Machine learning: creating the big picture by leveraging data
What you will learn:
- Different types of machine learning.
- Learn about the types of supervised and unsupervised machine learning approaches.
Machine learning (ML) is a data analysis method that automates the creation of analytical models. It is a branch of artificial intelligence (AI) based on the idea that systems can learn from data, identify patterns and make decisions with minimal human intervention. ML algorithms build a model based on sample data, or training data, to make predictions or decisions without being programmed to perform a given task.
These algorithms are used in a myriad of applications, including medicine, autonomous vehicles, voice recognition, and machine vision, where it is difficult or impossible to use traditional algorithms to perform required tasks. It’s also behind chatbots and predictive text, language translation apps, and even Netflix recommended shows and movies.
When companies use artificial intelligence programs, chances are they are using machine learning. So much so that the terms are often used interchangeably and sometimes ambiguously as an all-encompassing form of AI. This subfield aims to create computer models that exhibit intelligent behaviors similar to those of humans, i.e. they can recognize a visual scene, understand text written in natural language, or perform an action in the real world.
Forms of machine learning
ML is related to computational statistics, which focuses on making predictions using computers, but not all ML is statistical learning. Some ML implementations use data and neural networks in a way that mimics the functioning of a biological brain.
The study of mathematical optimization provides methods, theory, and areas of application to ML. Data mining is another related field of study, focusing on the exploratory analysis of data through unsupervised learning.
To this end, learning algorithms work on the basis that the strategies, algorithms, and interpretations have worked well in the past, so they will likely continue to work well in the future. These inferences can be obvious, such as “since the sky is blue today, it will most likely be blue tomorrow”.
They can also be nuanced, which means that while the platform may be the same, there may be subtle differences within the subset. For example, if X number of families have geographically separated species with different color variants, there is a good chance that several Y variants exist.
Machine learning uses a decision-making process that produces results based on input data, which may or may not be labeled. Most are equipped with an error function that evaluates the model’s prediction.
If there are known examples, an error function can perform a comparison to assess the accuracy of the model. If the model can better fit the data points in the training set, the weights are adjusted to reduce the differences between the known example and the model estimate. The algorithm will repeat the evaluation and optimization process, updating the weights on its own until a certain level of accuracy is achieved.
The methods (see figure above) used to achieve this specific result fall into four broad categories:
Supervised learning is defined by its use of labeled data sets to train algorithms that classify data or accurately predict outcomes. The learning algorithm receives a set of inputs and the corresponding correct outputs, and the algorithm learns by comparing its actual output with correct outputs to find errors. It then modifies the model accordingly. A cross-validation process is then used to ensure that the model avoids overfitting or underfitting.
Supervised learning helps organizations solve a variety of real-world problems at scale, such as classifying spam emails into a separate folder in your inbox. Some methods used in supervised learning include neural networks, naive arrays, linear regression, logistic regression, random forest, support vector machines (SVM), etc.
Unsupervised learning is used against data without historical labels, which means that the system does not receive the correct answer and the algorithm must understand what is displayed. The goal is to explore the data and find a hidden structure or pattern within. This method works well on transactional data.
For example, it can identify customer segments with similar attributes which can then be treated similarly in marketing campaigns. Or it can find the key attributes that separate customer segments from each other.
Popular techniques include self-organizing maps, nearest neighbor mapping, k-means clustering, and singular value decomposition. These algorithms are also used to segment text topics, recommend items, and identify outliers. In addition to this, they are used to reduce the number of features in a model through the process of dimensionality reduction, principal component analysis (PCA), and singular value decomposition (SVD). Other algorithms applied in unsupervised learning include neural networks, probabilistic clustering methods, etc.
This approach to ML offers a happy medium between supervised and unsupervised methods. During training, it uses a smaller labeled dataset to guide classification and feature extraction from a larger, unlabeled dataset.
This type of learning can be used with methods such as classification, regression, and prediction, and can solve the problem of not having enough labeled data (or not being able to afford to label enough data) to train a supervised learning algorithm. It is also useful when the cost associated with labeling is too high to allow for a fully labeled training process. Examples of semi-supervised learning include facial and object recognition.
Reinforcement learning is often associated with robotics, autonomous vehicles, games and navigation. This method allows the algorithm to discover, through trial and error, which actions produce the largest rewards.
Three main components are associated with this type of learning: the agent (the learner or the decision maker), the environment (everything with which the agent interacts) and the actions (what the agent can do). The goal is for the agent to choose actions that maximize the expected reward over a given time frame. The agent can achieve the goal quickly by following a good policy. Thus, the goal of reinforcement learning is to learn the best policy.
Dimensionality reduction is the process of reducing the number of features in a dataset. Often there are too many variables to deal with in ML tasks, such as regression or classification. These variables are also called characteristics: the greater the number of characteristics, the more difficult it is to model them. Additionally, some of these features may be redundant, adding unnecessary noise to the dataset.
Dimensionality reduction reduces the number of random variables considered by collecting a set of principal variables, which can then be divided into feature selection and feature extraction.
Many real-world applications take advantage of machine learning, including artificial neural networks (ANNs), which are modeled after their biological counterparts. These consist of thousands or millions of processing nodes that are densely interconnected to handle many tasks, including speech recognition/translation, gaming, social networking, medical diagnostics, etc.
With Facebook, for example, ML customizes how a member’s feed is delivered. If the member regularly stops to read messages from certain groups, they will prioritize those activities earlier in the stream.
Additionally, ML is used in speech applications, including text-to-speech, which uses natural language processing (NLP) to convert human speech into text. It can also be found with digital assistants such as Siri and Alexa, which use voice recognition for app interaction. Automated customer service, recommendation engines, computer vision, climate science, and even agriculture are among the many other applications.