The Daily Mail reports that the Computer Poker Research Group at the University of Alberta seems to have solved heads-up limit hold’em poker.
You can play against their AI online.
(My thanks to Glen for emailing me the story!)
Mathematician and Father. Into games, astronomy, psychology and philosophy.
CMU’s Professor Bhiksha Raj has a nice list of papers for his deep learning class. Check ‘em out.
Fernandez-Delgado, Cernadas, Barro, and Amorim tested 179 classfiers on 121 data sets and reported their results in “Do we Need Hundreds of Classi ers to Solve Real World Classi cation Problems?” The classifiers were drawn from the following 17 families:
“discriminant analysis, Bayesian, neural networks, support vector machines, decision trees, rule-based classi fiers, boosting, bagging, stacking, random forests and other ensembles, generalized linear models, nearest-neighbors, partial least squares and principal component regression, logistic and multinomial regression, multiple adaptive regression splines and other methods”
from the Weka, Matlab, and R machine learning libraries. The 121 datasets were drawn mostly from the UCI classification repository.
The overall result was that the random forest classifiers were best on average followed by support vector machines, neural networks, and boosting ensembles.
For more details, read the paper!
Christopher Clark and Amos Storkey wrote an interesting nine page article titled “Teaching Deep Convolutional Neural Networks to Play Go”. Their deep neural network correctly predicted the moves of experts on a 19×19 Go about 44% of the time. The previous record was 41% by Wistuba and Schmidt-Thieme in 2012. Furthermore, the Clark Storkey network was able to “consistently defeat the well-known Go program GNU Go.” This is the first time that a neural network was able to perform nearly as well as one of the better hand coded programs. It is still not as good at the better UCT programs, but it moves much more quickly than the UCT programs. I imagine that if there were a blitz version of computer Go, the Clark Storkey AI might win a computer competition.
The article reviews other recent attempts to train a neural network to play Go. The Clark Storkey network resembled the Wistuba Schmidt-Thieme network, but it had more 19×19 convolutional layers and the authors added one fully connected layer at the top before the final move decision. Also, known symmetries of the solution were hard-coded. Interestingly, they found that convolution seemed to be required.
“We briefly experimented with non-convolutional networks but found them to be much harder to train, often requiring more epochs of training and the use of approximate second order gradient descent methods, while getting worse results.”
Later they describe their training methods and network architecture as follows
“Networks were trained with mini-batch gradient descent with a batch size of 128, using a learning rate of 0.01 for 7 epochs, and 0.05 for 2 epochs which took about a day on a Nvidia GTX 780 GPU.”
“The best network had one convolutional layer with 64 7×7 filters, two convolutional layers with 64 5×5 filters, two layers with 48 5×5 filters, two layers with 32 5×5 filters, and one fully connected layer.”
They estimate that their AI would probably have a ranking near 4-5 kyu.
Mnih, Kavukcuoglu, Silver, Graves, Antonoglon, Wierstra, and Riedmiller authored the paper “Playing Atari with Deep Reinforcement Learning” which describes and an Atari game playing program created by the company Deep Mind (recently acquired by Google). The AI did not just learn how to pay one game. It learned to play seven Atari games without game specific direction from the programmers. (The same learning parameters, neural network topologies, and algorithms were used for every game).
The 2600 Atari gaming system was quite popular in the late 1970’s and the early 1980’s. The games ran with only four kilobytes of RAM and a 210 x 160 pixel display with 128 colors. Various machine learning techniques have been applied to the old Atari games using the Arcade Learning Environment which precisely reproduces the Atari 2600 gaming system. (See e.g. “An Object-Oriented Representation for Efficient Reinforcement Learning” by Diuk, Cohen, and Littman 2008, ”HyperNEAT-GGP:A HyperNEAT-based Atari General Game Player” by Hausknecht, Khandelwal, Miikkulainen, and Stone 2012, “Application of TEXPLORE on Atari Games “ by Shung Zhang , ”A Neuroevolution Approach to General Atari Game Playing” by Hausknect, Lehman,” Miikkulalianen, and Stone 2014, and “Replicating the Paper ‘Playing Atari with Deep Reinforcement Learning’ ” by Korjus, Kuzovkin, Tampuu, and Pungas 2014.)
Various papers have been written on how computers can learn to pay the Atari games, but most of them used the abstract representations of objects on the screen within the emulator. The Mnih et al AI learned to play the games using only the raw 210 x 160 video and the score. It seems to be the first successful attempt to learn arcade gaming from raw video.
To learn from raw video, they first converted the video to grayscale and then downsampled/cropped to 84 x 84 images. The last four frames were used to determine actions. The 28224 input pixels were run through two hidden convolution neural net layers and one fully connected (no convolution) 256 node hidden layer with a single output for each possible action. Training was done with stochastic gradient decent using random samples drawn from a historical database of previous games played by the AI to improve convergence (This technique known as “experience replay” is described in “Reinforcement learning for robots using neural nets” Long-Ji Lin 1993.)
