The three primary input domains discussed in this module are:

  • Computer Vision
  • Natural Language Processing
  • Statistical Data

Computer Vision

Computer vision uses visual inputs, such as images and video, to interpret meaningful information in artificial intelligence, computer science, and engineering.

Its goal is to emulate human vision, allowing machines to identify, process, and analyze visual information; make predictions and recommendations; and take actions.

Example: Face detection in phone cameras

A phone camera identifies faces in the frame and adjusts its focus and exposure accordingly. The system scans pixel patterns, detects features such as the arrangement of eyes and mouths, and draws a bounding box around each face in real time.

Essentially, computer vision converts image or video input into recognition judgments. It mimics tasks that the human visual system performs automatically, but assembles them piece by piece and without the broader understanding humans use to connect visual information.

How Computer Vision Works

  • Extracts information from the pixels in images and videos
  • Groups visual information to identify patterns

Key Components of Computer Vision

Image Classification

Assigns a category label to an entire image. In practice, the result is often a probability distribution, such as 50% husky, 12% wolf, and 3% Inuit dog.

Common examples include animal-species classification and facial recognition.

Object Detection

Identifies and locates objects from different categories within an image.

Examples include traffic-sign detection and person detection.

Image Segmentation

Divides an image into regions based on its content.

Examples include tumor detection in a brain MRI and lane detection for self-driving cars.

  • Semantic segmentation: Assigns a category label to every pixel in an image, such as labeling the pixels representing a chair in a living-room photograph.
  • Instance segmentation: Identifies objects at the pixel level while distinguishing each individual object, including objects from the same class. For example, it gives each dog in an image its own pixel-precise mask: dog 1, dog 2, and dog 3.

Image Captioning

Generates a natural-language description of the content of an image.

Visual Question Answering (VQA)

Answers natural-language questions about the content of an image.

Action Recognition

Identifies and classifies actions taking place in a video.

Examples include sports analytics and detecting suspicious human behavior in security footage.

Pose Estimation

Estimates the position and orientation of objects or people in an image or video, such as representing the human body as a skeleton.

Examples include:

  • Analyzing the biomechanics of pitching
  • Supporting Video Assistant Referee decisions in soccer

Image Enhancement and Restoration

Improves image quality by removing artifacts, noise, or blur; restoring missing content; or increasing visual quality.

Example: Google Pixel’s Magic Eraser.

Natural Language Processing

Natural Language Processing (NLP) is a field of AI concerned with interactions between computers and human languages. It enables computers to understand, manipulate, and generate languages spoken or written by humans.

NLP is used to build machines that understand text or voice data and respond with text or speech. It helps bridge the gap between natural human communication and the ways computers process information.

Key Applications of NLP

Text Classification

Assigns a category label to a piece of text.

Example: Gmail labeling an incoming email as spam or not spam.

Sentiment Analysis

Determines whether the emotional tone of a text is positive, negative, or neutral. Sentiment analysis is a specialized form of text classification.

Example: Classifying the product review “This product broke in two days” as negative.

Machine Translation

Automatically converts text from one language to another.

Example: Google Translate converting “Bonjour, comment ça va?” to “Hello, how are you?”

Text Summarization

Condenses a long document into a shorter version that preserves its key information.

Example: Generating a three-sentence abstract from a 20-page research paper.

Chatbots and Text Generation

Produce new, coherent text in response to an input, often in a conversational format.

Example: ChatGPT or Claude answering a question or drafting an email from a prompt.

Autocorrect

Detects and corrects spelling or typing errors using language models to predict the intended words.

Example: A phone changing “teh meetign” to “the meeting.”

Speech Recognition

Converts spoken audio into a form that a machine can understand. It is the broad field concerned with voice input.

Example: Siri understanding the command “Set a timer for 10 minutes.”

Text-to-Speech (TTS)

Converts written text into spoken audio.

Examples include:

  • A screen reader reading a webpage aloud
  • A GPS saying, “Turn left in 500 feet”

Speech-to-Text (STT)

Transcribes spoken audio into written text. It describes the transcript-producing output of speech recognition.

Examples include:

  • Zoom generating live captions during a meeting
  • Dictating a text message

Note: Speech recognition and speech-to-text are often used interchangeably. The distinction is that STT specifically refers to producing a written transcript.

Statistical Data

Statistical data is information collected, organized, and analyzed to describe, understand, and draw conclusions about a phenomenon or population.

It often begins as raw data and is transformed into structured data, such as tables, for statistical analysis.

Key Properties

  • Measurable: Structured and quantifiable information, such as numbers, percentages, and other measurable attributes—often called features
  • Contextual: Collected for a clear purpose, such as measuring the effectiveness of a treatment
  • Meaningful: Gathered to identify patterns, trends, and relationships and to support informed decisions
  • Representative: Drawn from a suitably chosen sample that reflects the characteristics of a population

Nature of Predictions

Classification

When a task predicts a category, such as a movie genre or flower species, it is called a classification problem. Its output is a categorical label.

Regression

When a task predicts a numerical value, such as a housing price or stock price, it is called a regression problem. Its output is a numerical label.

Examples of Statistical Data in AI

Data source Example features Prediction or use
Medical records Blood pressure, cholesterol, age, BMI Disease-risk prediction
Credit data Income, debt ratio, payment history Credit scoring
Weather measurements Temperature, humidity, pressure readings Weather forecasting
Sales figures Historical monthly revenue, seasonality statistics Demand prediction
Machine sensors Vibration and temperature averages Predictive maintenance

The common thread is the use of structured numerical data—such as counts, averages, rates, and measurements—as input for models including regression models, decision trees, and neural networks.