Online machine learning

Machine learning and data mining

Problems Classification Clustering Regression Anomaly detection Association rules Reinforcement learning Structured prediction Feature engineering Feature learning Online learning Semi-supervised learning Unsupervised learning Learning to rank Grammar induction
Supervised learning (classification • regression) Decision trees Ensembles (Bagging, Boosting, Random forest) k-NN Linear regression Naive Bayes Neural networks Logistic regression Perceptron Relevance vector machine (RVM) Support vector machine (SVM)
Clustering BIRCH Hierarchical k-means Expectation-maximization (EM) DBSCAN OPTICS Mean-shift
Dimensionality reduction Factor analysis CCA ICA LDA NMF PCA t-SNE
Structured prediction Graphical models (Bayes net, CRF, HMM)
Anomaly detection k-NN Local outlier factor
Neural nets Autoencoder Deep learning Multilayer perceptron RNN Restricted Boltzmann machine SOM Convolutional neural network
Reinforcement Learning Q-Learning SARSA Temporal Difference (TD)
Theory Bias-variance dilemma Computational learning theory Empirical risk minimization Occam learning PAC learning Statistical learning VC theory
Machine learning venues NIPS ICML IJMLC ML JMLR ArXiv:cs.LG
Machine learning portal

In computer science, online machine learning is a method of machine learning in which data becomes available in a sequential order and is used to update our best predictor for future data at each step, as opposed to batch learning techniques which generate the best predictor by learning on the entire training data set at once. Online learning is a common technique used in areas of machine learning where it is computationally infeasible to train over the entire dataset, requiring the need of out-of-core algorithms. It is also used in situations where it is necessary for the algorithm to dynamically adapt to new patterns in the data, or when the data itself is generated as a function of time, e.g. stock price prediction.

Two general modelling strategies exist for online learning models: statistical learning models and adversarial models. In statistical learning models (e.g. stochastic gradient descent, perceptrons), the data samples are assumed to be independent and identically distributed random variables (i.e. they are not adapting with time), and the algorithm just has a limited access to the data. In adversarial models, looking at the learning problem as a game between two players (the learner vs the data generator), the goal is to minimize losses regardless of the move played by the other player. In this model, the opponent is allowed to dynamically adapt the data generated based on the output of the learning algorithm. Spam filtering falls in this category, as the adversary will dynamically generate new spam based on the current behavior of the spam detector. Examples of algorithms in this model include follow the leader, follow the regularized leader, etc.

Introduction

In the setting of supervised learning, a function of $f:X\to Y$ is to be learned, where $X$ is thought of as a space of inputs and $Y$ as a space of outputs, that predicts well on instances that are drawn from a joint probability distribution $p(x,y)$ on $X\times Y$ . In reality, the learner never knows the true distribution $p(x,y)$ over instances. Instead, the learner usually has access to a training set of examples $(x_{1},y_{1}),\ldots ,(x_{n},y_{n})$ . In this setting, the loss function is given as $V:Y\times Y\to \mathbb {R}$ , such that $V(f(x),y)$ measures the difference between the predicted value $f(x)$ and the true value $y$ . The ideal goal is to select a function $f\in {\mathcal {H}}$ , where ${\mathcal {H}}$ is a space of functions called a hypothesis space, so that some notion of total loss is minimised. Depending on the type of model (statistical or adversarial), one can devise different notions of loss, which lead to different learning algorithms.

Statistical learning models

In statistical learning models, the training sample $(x_{i},y_{i})$ are assumed to have been drawn i.i.d. from the true distribution $p(x,y)$ and the objective is to minimize the expected "risk"

I[f]=\mathbb {E} [V(f(x),y)]=\int V(f(x),y)\,dp(x,y)\ .

A common paradigm in this situation is to estimate a function ${\hat {f}}$ through empirical risk minimization or regularized empirical risk minimization (usually Tikhonov regularization). The choice of loss function here gives rise to several well-known learning algorithms such as regularized least squares and support vector machines. For the case of online learning, the data is still assumed to be i.i.d without access to all the data. A purely online model in this category would learn based on just the new input $(x_{t+1},y_{t+1})$ , the current best predictor $f_{t}$ and some extra stored information (which is usually expected to have storage requirements independent of training data size). For many formulations, for example nonlinear kernel methods, true online learning is not possible, though a form of hybrid online learning with recursive algorithms can be used where $f_{t+1}$ is permitted to depend on $f_{t}$ and all previous data points $(x_{1},y_{1}),\ldots ,(x_{t},y_{t})$ . In this case, the space requirements are no longer guaranteed to be constant since it requires storing all previous data points, but the solution may take less time to compute with the addition of a new data point, as compared to batch learning techniques.

