There were in and AI in 2017, but few generated as much publicity and interest as DeepMind’s . This program was truly a shocking breakthrough: not only did it beat the prior version of AlphaGo — the program that beat 17 time world champion Lee Sedol just a year and a half earlier — 100–0, it was trained without any data from real human games. called it in . many advances Deep Learning AlphaGo Zero Xavier Amatrain “more [significant] than anything…in the last 5 years” Machine Learning So how did DeepMind do it? In this essay, I’ll try to give an intuitive idea of the techniques AlphaGo Zero used, what made them work, and what the implications for future AI research are. Let’s start with the general approach that both AlphaGo and AlphaGo Zero took to playing Go. DeepMind’s General Approach Both AlphaGo and AlphaGo Zero evaluated the Go board and chose moves using a combination of two methods: Performing “ ” search: looking ahead several moves by simulating games, and thus seeing which current move is most likely to lead to a “good” position in the future. lookahead Evaluating positions based on an “ ”, of whether a position is “good” or “bad” — that is, likely to lead to a win or a loss. intuition AlphaGo and AlphaGo Zero both worked by cleverly combining these two methods. Let’s look at each one in turn: Go-Playing Method #1: “Lookahead” Go is a sufficiently complex game that computers can’t simply search all possible moves using a brute force approach to find the best one (indeed, ). they can’t even come close The tree of possible positions in Go. Source The best Go programs prior to AlphaGo overcame this by using “ ” or MCTS. At a high level, this method involves initially exploring many possible moves on the board, and then focusing this exploration over time as certain moves are found to be more likely to lead to wins than others. Monte Carlo Tree Search Both AlphaGo and AlphaGo Zero use a relatively straightforward version of MCTS for their “lookahead”, simply using many of the best practices listed in the to properly manage the tradeoff between exploring new sequences of move or more deeply explore already-explored sequences (for more, see the details in the “Search” section under “Methods” in ). Monte Carlo Tree Search Wikipedia page the original AlphaGo Paper published in Nature Though, MCTS had been the core of all successful Go programs prior to AlphaGo, it was DeepMind’s clever combination of this technique with a neural network-based “intuition” that allowed it to surpass human performance. Go-Playing Method #2: “Intuition” DeepMind’s major innovation with AlphaGo was to use deep neural networks to understand the state of the game, and then use this understanding to intelligently guide the search of the MCTS. More specifically: they trained networks that could look at The current board position Which player was playing, The sequence of recent moves (necessary to rule out certain moves as illegal) Given this information, the neural networks could recommend: Which move should be played Whether the current player was likely to win or not. How did DeepMind train neural networks to do this? Here, AlphaGo and AlphaGo Zero used very different approaches; we’ll start first with AlphaGo’s: AlphaGo’s “Intuition”: Policy Network and Value Network AlphaGo had two separately trained neural networks. The two neural networks at the core of AlphaGo. Source The first neural network (initialized randomly) was trained to mimic the play of human experts by being shown 30 million moves from a large database of real games. Solving this was a difficult but straightforward pattern recognition problem of the kind ; and indeed, once trained, this network did indeed learn to recommend moves similar to those that it observed human experts playing in their games. at which deep neural networks excel DeepMind didn’t just want AlphaGo to mimic human players: they also wanted it to win. To learn to play moves more likely to lead to winning instead of losing, the network — that had been trained to play like a human expert — played games against itself. Moves were then randomly sampled from these “self-play” games; if a given move happened in a game in which the current player ended up winning, the network was trained to be more likely to play moves like that in the future, and vice versa. DeepMind then combined these two neural networks with MCTS — that is, the program’s “intuition” with its brute force “lookahead” search— in a very clever way: it used the network that had been trained to to of the game tree to search and used the network that had been trained to to the positions it encountered during its search. This allowed AlphaGo to intelligently search upcoming moves and ultimately allowed it to beat Lee Sedol. predict moves guide which branches predict whether a position was “winning” evaluate Monte Carlo Tree Search in AlphaGo, guided by neural networks. Source AlphaGo Zero, however, took this to a whole new level. The three tricks that made AlphaGo Zero work At a high level, AlphaGo Zero works the same way as AlphaGo: specifically, it plays Go by using MCTS-based lookahead search, intelligently guided by a neural network. However, AlphaGo Zero’s neural network — its “intuition” — was trained completely differently from that of AlphaGo: Trick #1: How to Train Your AlphaGo Zero, Part 