Unsupervised learning, is paradigm in machine learning where, in contrast to supervised learning and semi-supervised learning, algorithms learn patterns exclusively from unlabeled data.
Part of a series on |
Machine learning and data mining |
---|
Neural network tasks are often categorized as discriminative (recognition) or generative (imagination). Often but not always, discriminative tasks use supervised methods and generative tasks use unsupervised (see Venn diagram); however, the separation is very hazy. For example, object recognition favors supervised learning but unsupervised learning can also cluster objects into groups. Furthermore, as progress marches onward some tasks employ both methods, and some tasks swing from one to another. For example, image recognition started off as heavily supervised, but became hybrid by employing unsupervised pre-training, and then moved towards supervision again with the advent of dropout, ReLU, and adaptive learning rates.
During the learning phase, an unsupervised network tries to mimic the data it's given and uses the error in its mimicked output to correct itself (i.e. correct its weights and biases). Sometimes the error is expressed as a low probability that the erroneous output occurs, or it might be expressed as an unstable high energy state in the network.
In contrast to supervised methods' dominant use of backpropagation, unsupervised learning also employs other methods including: Hopfield learning rule, Boltzmann learning rule, Contrastive Divergence, Wake Sleep, Variational Inference, Maximum Likelihood, Maximum A Posteriori, Gibbs Sampling, and backpropagating reconstruction errors or hidden state reparameterizations. See the table below for more details.
An energy function is a macroscopic measure of a network's activation state. In Boltzmann machines, it plays the role of the Cost function. This analogy with physics is inspired by Ludwig Boltzmann's analysis of a gas' macroscopic energy from the microscopic probabilities of particle motion , where k is the Boltzmann constant and T is temperature. In the RBM network the relation is ,^{[1]} where and vary over every possible activation pattern and . To be more precise, , where is an activation pattern of all neurons (visible and hidden). Hence, early neural networks bear the name Boltzmann Machine. Paul Smolensky calls the Harmony. A network seeks low energy which is high Harmony.
This table shows connection diagrams of various unsupervised networks, the details of which will be given in the section Comparison of Networks. Circles are neurons and edges between them are connection weights. As network design changes, features are added on to enable new capabilities or removed to make learning faster. For instance, neurons change between deterministic (Hopfield) and stochastic (Boltzmann) to allow robust output, weights are removed within a layer (RBM) to hasten learning, or connections are allowed to become asymmetric (Helmholtz).
Hopfield | Boltzmann | RBM | Stacked Boltzmann |
---|---|---|---|
Helmholtz | Autoencoder | VAE |
---|---|---|
Of the networks bearing people's names, only Hopfield worked directly with neural networks. Boltzmann and Helmholtz came before artificial neural networks, but their work in physics and physiology inspired the analytical methods that were used.
1969 | Perceptrons by Minsky & Papert shows a perceptron without hidden layers fails on XOR |
1970s | (approximate dates) First AI winter |
1974 | Ising magnetic model proposed by WA Little | for cognition
1980 | Fukushima introduces the neocognitron, which is later called a convolutional neural network. It is mostly used in SL, but deserves a mention here. |
1982 | Ising variant Hopfield net described as CAMs and classifiers by John Hopfield. |
1983 | Ising variant Boltzmann machine with probabilistic neurons described by Hinton & Sejnowski following Sherington & Kirkpatrick's 1975 work. |
1986 | Paul Smolensky publishes Harmony Theory, which is an RBM with practically the same Boltzmann energy function. Smolensky did not give a practical training scheme. Hinton did in mid-2000s. |
1995 | Schmidthuber introduces the LSTM neuron for languages. |
1995 | Dayan & Hinton introduces Helmholtz machine |
1995-2005 | (approximate dates) Second AI winter |
2013 | Kingma, Rezende, & co. introduced Variational Autoencoders as Bayesian graphical probability network, with neural nets as components. |
Here, we highlight some characteristics of select networks. The details of each are given in the comparison table below.
