Introduction to Machine Learning

Machine Learning Definition

• Machine Learning (ML) involves programming computers to learn from data.
• ML gives computers the ability to learn without explicit programming.
• A computer program learns from experience E with respect to a task T and performance measure P if its performance on T improves with experience E.

Why Use Machine Learning

• Traditional approaches to problem-solving involve studying a problem, writing rules, analyzing errors, and evaluating.
• ML involves studying the problem, training an ML algorithm with data, and evaluating the solution.
• ML systems can adapt to changes automatically.
• With enough data, ML can uncover insights into problems that are difficult for humans to discern.
• ML is valuable when existing solutions require hand-tuning, for complex problems lacking good traditional solutions, in fluctuating environments, and for gaining insights from complex data.

Types of Machine Learning Systems

• ML systems can be classified based on whether they are trained with human supervision, their learning method, and their approach to generalization.
• Training methods include supervised, unsupervised, semi-supervised, and reinforcement learning.
• Learning method include online versus batch learning
• Generalization approach includes instance-based versus model-based learning.

Supervised Learning

• Involves training data that includes desired solutions or labels.
• A typical supervised learning task is classification, exemplified by spam filters.
• Regression is another typical task where the system predicts a target numeric value given predictors.
• Important supervised learning algorithms include k-Nearest Neighbors, Linear Regression, Logistic Regression, Support Vector Machines (SVMs), Decision Trees and Random Forests, and Neural networks.

Unsupervised Learning

• Utilizes unlabeled training data, where the system learns without a "teacher".
• Important unsupervised learning algorithms include:
  • Clustering (K-Means, DBSCAN, Hierarchical Cluster Analysis)
  • Anomaly detection and novelty detection (One-class SVM, Isolation Forest)

Semi-Supervised Learning

• Is where algorithms deal with partially labeled training data, typically a lot of unlabeled data and a little bit of labeled data.

Reinforcement Learning

• In reinforcement learning, an agent observes the environment, selects, and performs actions, receiving rewards or penalties.
• The agent learns a policy, which defines the best strategy to maximize rewards over time.

Batch Learning

• Batch learning involves training the system using all available data at once.
• Batch learning is typically done offline due to its time and resource requirements.
• The system learns then launches to production and applies what it has learned; called offline learning.

Online Learning

• Online learning trains incrementally by feeding data instances sequentially or in mini-batches.
• Each learning step is fast and cheap, allowing the system to learn about new data on the fly.

Instance-Based Learning vs. Model-Based Learning

• Instance-based learning involves the system learning examples by heart and generalizing to new cases by comparing them to learned examples.
• Model-based learning involves building a model from a set of examples and then using that model to make predictions.

Main Challenges of Machine Learning

• Selecting a learning algorithm and training it on data can run into two main problems: “bad algorithm” and “bad data”.
• This can lead to bad data problems like:
• Insufficient Quantity of Training Data
• Nonrepresentative Training Data
• Poor Quality Data
• Irrelevant Features
• Overfitting the Training Data
• Underfitting the Training Data