The objective function for supervised learning is usually a loss function representing the difference between the predicted label and the actual label. For these games the correct action is unknown, so reinforcement learning is used instead of supervised learning. The authors used a variant of Q-learning to train the weights in their neural network. They describe their algorithm in detail and compare it to several historical reinforcement algorithms, so this section of the paper can be used as a brief introduction to reinforcement learning.
The AI was trained to play seven games: Beam Rider, Breakout, Enduro, Pong, Q*bert, Seaquest, and Space Invaders. In six of the seven games, this general game learning algorithm outperformed all previously known reinforcement learning algorithms tested on those games and “surpasses a human expert on three” of the seven games.
The KDD 2014 article, written by Dong, Gabrilovich, Heitz, Horn, Lao, Murphy, Strohmann, Sun, and Zang, describes in detail Google’s knowledge database Knowledge Vault. Knowledge Vault is a probability triple store database. Each entry in the database is of the form subject-predicate-object-probability restricted to about 4500 predicates such as “born in”, “married to”, “held at”, or “authored by”. The database was built by combining the knowledge base Freebase with the Wikipedia and approximately one billion web pages. Knowledge Vault appears to be the largest knowledge base ever created. Dong et al. compared Knowledge Vault to NELL, YAGO2, Freebase, and the related project Knowledge Graph. (Knowledge Vault is probabilistic and contains many facts with less than 50% certainty. Knowledge Graph consists of high confidence knowledge.)
The information from the Wikipedia and the Web was extracted using standard natural language processing (NLP) tools including: “named entity recognition, part of speech tagging, dependency parsing, co-reference resolution (within each document), and entity linkage (which maps mentions of proper nouns and their co-references to the corresponding entities in the KB).” The text in these sources is mined using “distance supervision” (see Mintz, Bills, Snow, and Jurafksy “Distant Supervision for relation extraction without labeled data” 2009). Probabilities for each triple store are calculated using logistic regression (via MapReduce). Further information is extracted from internet tables (over 570 million tables) using the techniques in “Recovering semantics of tables on the web” by Ventis, Halevy, Madhavan, Pasca, Shen, Wu, Miao, and Wi 2012.
The facts extracted using the various extraction techniques are fused with logistic regression and boosted decision stumps (see “How boosting the margin can also boost classifier complexity” by Reyzin and Schapire 2006). Implications of the extracted knowledge are created using two techniques: the path ranking algorithm and a modified tensor decomposition.
The path ranking algorithm (see “Random walk inference and learning in a large scale knowledge base” by Lao, Mitchell, and Cohen 2011) can guess that if two people parent the same child, then it is likely that they are married. Several other examples of inferences derived from path ranking are provided in table 3 of the paper.
Tensor decomposition is just a generalization of singular value decomposition, a well-known machine learning technique. The authors used a “more powerful” modified version of tensor decomposition to derive additional facts. (See “Reasoning with Neural Tensor Networks for Knowledge Base Completion” by Socher, Chen, Manning, and Ng 2013.)
The article is very detailed and provides extensive references to knowledge base construction techniques. It, along with the references, can serve as a great introduction to modern knowledge engineering.
Even if you are not into chess, I think you will enjoy reading about Caruana’s amazing performance at Sinquefield.
(The cast comes off my hand today. I will be able to type with both hands soon!)
I read two nice articles this week: “Ten Simple Rules for Effective Computational Research” and “Seven Principles of Learning Better From Cognitive Science”.
In “Ten Simple Rules for Effective Computational Research”, Osborne, Bernabeu, Bruna, Calderhead, Cooper, Dalchau, Dunn, Fletcher, Freeman, Groen, Knapp, McInerny, Mirams, Pitt-Francis, Sengupta, Wright, Yates, Gavaghan, Emmott and Deane wrote up these guidelines for algorithm research:
- Look Before You Leap
- Develop a Prototype First
- Make Your Code Understandable to Others (and Yourself)
- Don’t Underestimate the Complexity of Your Task
- Understand the Mathematical, Numerical, and Computational Methods Underpinning Your Work
- Use Pictures: They Really Are Worth a Thousand Words
- Version Control Everything
- Test Everything
- Share Everything
- Keep Going!
Read the full PLOS 5 page article by clicking here!
Scott Young recently returned from a year abroad and wrote Up “Seven Principles of Learning Better From Cognitive Science” which is a review/summary of the book “Why Don’t Students Like School?” by Daniel Willingham. Here are the 7 Principles:
- Factual knowledge precedes skill.
- Memory is the residue of thought.
- We understand new things in the context of what we already know.
- Proficiency requires practice.
- Cognition is fundamentally different early and late in training.
- People are more alike than different in how we learn.
- Intelligence can be changed through sustained hard work.
Read the full 5 page article by clicking here! Get the great $10 book here.
Monte Carlo Tree Search (survey) is working rather well for Go.
broke my hand on thursday, so typing is difficult.
check out this video