Intro to AI Quiz 2 VOCAB

Neural Networks

Inspired by the structure and function of the human brain; attempt to mimic how biological neurons process and transmit information

Input Layer

Receives raw data

Hidden Layers

Process and transform data by finding patterns

Output layer

Produces the final result or classification

Neurons

Small computing units that combine inputs, weights, and activation functions to make simple decisions.

Weights

Connect neurons and determine how important each input is; Adjusted during training to improve accuracy. Represent the importance of each input.

Bias

allows flexibility by shifting the activation threshold up or down; Helps fine-tune neuron decisions so the model doesn’t always pass through the origin (adds flexibility).

Input Layer function

data intake

Hidden layer function

pattern recognition

output layer function

final prediction

artificial neuron

A mathematical model that takes inputs, multiplies them by weights, adds bias, and applies an activation function

shallow network

has no or few hidden layers

deep learning

Neural networks with many hidden layers, capable of learning complex features.

Activation functions

determine whether a neuron fires (ex. Sigmoid, ReLU, and Tanh); introduce nonlinearity, allowing networks to learn complex patterns

forward propagation

Data flows from input → hidden → output layer to generate a prediction.

loss calculation

Compares predicted output to the true value to measure error

backpropogation

Adjusts weights and biases in the opposite direction to minimize error — the key learning process.

training data

Input-output pairs used to “teach” the model.

Supervised Learning

The model learns with labeled examples (e.g., “This is a cat”).

Updating Weights

 After each training example, the network slightly changes weights to reduce future errors.

Neural Network Learning from Errors

Improve by identifying these, adjusting parameters, and iterating over many examples; allows predictions to become more accurate over time

Image recognition

Detects edges, shapes, and complex objects.

Speech recognition

Converts spoken language into text.

spam detection

Classifies emails as spam or not spam.

Autonomous Vehicles

Uses deep learning for perception and decision-making.

Medical diagnosis

Identifies disease patterns in scans or data

Multiple hidden layers

allow network to recognize complex, abstract features (e.g. from pixels —> edges —> shapes —> faces); deep learning models outperform shallow networks on complex tasks

overfitting

model memorizes training data and fails on new data; prevented using regularization, dropout, and more diverse training data

underfitting

Model is too simple and can’t learn the patterns.

high computation needs

Deep learning requires strong hardware (GPUs).

Ethical issues

Data bias can lead to unfair or inaccurate outcomes.

Ethical concerns

Bias in training data, privacy issues, and accountability for AI decisions

GPUs and TPUS

accelerate deep learning computations