COMP1008 Introduction to AI and Machine Learning

Vocabulary-style flashcards covering the basics of AI, machine learning types, milestones, and data processing techniques based on Professor Rong Qu's lecture notes.

Artificial Intelligence (AI)

Coined in 1956, defined as the science and engineering of making intelligent machines; it is a branch of computer science dealing with intelligent behavior, learning, and adaptation of machines.

Weak AI

AI that focuses on specific tasks and is limited to those tasks; it exists today and can be applied to various settings.

Strong AI

A broader, general-purpose, self-aware AI with the ability to learn like a human; currently considered purely theoretical.

Top-down approach

A method that involves pre-processes to get knowledge from the real world into a program to solve problems, often using hard-coded instructions (deduction).

Bottom-up approach

A method with no hard-wired instructions where models learn from sets of data to discover concepts and patterns by themselves (induction).

AI winter

A period of around 15 years starting in the 1970s when Artificial Neural Networks (ANNs) could not implement basic functions, leading to reduced interest and progress.

Turing Test

An operational definition of intelligence proposed in 1950 where a machine is considered intelligent if a person picks the machine system 30% of the time after asking questions.

Chinese Room Experiment

A behavioral definition of intelligence demonstrating a system that acts as if it understands Chinese through rule processing, despite having no actual understanding.

Machine Learning (Mitchell's Definition)

A program learns from experience (E) with respect to some class of tasks (T) and performance measure (P) if its performance at tasks in T, as measured by P, improves with experience E.

Data mining

The exploration and analysis of large quantities of data to discover valid, novel, useful, and ultimately understandable patterns.

Supervised learning

A type of machine learning that uses labeled training data to train a model to match an input to a desired output.

Unsupervised learning

A type of machine learning that uses data points without labels to group data or identify trends, such as K-means or the Apriori algorithm.

Reinforcement learning

A method where an agent takes feedback from an environment and is rewarded based on actions taken, gradually learning a policy via trial-and-error.

Imputation

A data cleaning process used to handle missing data by replacing the missing values.

K-means clustering

An unsupervised learning algorithm that initializes K cluster centroids randomly and iteratively assigns data points to the nearest centroid to update the mean.

Classification

A supervised learning task used when labels are categories, such as 'good' or 'bad'.

Regression

A supervised learning task used when labels are continuous data, such as integers and probabilities.

Polynomial linear regression

A regression model represented by the function $G(x) = b_0 + b_1 \times x + b_2 \times x^2$.

Underfitting

A situation, often occurring with simple linear regression, where a model does not fully represent the underlying data.

Overfitting

A situation where a model overrepresents the training data, reflecting randomness and granularity rather than the general pattern.

Decision Tree

A recursive data structure where root nodes and branches are used to group feature values through a series of compared decisions.

Confusion Matrix

A matrix used to compare actual and predicted outputs of test instances to reveal false positives and false negatives.

ANN (Artificial Neural Network)

A subset of machine learning described as a 'black box' because it is very difficult to visualize or understand the internal processing compared to other models.