Hopfield | Boltzmann | RBM | Stacked RBM | Helmholtz | Autoencoder | VAE | |
---|---|---|---|---|---|---|---|
Usage & notables | CAM, traveling salesman problem | CAM. The freedom of connections makes this network difficult to analyze. | pattern recognition. used in MNIST digits and speech. | recognition & imagination. trained with unsupervised pre-training and/or supervised fine tuning. | imagination, mimicry | language: creative writing, translation. vision: enhancing blurry images | generate realistic data |
Neuron | deterministic binary state. Activation = { 0 (or -1) if x is negative, 1 otherwise } | stochastic binary Hopfield neuron | ← same. (extended to real-valued in mid 2000s) | ← same | ← same | language: LSTM. vision: local receptive fields. usually real valued relu activation. | middle layer neurons encode means & variances for Gaussians. In run mode (inference), the output of the middle layer are sampled values from the Gaussians. |
Connections | 1-layer with symmetric weights. No self-connections. | 2-layers. 1-hidden & 1-visible. symmetric weights. | ← same. no lateral connections within a layer. |
top layer is undirected, symmetric. other layers are 2-way, asymmetric. | 3-layers: asymmetric weights. 2 networks combined into 1. | 3-layers. The input is considered a layer even though it has no inbound weights. recurrent layers for NLP. feedforward convolutions for vision. input & output have the same neuron counts. | 3-layers: input, encoder, distribution sampler decoder. the sampler is not considered a layer |
Inference & energy | Energy is given by Gibbs probability measure : | ← same | ← same | minimize KL divergence | inference is only feed-forward. previous UL networks ran forwards AND backwards | minimize error = reconstruction error - KLD | |
Training | Δw_{ij} = s_{i}*s_{j}, for +1/-1 neuron | Δw_{ij} = e*(p_{ij} - p'_{ij}). This is derived from minimizing KLD. e = learning rate, p' = predicted and p = actual distribution. | Δw_{ij} = e*( < v_{i} h_{j} >_{data} - < v_{i} h_{j} >_{equilibrium} ). This is a form of contrastive divergence w/ Gibbs Sampling. "<>" are expectations. | ← similar. train 1-layer at a time. approximate equilibrium state with a 3-segment pass. no back propagation. | wake-sleep 2 phase training | back propagate the reconstruction error | reparameterize hidden state for backprop |
Strength | resembles physical systems so it inherits their equations | ← same. hidden neurons act as internal representatation of the external world | faster more practical training scheme than Boltzmann machines | trains quickly. gives hierarchical layer of features | mildly anatomical. analyzable w/ information theory & statistical mechanics | ||
Weakness | hard to train due to lateral connections | equilibrium requires too many iterations | integer & real-valued neurons are more complicated. |
The classical example of unsupervised learning in the study of neural networks is Donald Hebb's principle, that is, neurons that fire together wire together.^{[4]} In Hebbian learning, the connection is reinforced irrespective of an error, but is exclusively a function of the coincidence between action potentials between the two neurons.^{[5]} A similar version that modifies synaptic weights takes into account the time between the action potentials (spike-timing-dependent plasticity or STDP). Hebbian Learning has been hypothesized to underlie a range of cognitive functions, such as pattern recognition and experiential learning.
Among neural network models, the self-organizing map (SOM) and adaptive resonance theory (ART) are commonly used in unsupervised learning algorithms. The SOM is a topographic organization in which nearby locations in the map represent inputs with similar properties. The ART model allows the number of clusters to vary with problem size and lets the user control the degree of similarity between members of the same clusters by means of a user-defined constant called the vigilance parameter. ART networks are used for many pattern recognition tasks, such as automatic target recognition and seismic signal processing.^{[6]}
Two of the main methods used in unsupervised learning are principal component and cluster analysis. Cluster analysis is used in unsupervised learning to group, or segment, datasets with shared attributes in order to extrapolate algorithmic relationships.^{[7]} Cluster analysis is a branch of machine learning that groups the data that has not been labelled, classified or categorized. Instead of responding to feedback, cluster analysis identifies commonalities in the data and reacts based on the presence or absence of such commonalities in each new piece of data. This approach helps detect anomalous data points that do not fit into either group.
A central application of unsupervised learning is in the field of density estimation in statistics,^{[8]} though unsupervised learning encompasses many other domains involving summarizing and explaining data features. It can be contrasted with supervised learning by saying that whereas supervised learning intends to infer a conditional probability distribution conditioned on the label of input data; unsupervised learning intends to infer an a priori probability distribution .
Some of the most common algorithms used in unsupervised learning include: (1) Clustering, (2) Anomaly detection, (3) Approaches for learning latent variable models. Each approach uses several methods as follows:
One of the statistical approaches for unsupervised learning is the method of moments. In the method of moments, the unknown parameters (of interest) in the model are related to the moments of one or more random variables, and thus, these unknown parameters can be estimated given the moments. The moments are usually estimated from samples empirically. The basic moments are first and second order moments. For a random vector, the first order moment is the mean vector, and the second order moment is the covariance matrix (when the mean is zero). Higher order moments are usually represented using tensors which are the generalization of matrices to higher orders as multi-dimensional arrays.
In particular, the method of moments is shown to be effective in learning the parameters of latent variable models. Latent variable models are statistical models where in addition to the observed variables, a set of latent variables also exists which is not observed. A highly practical example of latent variable models in machine learning is the topic modeling which is a statistical model for generating the words (observed variables) in the document based on the topic (latent variable) of the document. In the topic modeling, the words in the document are generated according to different statistical parameters when the topic of the document is changed. It is shown that method of moments (tensor decomposition techniques) consistently recover the parameters of a large class of latent variable models under some assumptions.^{[11]}
The Expectation–maximization algorithm (EM) is also one of the most practical methods for learning latent variable models. However, it can get stuck in local optima, and it is not guaranteed that the algorithm will converge to the true unknown parameters of the model. In contrast, for the method of moments, the global convergence is guaranteed under some conditions.