Do you know all these artificial intelligence

Do these AI know that the world cup is hot and cold

the national teams of the major powers of the Russian World Cup have been repeatedly cold. For example, Brazil was the favorite team to win the world cup, but it lost to Belgium in the quarter finals. Samba Legion even left Russia on the way back to Brazil, by the country's extreme fans stoned the team's bus to vent their anger

on the other hand, there are media reports that the Brazilian media believe that the reason why the Brazilian team was able to win the world cup five times is related to sticking to eating black bean rice, a local specialty food, and the reason why it did not perform well this time is because there was a problem in the dietary arrangement

is it because of diet, rest before the game, or (7) the guide wheel on the workbench collides with the column for other reasons? What determines the final outcome of the game? With the rapid development of data intelligence and artificial intelligence, how do data scientists view the prediction of football matches? For the development of this field, we discussed with Xie Bo, CEO of Beijing guess technology

"the prediction of the world cup is very difficult. The fundamental reason is that there are too few battle records between teams. At the same time, the player lineup is changed every four years, and there is not enough historical data to help model", Xiebo said. "If anyone claims that he can predict the victory or defeat of the world cup, it is basically the reconstruction and deep modeling of fundamental data, but in terms of specific samples, the effect is also unsatisfactory."

"so the world cup is still one of the most uncertain international events." Xiebo added, "for similar matches, we have made a very good prediction of the five major leagues based on the unique historical sample data."

"in football matches, every player can't ensure his stable play, and the coach's tactical execution often changes... There are so many mobile factors in the game, how do you model and predict the game?"

"first of all, we need to popularize several basic concepts: odds, prediction probability and return on value. Then we can clearly see how AI is changing this industry."

we know that the odds of a football lottery is the betting return of a football game given comprehensive consideration by the gambling company. For example, the gambling company offers a odds of 2.25 for the home team to win, which means that if the gambler wins the home team and the game result is really the home team wins, the gambler can get a return of 2.25 times his betting amount. As for the judgment of the competition results, we give a certain probability estimation for the three results of victory and draw, that is, the prediction probability. Football lottery odds and prediction probability become the core two points. On this basis, we will also pay attention to the return on value, which comprehensively considers the odds and result probability. The return on value represents the expected return of betting on this result

therefore, we look at this problem from the perspective of Chinese sports lottery players: how to exchange the minimum money for the maximum rate of return? Similar to stock investment, there is the best portfolio, and artificial intelligence can efficiently select these portfolios! There are relatively mature industry theories:

returns = max{ σ odds | pro, val, α } ， Where odds is the odds of the prediction result, Pro is the probability of prediction, Val is the return on value of prediction, α To choose a strategy... In a word, this is a hard-working industry formula, which is what Actuaries of major gambling companies rely on

"To this end, guess sports National Engineering Laboratory is an important part of the national scientific and technological innovation system. The education team has developed a game prediction model based on deep learning technology and a betting strategy model based on value return. Guess sports' core comes from Baidu, Tencent and Didi's AI technology, algorithm and product teams, which introduce the advantages of AI into the field of football lottery to provide users with the best betting opportunities and unexpected reminders And real-time bowling strategy. "

1. Game result prediction model based on deep learning

deep learning has achieved great success in image recognition, speech recognition, natural language processing and other fields. By introducing deep learning technology into the prediction of football match results, we can make comprehensive use of historical game records and various real-time data information for training and calculation. Whether it is the injury of players before the game or the strange move of the coach, many clues can be obtained from the massive data to help the model get accurate prediction results

figure: Game prediction based on deep neural networks

further, the team also uses multiple sub models to fuse the results at multiple levels, just like making a perfect meal. It must be constantly tried to make the most delicious food and obtain the maximum sexual benefits. Xie Bo said that the current guessing model can achieve 68% prediction accuracy

during the world cup, based on such technical advantages, guess the small program opened by the sports team during the world cup launched a challenge to the majority of lottery fans and fans. AI and everyone predicted the goal time of a game at the same time

"after the start of the game, the data that is far from constant is produced rapidly. In fact, the accuracy of prediction in the game is very high, such as predicting the goal time in the game." Xie Bo added

figure: guess the small program launched by science and technology during the world cup

2. Search for the optimal portfolio with the introduction of value return

under the condition that the result probability and odds are known, which games are selected for betting, and what is the proportion of betting, which is the concern of betting strategy. Different prediction probabilities will have different value return values. According to different value return rates, we will adopt different betting strategies, which involves the search of an optimal portfolio. The search space is based on the combination of odds, prediction probability, return on value and different prediction models. How to carry out the optimal search in millions of portfolios has become the core problem. Of course, these complex calculation processes should be left to our AI

figure: parameter combination search based on genetic algorithm

guess the sports team uses pruning optimization strategy, combined with genetic algorithm and other search strategies (in short, some algorithms and strategies) to retrieve the optimal parameter space. In order to test this model, the guess sports team conducted simulation training for 20000 football matches from 2012 to 2018, and tested 300 matches in the test set, with a return rate of 41%. All bets are 1000 yuan, 300 bets, 300000 yuan invested, and 123000 yuan returned

finally, guess sports will create a deep learning model that can stably predict the results through the correction and leveling of ⑨ platform in this accidental sports competition. For sports fans, it is naturally very interesting to bet a few small bets to support their beloved team. Through the blessing of AI, they can get higher and more stable returns. Although we mentioned many professional terms above, its prototype can be found in life. The majority of sports fans only need to choose the team they support, and the rest is left to our AI

"the engineers from Baidu basically have some feelings of artificial intelligence. Many engineers hope to try various projects in the way of data intelligence. In fact, the prediction of competitive sports events can especially use the power of big data and neural networks."

according to Guo Yang, guess technology CTO, at present, they have accumulated tens of thousands of historical data of European mainstream leagues in the past 10 years, as well as millions of real-time data in the game. The accuracy of prediction for the game results has exceeded that of most human experts

"how well do you predict the football match?"

"The prediction results of our neural network are compared with the betting strategy samples of many senior lottery players, and many interesting details are found. For example, most people make some statistical mistakes when buying football betting. In some cases, there are many inducements in the opening given by the gambling company itself. If we do not make an objective analysis from the data, it is difficult for us to realize that in the long run, the general It is difficult for fans and lottery fans of Tong to obtain positive returns in terms of expected returns. But our neural network model captures a large number of similar factors. For example, most people subconsciously prefer to bet on the hanging wall, and the results of the neural network show that the proportion of the machine choosing the footwall will be much higher. In the final analysis, the machine has maintained absolute rationality and objectivity in the process of predicting the outcome of the game. "