Machine Learning

The fundamental goal of machine learning is to make computer systems that can adapt and learn from their experience.

Machine Learning isn't new; it has been around at least since the 1970s, when the first related algorithms appeared. What has changed is that the explosion in computing power has allowed us to use machine learning to tackle ever-more-complex problems, while the explosion of data being captured and stored has allowed us to apply machine learning to an ever-expanding range of domains.

The modern history of deep learning started in 1943 when the McCulloch-Pitts (MCP) model was introduced and became known as the prototype of artificial neural models

Machine learning focuses more on understanding—measuring or making predictions—while AI is thinking one step further. Once we understand what’s going on, how can we make better decisions? How can we change the way we do things to take advantage of those insights? AI adds a layer of decision making on top of machine learning.

Machine learning employs the following two strategies: supervised learning and unsupervised learning.

In supervised learning, the training set contains data and the correct output of the task with that data. This is like giving a student a set of problems and their solutions and telling that student to figure out how to solve other problems he or she will have to deal with in the future.

Supervised learning includes classification algorithms, which take as input a dataset and the class of each piece of data so that the computer can learn how to classify new data. For example, the input might be a set of past loan applications with an indication of which of them went bad. On the basis of this information, the computer classifies new loan applications. Classification can employ logic regression, classification trees, support vector machines, random forests, artificial neural networks (ANNs), or other algorithms.

Regression algorithms predict a value of an entity's attribute (“regression” here has a wider sense than merely statistical regression). Regression algorithms include linear regression, decision trees, Bayesian networks, fuzzy classification, and ANNs.

In unsupervised learning, the training set contains data but no solutions; the computer must find the solutions on its own. This is like giving a student a set of patterns and asking him or her to figure out the underlying motifs that generated the patterns.

Unsupervised learning includes clustering algorithms, which take as input a dataset covering various dimensions and partition it into clusters satisfying certain criteria. A popular algorithm is k-means clustering, which aims to partition the dataset so that each observation lies closest to the mean of its cluster. Other clustering approaches include hierarchical clustering, Gaussian mixture models, genetic algorithms (in which the computer learns the best way for a task through artificial selection), and ANNs.

Dimensionality reduction algorithms take the initial dataset covering various dimensions and project the data to fewer dimensions. These fewer dimensions try to better capture the data's fundamental aspects. Dimensionality reduction algorithms include principal component analysis, tensor reduction, multidimensional statistics, random projection, and ANNs.

ANNs can be used across the spectrum of machine learning: classification, regression, clustering, and dimensionality reduction.

Innovations in ANN architectures and the availability of cheap computing resources to run ANNs has brought about the burgeoning of deep learning—using big ANNs to perform machine learning.

Algorithms inspired by how the brain works, so-called artificial neural networks, date back to the 1960s; most computer science students still encounter them in introductory AI classes. These networks are formed by connections of artificial "neurons," which are basically just weighted links between nodes in a graph.

The actual network itself does not have any inherent meaning or knowledge. But by subjecting the network to stimuli and reinforcing the links that are used when it makes the correct choices, it is possible to train the network to make choices.

For instance, etwork to a sequence of pictures with simple geometrical shapes but reinforcing it only when the network selects those that contain a circle, it would be possible to teach it to pick only images that depict a circle.

Deep learning is particularly good at general tasks requiring the elicitation of higher-level, abstract concepts from the input data, which is what the many layers of an ANN excel at.

How you represent your data has an intimate connection with the type of algorithm you’re going to use to do the supervised classification. The way the data is formatted can make it very easy for us to find the truth we’re seeking, or it can make it very hard. It’s worth thinking about carefully.