An important generalization of these techniques is mini-batch techniques, which process a small batch of $b\geq 1$ data points at a time, but can be considered as online algorithms for $b$ much smaller than the total number of training points. Mini-batch techniques are used with repeated passing over the training data (called incremental methods) to obtain optimized out-of-core versions of machine learning algorithms, for e.g. Stochastic gradient descent. When combined with backpropogation, this is currently the de facto training method for training artificial neural networks.

Example: linear least squares

Main article: Linear least squares (mathematics)

The simple example of linear least squares is used to explain a variety of ideas in online learning. The ideas are general enough to be applied to other settings, for e.g. with other convex loss functions.

Batch learning

In the setting of supervised learning with the square loss function, the intent is to minimize the empirical loss,

I_{n}[w]=\sum _{j=1}^{n}V(\langle w,x_{j}\rangle ,y_{j})=\sum _{j=1}^{n}(x_{j}^{T}w-y_{j})^{2}

where

x_{j}\in \mathbb {R} ^{d},w\in \mathbb {R} ^{d},y_{j}\in \mathbb {R}

Let $X$ be the $i\times d$ data matrix and $Y$ is the $i\times 1$ matrix of target values after the arrival of the first $i$ data points. Assuming that the covariance matrix $\Sigma _{i}=X^{T}X$ is invertible (otherwise it is preferential to proceed in a similar fashion with Tikhonov regularization), the best solution $f^{*}(x)=\langle w^{*},x\rangle$ to the linear least squares problem is given by

w^{*}=(X^{T}X)^{-1}X^{T}Y=\Sigma _{i}^{-1}\sum _{j=1}^{i}x_{j}y_{j}

Now, calculating the covariance matrix $\Sigma _{i}=\sum _{j=1}^{i}x_{j}x_{j}^{T}$ takes time $O(id^{2})$ , inverting the $d \times d$ matrix takes time $O(d^{3})$ , while the rest of the multiplication takes time $O(d^{2})$ , giving a total time of $O(id^{2}+d^{3})$ . When $n$ total points in the dataset and having to recompute the solution after the arrival of every datapoint $i=1,\ldots ,n$ , the naive approach will have a total complexity $O(n^{2}d^{2}+nd^{3})$ . Note that when storing the matrix $\Sigma _{i}$ , then updating it at each step needs only adding $x_{i+1}x_{i+1}^{T}$ , which takes $O(d^{2})$ time, reducing the total time to $O(nd^{2}+nd^{3})=O(nd^{3})$ , but with an additional storage space of $O(d^{2})$ to store $\Sigma _{i}$ .^[1]

Online learning: recursive least squares

The recursive least squares algorithm considers an online approach to the least squares problem. It can be shown that by initialising $\textstyle w_{0}=0\in \mathbb {R} ^{d}$ and $\textstyle \Gamma _{0}=I\in \mathbb {R} ^{d\times d}$ , the solution of the linear least squares problem given in the previous section can be computed by the following iteration:

\Gamma _{i}=\Gamma _{i-1}-{\frac {\Gamma _{i-1}x_{i}x_{i}^{T}\Gamma _{i-1}}{1+x_{i}^{T}\Gamma _{i-1}x_{i}}}

w_{i}=w_{i-1}-\Gamma _{i}x_{i}(x_{i}^{T}w_{i-1}-y_{i})

The above iteration algorithm can be proved using induction on $i$ .^[2] The proof also shows that $\Gamma _{i}=\Sigma _{i}^{-1}$ . One can look at RLS also in the context of adaptive filters (see RLS).

The complexity for $n$ steps of this algorithm is $O(nd^{2})$ , which is an order of magnitude faster than the corresponding batch learning complexity. The storage requirements at every step $i$ here are to store the matrix $\Gamma _{i}$ , which is constant at $O(d^{2})$ . For the case when $\Sigma _{i}$ is not invertible, consider the regularised version of the problem loss function $\sum _{j=1}^{n}(x_{j}^{T}w-y_{j})^{2}+\lambda ||w||_{2}^{2}$ . Then, it's easy to show that the same algorithm works with $\Gamma _{0}=(I+\lambda I)^{-1}$ , and the iterations proceed to give $\Gamma _{i}=(\Sigma _{i}+\lambda I)^{-1}$ .^[1]