1 Let’s say you have a neural network that is attempting to “understand” the game of Go: that is, for every board position, it is using a deep neural network to generate evaluations of what the best moves are. What DeepMind realized is that no matter how intelligent this neural network is — whether it is completely clueless or a Go master — . its evaluations can always be made better by MCTS Fundamentally, MCTS performs the kind of lookahead search that we would imagine a human master would perform if given enough time: it intelligently guesses which variations— sequences of moves — are most promising, simulates those variations, evaluates how good they are, and updates its assessments of its best moves accordingly. future actually current An illustration of this is below. Suppose we have a neural network that is reading the board and determining that a given move results in a game being even, with an evaluation of 0.0. Then, the network intelligently looks ahead a few moves and finds a sequence of moves that can be forced from the current position that ends up resulting in an evaluation of 0.5. It can then update its evaluation of the board position to reflect that it leads to a more favorable position down the road. current How MCTS can always continually improve programs’ evaluations. This lookahead search, therefore, can always give us improved data on how good the various moves in the position that the neural network is evaluating are. This is true whether our neural network is playing at an amateur level or an expert level: we can generate improve evaluations for it by looking ahead and seeing which of its current options lead to better positions. current always actually Trick #1 (continued): How to Train Your AlphaGo Zero, Part 2 In addition, just as in AlphaGo, we would also want our neural network to learn which moves are likely to lead to wins. So, also as before, our agent—using its MCTS-improved evaluations and the current state of its neural network — could play games against itself, winning some and losing others. Self-play in AlphaGo Zero. Diagram courtesy of . DeepMind This data, generated purely via lookahead and self-play, is what DeepMind used to train AlphaGo Zero. More specifically: The neural network was trained to play moves that reflected the improved evaluations from performing the “lookahead” search. The neural network was adjusted so that it was more likely to play moves similar to those that led to wins and less likely to play moves similar to those that led to losses during the self-play games. Much was made of the fact that no games between humans were used to train AlphaGo Zero, and this first “trick” was the reason why: for a given state of a Go agent, it can always be made smarter by performing MCTS-based lookahead and using the results of that lookahead to improve the agent. This is how AlphaGo Zero was able to continuously improve, from when it was an amateur all the way up to when it better than the best human players. The second trick was a novel neural network structure that I’ll call the “Two Headed Monster”. Trick #2: The Two Headed Monster AlphaGo Zero’s was its neural network architecture, a “two-headed” architecture. Its first 20 layers or so were layer “blocks” of a type often seen in modern neural net architecures. These layers were followed by : one head that took the output of the first 20 layers and produced probabilities of the Go agent making certain moves, and another that took the output of the first 20 layers and outputted a probability of the current player winning. two “heads” AlphaGo Zero’s two headed neural network architecture. Diagram courtesy of . DeepMind This is quite unusual. In neural networks output a single, fixed output — such as the probability of an image containing a dog, or a vector containing the probabilities of an image containing one of 10 types of objects. How can a net learn if it is receiving two sets of signals: one on how good its evaluations of the board are, and another how good the specific moves it is selecting are? almost all applications, The answer is simple: remember that neural networks are fundamentally just mathematical functions with a bunch of parameters that determine the predictions that they make; we “teach” them by repeatedly showing them “correct answers” and having them update their parameters so the answers they produce more closely match these correct answers. So, when we use the two headed neural net to make a prediction using Head #1, we simply update the parameters that led to making that prediction, namely the parameters in the “Body” and in “Head #1”. Similarly, when we make a prediction using Head #2, we update the parameters in the “Body” and in “Head #2”. Training the two headed neural network, one head at a time. This is how DeepMind trained its that it used to guide MCTS during its search, just as AlphaGo did with two separate neural networks. This trick accounted for half of AlphaGo Zero’s increase in playing strength over AlphaGo. single, “two-headed” neural network (this trick is known more technically as Multi-Task Learning with Hard Parameter Sharing. ). Sebastian Ruder has a great overview here The other half of the increase in playing strength simply came from bringing the neural network architecture up-to-date with the latest advances in the