In many reinforcement learning (RL) papers, Markov Decision Process (MDP) is a typical problem setting for RL problem. What is the real benefit of this setting? Some papers use LSTM as their policy network structure which obviously violate the MDP assumption and make more sense.
Basically, Markov Decision Processes provide a theoretical framework that allows to analyze the convergence guarantees of the algorithms as well as other theoretical properties. Although LSTM and other deep learning approaches combined with RL have reached impressive results, they lack from a solid theoretical background that allow understand or ensure when the algorithm is going to learn something useful, or how far the learned policy will be from the optimal one.
Related
As far as I know, NEAT (NeuroEvolution of Augmenting Topologies) is an algorithm that uses the concept of evolution to train a neural network. On the other hand, reinforcement learning is a type of machine learning with the concept of "rewarding" more successful nodes.
What is the difference between these two fields as they seem to be quite similar? Or is NEAT derived from reinforcement learning?
In short they have barely anything in common.
NEAT is an evolutionary method. This is a black box approach to optimization of functions. In this case - performance of the neural net (which can be easily measured) wrt. to its architecture (which you alter during evolution).
Reinforcement learning is about agents, learning policies to behave well in the environment. Thus they solve different, more complex problem. In theory you could learn NEAT using RL, as you might pose the problem of "given a neural network as a state, learn how to modify it over time to get better performance". The crucial difference will be - NEAT output is a network, RL output is a policy, strategy, algorithm. Something that can be used multiple times to work in some environment, take actions and obtain rewards.
I am a beginner in machine learning and recently read about supervised and unsupervised machine learning. It looks like supervised learning is synonymous to classification and unsupervised learning is synonymous to clustering, is it so?
No.
Supervised learning is when you know correct answers (targets). Depending on their type, it might be classification (categorical targets), regression (numerical targets) or learning to rank (ordinal targets) (this list is by no means complete, there might be other types that I either forgot or unaware of).
On the contrary, in unsupervised learning setting we don't know correct answers, and we try to infer, learn some structure from data. Be it cluster number or low-dimensional approximation (dimensionality reduction, actually, one might think of clusterization as of extreme 1D case of dimensionality reduction). Again, this might be far away from completeness, but the general idea is about hidden structure, that we try to discover from data.
Supervised learning is when you have labeled training data. In other words, you have a well-defined target to optimize your method for.
Typical (supervised) learning tasks are classification and regression: learning to predict categorial (classification), numerical (regression) values or ranks (learning to rank).
Unsupservised learning is an odd term. Because most of the time, the methods aren't "learning" anything. Because what would they learn from? You don't have training data?
There are plenty of unsupervised methods that don't fit the "learning" paradigm well. This includes dimensionality reduction methods such as PCA (which by far predates any "machine learning" - PCA was proposed in 1901, long before the computer!). Many of these are just data-driven statistics (as opposed to parameterized statistics). This includes most cluster analysis methods, outlier detection, ... for understanding these, it's better to step out of the "learning" mindset. Many people have trouble understanding these approaches, because they always think in the "minimize objective function f" mindset common in learning.
Consider for example DBSCAN. One of the most popular clustering algorithms. It does not fit the learning paradigm well. It can nicely be interpreted as a graph-theoretic construct: (density-) connected components. But it doesn't optimize any objective function. It computes the transitive closure of a relation; but there is no function maximized or minimized.
Similarly APRIORI finds frequent itemsets; combinations of items that occur more than minsupp times, where minsupp is a user parameter. It's an extremely simple definition; but the search space can be painfully large when you have large data. The brute-force approach just doesn't finish in acceptable time. So APRIORI uses a clever search strategy to avoid unnecessary hard disk accesses, computations, and memory. But there is no "worse" or "better" result as in learning. Either the result is correct (complete) or not - nothing to optimize on the result (only on the algorithm runtime).
Calling these methods "unsupervised learning" is squeezing them into a mindset that they don't belong into. They are not "learning" anything. Neither optimizes a function, or uses labels, or uses any kind of feedback. They just SELECT a certain set of objects from the database: APRIORI selects columns that frequently have a 1 at the same time; DBSCAN select connected components in a density graph. Either the result is correct, or not.
Some (but by far not all) unsupervised methods can be formalized as an optimization problem. At which point they become similar to popular supervised learning approaches. For example k-means is a minimization problem. PCA is a minimization problem, too - closely related to linear regression, actually. But it is the other way around. Many machine learning tasks are transformed into an optimization problem; and can be solved with general purpose statistical tools, which just happen to be highly popular in machine learning (e.g. linear programming). All the "learning" part is then wrapped into the way the data is transformed prior to feeding it into the optimizer. And in some cases, like for PCA, a non-iterative way to compute the optimum solution was found (in 1901). So in these cases, you don't need the usual optimization hammer at all.
I have little background knowledge of Machine Learning, so please forgive me if my question seems silly.
Based on what I've read, the best model-free reinforcement learning algorithm to this date is Q-Learning, where each state,action pair in the agent's world is given a q-value, and at each state the action with the highest q-value is chosen. The q-value is then updated as follows:
Q(s,a) = (1-α)Q(s,a) + α(R(s,a,s') + (max_a' * Q(s',a'))) where α is the learning rate.