Stochastic gradient descent

Main article: Stochastic gradient descent

When this is replaced,

\textstyle w_{i}=w_{i-1}-\Gamma _{i}x_{i}(x_{i}^{T}w_{i-1}-y_{i})

\textstyle w_{i}=w_{i-1}-\gamma _{i}x_{i}(x_{i}^{T}w_{i-1}-y_{i})=w_{i-1}-\gamma _{i}\nabla V(\langle w_{i-1},x_{i}\rangle ,y_{i})

or $\Gamma _{i}\in \mathbb {R} ^{d\times d}$ by $\gamma _{i}\in \mathbb {R}$ , this becomes the stochastic gradient descent algorithm. In this case, the complexity for $n$ steps of this algorithm reduces to $O(nd)$ . The storage requirements at every step $i$ are constant at $O(d)$ .

However, the stepsize $\gamma _{i}$ needs to be chosen carefully to solve the expected risk minimization problem, as detailed above. By choosing a decaying step size $\gamma _{i}\approx {\frac {1}{\sqrt {i}}},$ one can prove the convergence of the average iterate ${\overline {w}}_{n}={\frac {1}{n}}\sum _{i=1}^{n}w_{i}$ . This setting is a special case of stochastic optimization, a well known problem in optimization.^[1]

Incremental SGD

In practice, one can perform multiple stochastic gradient passes (also called cycles or epochs) over the data. The algorithm thus obtained is called incremental gradient method and corresponds to an iteration

\textstyle w_{i}=w_{i-1}-\gamma _{i}\nabla V(\langle w_{i-1},x_{t_{i}}\rangle ,y_{t_{i}})

The main difference with the stochastic gradient method is that here a sequence $t_i$ is chosen to decide which training point is visited in the $i$ -th step. Such a sequence can be stochastic or deterministic. The number of iterations is then decoupled to the number of points (each point can be considered more than once). The incremental gradient method can be shown to provide a minimizer to the empirical risk.^[3] Incremental techniques can be advantageous when considering objective functions made up of a sum of many terms e.g. an empirical error corresponding to a very large dataset.^[1]

Kernel methods

Progressive learning

Progressive learning is an effective learning model which is demonstrated by the human learning process. It is the process of learning continuously from direct experience. Progressive learning technique (PLT) in machine learning can learn new classes/labels dynamically on the run.^[4] Though online learning can learn new samples of data that arrive sequentially, they cannot learn new classes of data being introduced to the model. The learning paradigm of progressive learning, is independent of the number of class constraints and it can learn new classes while still retaining the knowledge of previous classes. Whenever a new class (non-native to the knowledge learnt thus far) is encountered, the classifier gets remodeled automatically and the parameters are calculated in such a way that it retains the knowledge learnt thus far. This technique is suitable for real-world applications where the number of classes is often unknown and online learning from real-time data is required.

Adversarial models: sequential learning

Online convex optimisation

In OCO, the hypothesis set and the loss functions are forced to be convex to obtain stronger learning bounds. The modified sequential game is now as follows:

For $t=1,2,...,T$

Learner receives input $x_{t}$
Learner outputs $w_{t}$ from a fixed convex set $S$
Nature sends back a convex loss function $v_{t}:S\rightarrow \mathbb {R}$ .
Learner suffers loss $v_{t}(w_{t})$ and updates its model

Thus, when regret is minimised, we are now competing against the best weight vector $u\in H$ . As an example, consider the case of online least squares linear regression. Here, the weight vectors come from the convex set $S=\mathbb {R} ^{d}$ , and nature sends back the convex loss function $v_{t}(w)=(\langle w,x_{t}\rangle -y_{t})^{2}$ . Note here that $y_{t}$ is implicitly sent with $v_{t}$ .

Some online prediction problems however cannot fit it the framework of OCO. For example, in online classification, the prediction domain and the loss functions are not convex. In such scenarios, two simple techniques for convexification are convexification by randomisation and convexification by use of surrogate loss functions.^[5]

Some simple online convex optimisation algorithms are to be considered:

Follow the leader (FTL)

The simplest learning rule to try is to select (at the current step) the hypothesis that has the least loss over all past rounds. This algorithm is called Follow the leader, and is simply given by:

In round $t$ , set

w_{t}=\operatorname {arg\,min} _{w\in S}\sum _{i=1}^{t-1}v_{i}(w)

Here, ties are broken arbitrarily. This method can thus be looked as a greedy algorithm. For the case of online quadratic optimization (where the loss function is $v_{t}(w)=||w-x_{t}||_{2}^{2}$ ), one can show a regret bound that grows as $\log(T)$ .^[5] However, similar bounds cannot be obtained for the FTL algorithm for other important families of models like online linear optimization etc. To do so, one modifies FTL by adding regularisation.

