HS1502 Conceptual Introduction to Machine Learning - Vocabulary Flashcards

Revision

Artificial Intelligence (AI)

The mimicking of biological (often human) intelligence using algorithms implemented on machines; broader than automation; has roots in neuroscience, logic, psychology, and linguistics.

Machine Learning (ML)

An approach to AI where the performance of a computer program on a task improves with additional “experience” (data); learns from data and can adapt with more data, without necessarily a predefined reasoning system.

Deep Learning

A subset of ML that uses deep artificial neural networks to learn representations (features) from data; simple representations are combined across layers to form complex concepts.

Artificial Neural Network (ANN)

A network of interconnected artificial neurons; each neuron is a mathematical function that maps inputs to outputs.

Perceptron

The first artificial neuron model (1957) developed by Rosenblatt; an early mathematical model of biological learning used in early neural networks.

Computational Neuroscience

An active field studying how biological intelligence works, providing early motivation for AI by viewing intelligence as computational processes.

Knowledge-based Systems

AI programs that use a knowledge base and a reasoning system to mimic intelligence; they do not learn from data, though new data can be provided.

Algorithm

A central concept in AI: a process or set of rules to be followed to perform calculations or problem solving.

Artificial General Intelligence (AGI)

AI that matches or surpasses human-level general intelligence across a wide range of tasks and domains.

Artificial Narrow Intelligence (ANI)

AI systems designed to perform specific tasks; the current mainstream form of AI, not general across domains.

Turing Test

A test proposed by Alan Turing to assess whether a machine can imitate human conversation well enough to be indistinguishable from a human.

AlexNet

A deep neural network that won an image recognition contest in 2012 with about 85% accuracy, catalyzing the deep learning revolution.

GPUs (General-Purpose Computing on GPUs)

Hardware that enables efficient training of large neural networks via parallel processing, fueling the deep learning era.

Convolutional Neural Networks (CNNs)

A class of neural networks particularly effective for processing visual data through convolutional layers.

Recurrent Neural Networks (RNNs)

Neural networks designed to handle sequential data by maintaining a hidden state across time steps.

Transformers

A neural network architecture based on attention mechanisms, widely used in natural language processing and beyond.

AlphaGo

Go-playing AI that defeated world champion Ke Jie in 2017, illustrating the power of deep learning plus search techniques.

MuZero

A game-playing algorithm that dominates Go, chess, shogi, and arcade games by planning and learning without a fixed game model.

ChatGPT

A chatbot released in 2022 using a large language model (LLM); broad knowledge base and versatile, human-like responses.

Large Language Models (LLMs)

AI models trained on vast text data to understand and generate human-like language; capable across tasks but with reliability and context limits.

Deepfakes

AI-generated media that convincingly mimics real people or events, posing societal and ethical risks.

Big Data

Extremely large datasets used to train modern ML/DL models, enabling better performance with more diverse data.

Deep Representation Learning

Learning hierarchical representations (features) from data, enabling models to build complex concepts from simple ones.

Feature Learning

The process by which models automatically discover useful representations (features) from raw data.

Moore’s Law

The observation that computing power grows exponentially with a doubling time of about 2 years, influencing AI progress.

Clinical Decision Support Systems (CDSS)

Knowledge-based AI systems that use patient data and reasoning to suggest clinical decisions to physicians.

DL Revolution (2010s–present)

The period when deep learning became dominant due to GPUs, big data, and large neural networks.

Natural Language Processing (NLP)

A subfield of AI focused on enabling computers to understand, interpret, and generate human language.

Ethical and Societal Risks of AI

Concerns such as carbon footprint, privacy and data misuse, misinformation, and concentration of power.