AI ML Terminology

Machine Learning Terminology

Understanding Machine Learning (ML) requires knowledge of key concepts like relationships, labels, features, and different learning approaches. Below is a structured breakdown of these terms.

ML Relationships

Machine learning models identify connections between input variables to generate predictions. In mathematics, a linear equation is often expressed as:

y=mx+c

• y represents the target output
• 𝑚 m is the coefficient (slope)
• x stands for the independent variable
• 𝑐 c is the constant

Similarly, in ML, the relationship is often defined as:

y=b+wx

• y is the expected result
• 𝑤 w represents the weight (importance) of an input
• 𝑥 x denotes the features (inputs)
• 𝑏 b is the bias (offset)

For multiple inputs, the equation extends to:

y=b+w  1 x  1  +w  2  x  2   +w  3 x  3 

This functionality allows ML models to map inputs to outputs efficiently.

ML Labels

A label is the value a model is trained to determine.

• It is the outcome the system aims to predict.
• Labels correspond to the dependent variable in an equation.

Example in ML:

y=b+wx

Where y is the predicted label.

In classification, labels might be categories (e.g., "spam" or "not spam").

For regression, labels are often numeric values (e.g., predicting house prices).

ML Features

Features are the characteristics or attributes used as inputs in a model.

• In linear equations, they represent x values.
• Features influence the predicted result.

Example:

y=b+w  1  x  1  +w  2  x  2  +w  3  x  3 

Where y is the predicted label.

In classification, labels might be categories (e.g., "spam" or "not spam").

For regression, labels are often numeric values (e.g., predicting house prices).

ML Features

Features are the characteristics or attributes used as inputs in a model.

• In linear equations, they represent x values.
• Features influence the predicted result.

Example:

y=b+w  1  x  1  +w  2  x  2  +w  3  x  3 

Where:

• x 1 ,x 2 ,x 3 are different inputs
• w 1 ,w 2 ,w 3 are weights assigned to each input

Features help models learn patterns from data.

ML Models

A model represents the relationship between inputs and outputs.

Every model undergoes three key stages:

• Data Preparation – Gathering and processing relevant information.
• Training Phase – The system learns from existing datasets.
• Inference Process – The trained model is used to make real-world predictions.

ML Training

The purpose of training is to build a model that can answer a specific question (e.g., estimating a car’s resale value).

• Data is fed into the algorithm.
• The system adjusts parameters to enhance accuracy.
• It refines itself by reducing errors.

ML Inference

The purpose of training is to build a model that can answer a specific question (e.g., estimating a car’s resale value).

• Data is fed into the algorithm.
• The system adjusts parameters to enhance accuracy.
• It refines itself by reducing errors.

ML Inference

Inference happens when a trained model is applied to new data.

• The model analyzes input.
• It predicts the correct output.
• This phase is where ML is used in real-world applications (e.g., face recognition, spam filtering).

Phases of ML

Machine Learning progresses through two primary phases:

1. Training Process

• The system learns from labeled data.
• It discovers patterns and optimizes predictions.

2. Inference Process

• The finalized model is applied to new inputs.
• It generates accurate outputs based on prior learning.

Types of ML Approaches

Supervised Learning

A structured method where models learn from labeled examples.

• Uses known data to train models.
• Applied in classification (e.g., recognizing handwritten digits).
• Used for prediction (e.g., forecasting sales trends).

Unsupervised Learning

A pattern-finding method where the system works with unlabeled data.

• Detects relationships without predefined answers.
• Groups similar data (clustering) without external guidance.
• Helps in data segmentation (e.g., customer behavior analysis).

Reinforcement Learning

A technique where an agent learns by trial and error, receiving rewards for positive outcomes.

• It adapts dynamically based on success or failure.
• Used in robotics, gaming, and self-driving cars.

Self-Supervised Learning

A hybrid approach where models generate their own labels from data.

• Unlike traditional unsupervised learning, it focuses on classification and regression tasks.
• Used in language processing (e.g., AI chatbots).

Final Thoughts

These ML principles lay the groundwork for understanding machine intelligence. Whether models are predicting values, recognizing images, or detecting anomalies, relationships, features, and learning types play crucial roles in shaping their accuracy.