Follow the regularised leader (FTRL)

This is a natural modification of FTL that is used to stabilise the FTL solutions and obtain better regret bounds. We choose a regularisation function $R:S\rightarrow \mathbb {R}$ and then perform the learning as follows:

In round $t$ , set

w_{t}=*{\underset {w\in S}{\operatorname {arg\,min} }}\sum _{i=1}^{t-1}v_{i}(w)+R(w)

As a special example, consider the case of online linear optimisation i.e. where nature sends back loss functions of the form $v_{t}(w)=\langle w,z_{t}\rangle$ . Also, let $S=\mathbb {R} ^{d}$ . Suppose the regularisation function $R(w)={\frac {1}{2\eta }}||w||_{2}^{2}$ is chosen for some positive number $\eta$ . Then, one can show that the regret minimising iteration becomes ^[5]

w_{t+1}=-\eta \sum _{i=1}^{t}z_{i}=w_{t}-\eta z_{t}

Note that this can be rewritten as $w_{t+1}=w_{t}-\eta \nabla v_{t}(w_{t})$ , which looks exactly like online gradient descent. If $S$ is instead some convex subspace of $\mathbb {R} ^{d}$ , $S$ would need to be projected onto, leading to the modified update rule

w_{t+1}\in \Pi _{S}(-\eta \sum _{i=1}^{t}z_{i})=\Pi _{S}(\eta \theta _{t+1})

This algorithm is known as lazy projection, as the vector $\theta _{t+1}$ accumulates the gradients. It is also known as Nesterov's dual averaging algorithm. In this scenario of linear loss functions and quadratic regularisation, the regret is bounded by $O({\sqrt {T}})$ , and thus the average regret goes to $0$ as desired.^[6]

Online subgradient descent (OSD)

Other sequential algorithms

Quadratically regularised FTRL algorithms lead to lazily projected gradient algorithms as described above. To use the above for arbitrary convex functions and regularisers, one uses online mirror descent. Another algorithm is called prediction with expert advice. In this case, the hypothesis set consists of $d$ functions. We maintain a distribution $w_{t}\in \Delta _{d}$ over the $d$ experts, and predict by sampling an expert from this distribution. For the Euclidean regularisation, one can show a regret bound of $O({\sqrt {T}})$ , which can be improved further to a $O({\sqrt {\log T}})$ bound by using a better regulariser. For further reading about these algorithms, refer to ^[5]^[6]

Comparison of the models

The paradigm of online learning interestingly has three distinct interpretations depending on the choice of the learning model, each of which has distinct implications about the predictive quality of the sequence of functions $f_{1},f_{2},\ldots ,f_{n}$ . The prototypical stochastic gradient descent algorithm is used for this discussion. As noted above, its recursion is given by

\textstyle w_{t}=w_{t-1}-\gamma _{t}\nabla V(\langle w_{t-1},x_{t}\rangle ,y_{t})

Statistical learning model

The first interpretation consider the stochastic gradient descent method as applied to the problem of minimizing the expected risk $I[w]$ defined above.^[7] Indeed, in the case of an infinite stream of data, since the examples $(x_{1},y_{1}),(x_{2},y_{2}),\ldots$ are assumed to be drawn i.i.d. from the distribution $p(x,y)$ , the sequence of gradients of $V(\cdot ,\cdot )$ in the above iteration are an i.i.d. sample of stochastic estimates of the gradient of the expected risk $I[w]$ and therefore one can apply complexity results for the stochastic gradient descent method to bound the deviation $I[w_{t}]-I[w^{\ast }]$ , where $w^{\ast }$ is the minimizer of $I[w]$ .^[8] This interpretation is also valid in the case of a finite training set; although with multiple passes through the data the gradients are no longer independent, still complexity results can be obtained in special cases.

The second interpretation applies to the case of a finite training set and considers the SGD algorithm as an instance of incremental gradient descent method.^[3] In this case, one instead looks at the empirical risk:

I_{n}[w]={\frac {1}{n}}\sum _{i=1}^{n}V(\langle w,x_{i}\rangle ,y_{i})\ .

Since the gradients of $V(\cdot ,\cdot )$ in the incremental gradient descent iterations are also stochastic estimates of the gradient of $I_{n}[w]$ , this interpretation is also related to the stochastic gradient descent method, but applied to minimize the empirical risk as opposed to the expected risk. Since this interpretation concerns the empirical risk and not the expected risk, multiple passes through the data are readily allowed and actually lead to tighter bounds on the deviations $I_{n}[w_{t}]-I_{n}[w_{n}^{\ast }]$ , where $w_{n}^{\ast }$ is the minimizer of $I_{n}[w]$ .

