Google DeepMind discovered in a system is the one that won the award of the silver in the International Olympiad of Mathematics (IMO) for this year.
The system in the hybrid's DeepMind, which combines the AlphaProof and AlphaGeometry 2, it was resolved in four of the six problems, the complexity of the proposed and achieved a result in history to artificial intelligence in mathematics, which is the word, for the word may have been competition for the human!
IMO, he is considered as a contest in the most challenging of math in the world, the students of the high school, has become a point of reference for the evaluation of the ability of IT to the solution of the problem. System of DeepMind, the Google gained 28 points out of 42 possible, it is a little bit less than the threshold of the coin, of art, of the 29-point to be made out of just 58 out of the 609 competitors of the human in this year. (See all choices here).
AlphaProof and AlphaGeometry 2, is a winning combination
At the center of this achievement is the AlphaProof, a sample space is one that brings together a template for the language of the pre-trained and therefore, the teaching and learning of the strengthening of the AlphaZero, and the same technology, including algorithms for competing at the games of the complex as chess and Go all the reasoning math with similar strategies in the game by looking through the stages of the available evidence as if it was to move into a tree of decisions.
Different from the approaches of the former, to be limited by the inadequacy of the data to be written by a man, AlphaProof connects to bridge the gap between the language of a natural and formal. It uses a version of the optimized model Gemini to Google translate, the problems of language and a natural, in a formal statement, which has created an extensive library of material for coaching.
An additional AlphaProof is AlphaGeometry 2, an updated version of the zgjidhësit of the geometry of the DeepMind. This system demonstrated the effectiveness of the extraordinary, to solve a problem, IMO, in just 19 seconds. The performance of the its of the upgrade comes from the details of the programme of the training, and a new division of knowledge, to help solve the problems more complex.
Ahead of the announcement of this year's AlphaGeometry 2 be in a position to tackle some 83% of all the problems in the historical, the geometry IMO over the last 25 years, a significant progress on the extent of the success of the 53% of the ancestors of it.
The math there is no secret about IT
Problems in mathematical Olympiad, International Mathematics were translated manually into one of the official languages to be understood by the systems of the artificial intelligence of the DeepMind. While the students are people of two sessions of about 4 and a half hours each, are available for solving the problems, the DeepMind was able to find a solution to a problem in a couple of minutes, taking up to 3 days for the question to be more complex.
This is the result of the extraordinary highlights of the progress in the rapidly become of HIM in the handling of the tasks of the advanced reasoning, logical-mathematical. Systems such as these have the potential, while the evolve further, to speed up innovation in science and to push the boundaries of the capabilities of the human. DeepMind, and will continue to explore new approaches to artificial intelligence for advanced reasoning and math.
Discussion about this post