Twenty-four years ago, chess master Garry Kasparov, who lost to IBM's "Deep Blue" computer in his area of best, was the first intellectual worker whose job was threatened by a machine.
In March 2016, the computer went down to the next city. AlphaGo, the artificial intelligence of Google's DeepMind team, defeated South Korean Go player Lee Sedol by an aggregate score of 4:1 in the "Go Man-Machine War", and Lee Sedol's weak appearance is still vividly remembered.
Lee Sedol made a forehead gesture
In May 2017, AlphaGo and Chinese Go player Ke Jie opened the second "Go Man-Machine War". Although Ke Jie played a game of chess with AlphaGo that even AphlaGo evaluated perfectly in the second game of the triple game, in terms of results, the Chinese player did not win a game from AphlaGo! In the end, AlphaGo defeated Ke Jie 3:0. Artificial intelligence has gained further improvements in the dominance of chess sports.
Ke Jie wept silently
In 2018, AlphaFold, an AI program developed by DeepMind, made its debut in the Global Protein Structure Prediction Competition (CASP). Although "AlphaFold" won the first place, it did not show any revolutionary difference than the traditional thinking.
AlphaFold's neural network model architecture
In 2020, the "AlphaFold2" king returned, using machine learning, to predict the correct topological structure for almost all proteins, of which about 2/3 of the protein prediction accuracy reached the measurement accuracy of structural biology experiments. With this result, he won the crown again.
In 2021, according to foreign media reports on October 7, DeepMind, as one of the world's important artificial intelligence laboratories, has finally achieved its first profit in history. As a reference, take JD.com as an example. JD.com has spent more than a decade since it set foot in its own e-mall, and it didn't start to make a profit until 2017.
DeepMind
OK, after reading the achievements of artificial intelligence over the years. Let's talk about the machine learning behind AlphaGo.
In terms of specific algorithms, AlphaGo uses deep convolutional neural networks (CNNs) to train value networks and policy networks, uses the feature learning capabilities of deep learning, only uses general supervised learning algorithms and reinforcement learning to achieve domain modeling, and also uses Monte Carlo tree search (MCTS) to greatly improve search efficiency.
AlphaFold2 borrows from the Transformer architecture in AI research. Transformer emerged from the NLP field using a attention mechanism to process a series of sequences of text.
I believe that many friends are very interested in artificial intelligence, but suffering from their own lack of relevant skills, here, I recommend a book, is Zhou Zhihua's "Machine Learning", which is commonly known as the watermelon book hhh.
Zhou Zhihua", "Machine Learning"
● Friends who need to receive information related to the article
○1, pay attention to I remember the comments.
●2. The private message author "information" to obtain contact information to collect information.