Introduction to Machine Learning

• Machine Learning (ML) programs computers so they can learn from data

What Is Learning?

• A computer program learns from experience E on task T, with performance measure P, if its performance on T improves with experience E

Why Use Machine Learning?

• ML can be used instead of traditional programming to create solutions
• First, the structure of typical spam is considered, such as words and phrases
• Then, a detection algorithm is written for each noticed pattern to flag spam emails
• The program is tested, repeating steps until satisfactory
• Machine learning algorithm can simplify code and improve performance
• It can find solutions to complex problems for which there is no good traditional approach
• Machine Learning is great for fluctuating environments to adapt to new data
• It helps in getting insights about complex problems and large amounts of data

Types of Machine Learning Systems

• Machine Learning systems are useful to classify in broad categories based on training, learning incrementally, and comparing data points
• Systems are classified based on whether they are trained with human supervision or not
• Classifications include supervised, unsupervised, semi-supervised, and reinforcement learning

Supervised Learning

• Training data fed to the algorithm includes the desired solutions
• Classification is a typical supervised learning task, which includes training with example emails to classify new emails
• Regression is a typical task to predict a target numeric value, given a set of features called predictors
• Some supervised learning algorithms: k-Nearest Neighbors, Linear Regression, Logistic Regression, Support Vector Machines (SVMs), Decision Trees and Random Forests and Neural networks2

Unsupervised Learning

• Training data is unlabeled
• Unsupervised algorithms: K-Means, DBSCAN, Hierarchical Cluster Analysis (HCA)
• Anomaly detection and novelty detection: One-class SVM and Isolation Forest

Semi-Supervised Learning

• Algorithms deal with partially labeled training data composed of unlabeled and a little bit of labeled data

Reinforcement Learning

• The learning system, an agent, observes the environment, selects, performs actions, and gets rewards (or penalties) in return
• The agent learns the best strategy (policy) to get the most reward over time
• A policy defines when the agent should choose when it is in a given solution

Batch and Online Learning

• Another criterion used is whether or not the system can learn incrementally from a stream of incoming data

Batch Learning

• Cannot learn incrementally; it must be trained with all available data
• It takes a lot of time and computing resources, so it is typically done offline
• The system is trained, launched into production, and runs without learning anymore, just applying what it has learned
• It's also known as offline learning

Online Learning

• Trains incrementally by feeding data instances sequentially, individually or by small groups called mini-batches
• Each learning step is fast and cheap, so the system can learn about new data on the fly, as it arrives

Instance Based vs Model Based Learning

• Machine Learning tasks are about making predictions and generalizing to examples never seen before
• Instance-based learning generalizes to new cases by comparing them to the learned instances
• Model-based learning builds a model from a set of examples

Main Challenges of ML

• Selecting a learning algorithm and training it on some data can cause "bad algorithm" and "bad data" results
• The challenges include insufficient quantity of training data, nonrepresentative training data, and poor quality data
• Other challenges are irrelevant features, overfitting the training data, underfitting the training data, stepping back, and testing/validating