Apparently, for problems with high dimensionality, the number of states become astronomically large making q-value table storage infeasible.
So the practical implementation of Q-Learning requires using Q-value approximation via generalization of states aka features. For example if the agent was Pacman then the features would be:
Distance to closest dot
Distance to closest ghost
Is Pacman in a tunnel?
And then instead of q-values for every single state you would only need to only have q-values for every single feature.
So my question is:
Is it possible for a reinforcement learning agent to create or generate additional features?
Some research I've done:
This post mentions A Geramifard's iFDD method
http://www.icml-2011.org/papers/473_icmlpaper.pdf
http://people.csail.mit.edu/agf/Files/13RLDM-GQ-iFDD+.pdf
which is a way of "discovering feature dependencies", but I'm not sure if that is feature generation, as the paper assumes that you start off with a set of binary features.
Another paper that I found was apropos is Playing Atari with Deep Reinforcement Learning, which "extracts high level features using a range of neural network architectures".
I've read over the paper but still need to flesh out/fully understand their algorithm. Is this what I'm looking for?
Thanks
It seems like you already answered your own question :)
Feature generation is not part of the Q-learning (and SARSA) algorithm. In a process which is called preprocessing you can however use a wide array of algorithms (of which you showed some) to generate/extract features from your data. Combining different machine learning algorithms results in hybrid architectures, which is a term you might look into when researching what works best for your problem.
Here is an example of using features with SARSA (which is very similar to Q-learning).
Whether the papers you cited are helpful for your scenario, you'll have to decide for yourself. As always with machine learning, your approach is highly problem-dependent. If you're in robotics and it's hard to define discrete states manually, a neural network might be helpful. If you can think of heuristics by yourself (like in the pacman example) then you probably won't need it.
Many machine learning competitions are held in Kaggle where a training set and a set of features and a test set is given whose output label is to be decided based by utilizing a training set.
It is pretty clear that here supervised learning algorithms like decision tree, SVM etc. are applicable. My question is, how should I start to approach such problems, I mean whether to start with decision tree or SVM or some other algorithm or is there is any other approach i.e. how will I decide?
So, I had never heard of Kaggle until reading your post--thank you so much, it looks awesome. Upon exploring their site, I found a portion that will guide you well. On the competitions page (click all competitions), you see Digit Recognizer and Facial Keypoints Detection, both of which are competitions, but are there for educational purposes, tutorials are provided (tutorial isn't available for the facial keypoints detection yet, as the competition is in its infancy. In addition to the general forums, competitions have forums also, which I imagine is very helpful.
If you're interesting in the mathematical foundations of machine learning, and are relatively new to it, may I suggest Bayesian Reasoning and Machine Learning. It's no cakewalk, but it's much friendlier than its counterparts, without a loss of rigor.
EDIT:
I found the tutorials page on Kaggle, which seems to be a summary of all of their tutorials. Additionally, scikit-learn, a python library, offers a ton of descriptions/explanations of machine learning algorithms.
This cheatsheet http://peekaboo-vision.blogspot.pt/2013/01/machine-learning-cheat-sheet-for-scikit.html is a good starting point. In my experience using several algorithms at the same time can often give better results, eg logistic regression and svm where the results of each one have a predefined weight. And test, test, test ;)
There is No Free Lunch in data mining. You won't know which methods work best until you try lots of them.
That being said, there is also a trade-off between understandability and accuracy in data mining. Decision Trees and KNN tend to be understandable, but less accurate than SVM or Random Forests. Kaggle looks for high accuracy over understandability.
It also depends on the number of attributes. Some learners can handle many attributes, like SVM, whereas others are slow with many attributes, like neural nets.
You can shrink the number of attributes by using PCA, which has helped in several Kaggle competitions.
I am working on binary classification of data and I want to know the advantages and disadvantages of using Support vector machine over decision trees and Adaptive Boosting algorithms.
Something you might want to do is use weka, which is a nice package that you can use to plug in your data and then try out a bunch of different machine learning classifiers to see how each works on your particular set. It's a well-tread path for people who do machine learning.
Knowing nothing about your particular data, or the classification problem you are trying to solve, I can't really go beyond just telling you random things I know about each method. That said, here's a brain dump and links to some useful machine learning slides.
Adaptive Boosting uses a committee of weak base classifiers to vote on the class assignment of a sample point. The base classifiers can be decision stumps, decision trees, SVMs, etc.. It takes an iterative approach. On each iteration - if the committee is in agreement and correct about the class assignment for a particular sample, then it becomes down weighted (less important to get right on the next iteration), and if the committee is not in agreement, then it becomes up weighted (more important to classify right on the next iteration). Adaboost is known for having good generalization (not overfitting).
SVMs are a useful first-try. Additionally, you can use different kernels with SVMs and get not just linear decision boundaries but more funkily-shaped ones. And if you put L1-regularization on it (slack variables) then you can not only prevent overfitting, but also, you can classify data that isn't separable.
Decision trees are useful because of their interpretability by just about anyone. They are easy to use. Using trees also means that you can also get some idea of how important a particular feature was for making that tree. Something you might want to check out is additive trees (like MART).