube developed its own reward system. By observing the changes in the cube, it learned to evaluate the possible success of its proposed moves.
Once it has collected enough data about its current position, it uses a traditional tree search method to examine each possible move to determine the best one. It repeats this until it finishes the cube.
The result of this algorithm is remarkably well. “Our algorithm is able to solve 100% of randomly scrambled cubes while achieving a median solve length of 30 moves—less than or equal to solvers that employ human domain knowledge,” say McAleer and co.
DeepCube took only 44 hours to learn how to complete solve any standard, randomly scrambled Rubik’s Cube in an average of 30 moves or less, which is about the same (or better than) what we can do.
Along with earning how to solve a Rubik’s Cube, DeepCube picked up a lot of knowledge about the cube along the way. The system discovered “a notable amount of Rubik’s Cube knowledge during its training process,” write the researchers, including a strategy used by advanced speedcubers, namely a technique in which the corner and edge cubelets are matched together before they’re placed into their correct location
The researchers now plan on testing the Autodidactic Iteration technique by trying to teach DeepCube how to solve harder 16-sided cubes. The research could also be used to solve real-world problems. Some applications include predicting the 3D shape of proteins. Instead of figuring out the next move of a Rubik’s Cube, the system could determine the proper sequence of amino acids along a 3D lattice.
McAleer said “We are working on extending this method to find approximate solutions to other combinatorial optimization problems such as prediction of protein tertiary structure.”
More in AI :