field: Trick #3: “Residual” Nets AlphaGo Zero used a more “cutting edge” neural network architecture than AlphaGo. Specifically, they used a “residual” neural network architecture instead of a purely “convolutional” architecture. Residual nets were , right around the time work on the first version of AlphaGo would have wrapped up, so it both understandable that DeepMind did not use them in the original AlphaGo program. pioneered by Microsoft Research in late 2015 Diagram comparing residual to convolutional architectures, from the original “ResNet” paper. Source Interestingly, as the chart below shows, each of these two neural network-related tricks — switching from convolutional to residual architecture and using the “Two Headed Monster” neural network architecture instead of separate neural networks — would have resulted in about half of the increase in playing strength as was achieved when both were combined. Improvmenet of AlphaGo Zero — which used a “Dual-Residual” neural network architecture — over AlphaGo, which used a “Separate-Convolutional” architecture. Chart courtesy of . DeepMind Summary of Tricks These three tricks are what enabled AlphaGo Zero to achieve its incredible performance that blew away even Alpha Go: Using the evaluations provided by Monte Carlo Tree Search— “intelligent lookahead” — to continually improve the neural network’s evaluation of board positions, instead of using human games. Using one neural network — the “Two Headed Monster” that simultaneously learns both which moves “intelligent lookahead” would recommend which moves are likely to lead to victory — instead of two separate neural networks. and Using a more cutting edge neural network architecture — a “residual” architecture rather than a “convolutional” architecture. One comment It is worth noting that AlphaGo did not use any classical or even “cutting edge” reinforcement learning concepts — no Deep Q Learning, Asynchronous Actor-Critic Agents, or anything else we typically associate with reinforcement learning. It simply used simulations to generate training data for its neural nets to then learn from in a supervised fashion. sums this idea up well in this Tweet from just after when the AlphaGo Zero paper was released: Denny Britz Follow Denny . here The Numbers: Training AlphaGo Zero, Step-by-Step Here’s a “step-by-step” timeline of how AlphaGo Zero was trained: Initialize neural network. Play self-play games, using MCTS simulations per move (which takes about 0.4 seconds). 1,600 One MCTS iteration in AlphaGo Zero. Diagram courtesy of . DeepMind 3. As these self-play games are happening, sample from the most recent games, along with whether the game was won or lost. For each move, record both A) the results of the MCTS evaluations of those positions — how “good” the various moves in these positions were based on lookahead — and B) whether the current player won or lost the game. 2,048 positions 500,000 4. Train the neural network, using both A) the move evaluations produced by the MCTS lookahead search and B) whether the current player won or lost. 5. Finally, every 1,000 iterations of steps 3–4, evaluate the current neural network against the previous best version; if it wins at least 55% of the games, begin using to generate self-play games instead of the prior version. it Repeat steps 3–4 times, while the self-play games are continuously being played — after three days, you’ll have yourself an AlphaGo Zero! 700,000 Implications for the rest of AI There are many implications of DeepMind’s incredible achievement for the future of AI research. Here are a couple of key ones: First, the fact that self-play data generated from simulations was “good enough” to be able to train the network suggests that — data generated from human experts may not be needed. simulated self-play data train agents to surpass human performance in extremely complex tasks, even starting completely from scratch can Second, , since it seems to prevent the agents from overfitting their behavior to any individual task. DeepMind seems to really like this trick, and has used it and more advanced versions of it to build agents that can learn multiple tasks in . the “Two Headed Monster” trick seems to significantly help agents learn to perform several related tasks in many domains several different domains DeepMind’s AI learning to solve mazes, using the “Distral” framework for multitask reinforcement learning. Source Many projects in robotics, especially the burgeoning field of using simulations to teach robotic agents to use their limbs to accomplish tasks, are using these two tricks to great effect. highlights many impressive new results that use these tricks many bleeding edge reinforcement learning techniques. Indeed, locomotion seems like a perfect use case for the “Two Headed Monster” trick in particular: for example, robotic agents could be simultaneously trained to hit a baseball using a bat and to throw a punch to hit a moving target, since the two tasks require learning some common skills (e.g. balance, torso rotation). Pieter Abbeel’s recent NIPS keynote along with The tricks DeepMind used to train AlphaGo Zero have already been applied to locomotion. Source DeepMind’s AlphaGo Zero was one of the most intriguing advancements in AI and Deep Learning in 2017. I can’t wait to see what 2018 brings!