Adversarial model

The third interpretation of the above recursion is distinctly different from the first two and concerns the case of sequential trials where the data are potentially not i.i.d. and can perhaps be selected in an adversarial manner. Since no distributional assumptions are made about the data, the goal here is to perform as well as if the entire sequence of examples could be viewed ahead of time, and the regret is to be minimised on the hypothesis set ${\mathcal {H}}$

R_{T}({\mathcal {H}})=\sum _{t=1}^{T}V(\langle w_{t},x_{t}\rangle ,y_{t})-\min _{w\in H}\sum _{t=1}^{T}V(\langle w,x_{t}\rangle ,y_{t})\ .

In this setting, the above recursion can be considered as an instance of the online subgradient descent method for which there are complexity bounds that guarantee $O({\sqrt {T}})$ regret.^[5]

It should be noted that although the three interpretations of this algorithm yield complexity bounds in three distinct settings, each bound depends on the choice of step-size sequence $\{\gamma _{t}\}$ in a different way, and thus the consequences of all three interpretations cannot be simultaneously applied; the step-size sequence should be selected in a way that is tailored for the interpretation that is most relevant instead. Furthermore, the above algorithm and these interpretations can be extended to the case of a nonlinear kernel by simply considering $X$ to be the feature space associated with the kernel. Although in this case the memory requirements at each iteration are no longer $O(d)$ , but are rather on the order of the number of data points considered so far.

Implementations

Vowpal Wabbit: Open-source fast out-of-core online learning system which is notable for supporting a number of machine learning reductions, importance weighting and a selection of different loss functions and optimisation algorithms. It uses the hashing trick for bounding the size of the set of features independent of the amount of training data.
scikit-learn: Provides out-of-core implementations of algorithms for
- Classification: Perceptron, SGD classifier, Naive bayes classifier.
- Regression: SGD Regressor, Passive Aggressive regressor.
- Clustering: Mini-batch k-means.
- Feature extraction: Mini-batch dictionary learning, Incremental PCA.

Books with substantial treatment of online machine learning

Algorithmic Learning in a Random World by Vladimir Vovk, Alex Gammerman, and Glenn Shafer. Published by Springer Science+Business Media, Inc. 2005 ISBN 0-387-00152-2
Prediction, learning, and games by Nicolò Cesa-Bianchi and Gábor Lugosi. Cambridge University Press, 2006 ISBN 0-521-84108-9

References

1 2 3 4 5 6 7 L. Rosasco, T. Poggio, Machine Learning: a Regularization Approach, MIT-9.520 Lectures Notes, Manuscript, Dec. 2015. Chapter 7 - Online Learning
↑ Yin, Harold J. Kushner, G. George (2003). Stochastic approximation and recursive algorithms and applications (Second edition. ed.). New York: Springer. pp. 8–12. ISBN 978-0-387-21769-7.
1 2 Bertsekas, D. P. (2011). Incremental gradient, subgradient, and proximal methods for convex optimization: a survey. Optimization for Machine Learning, 85.
↑ Venkatesan, Rajasekar; Meng Joo, Er (2016). "A novel progressive learning technique for multi-class classification". Neurocomputing. 207: 310–321. doi:10.1016/j.neucom.2016.05.006.
1 2 3 4 5 6 Shalev-Shwartz, Shai (2011). "Online Learning and Online Convex Optimization". Foundations and Trends® in Machine Learning. pp. 107–194. doi:10.1561/2200000018. Missing or empty |url= (help)
1 2 3 4 Liang, Percy. "CS229T/STAT231: Statistical Learning Theory (Winter 2015)" (PDF).
↑ Bottou, Léon (1998). "Online Algorithms and Stochastic Approximations". Online Learning and Neural Networks. Cambridge University Press. ISBN 978-0-521-65263-6
↑ Stochastic Approximation Algorithms and Applications, Harold J. Kushner and G. George Yin, New York: Springer-Verlag, 1997. ISBN 0-387-94916-X; 2nd ed., titled Stochastic Approximation and Recursive Algorithms and Applications, 2003, ISBN 0-387-00894-2.

External links

http://onlineprediction.net/, Wiki for On-Line Prediction.

This article is issued from Wikipedia - version of the 11/24/2016. The text is available under the Creative Commons Attribution/Share Alike but additional terms may apply for the media files.