Network Architecture

62

What the inside of an AI model looks like. Neurons, layers, weights, activation functions, RNNs, LSTMs, parameters: what people actually mean when they say a model is "big" or "deep".

Level Intro 15 Basics 30 Advanced 17

Activation Function

A nonlinear function that bends a neuron's summed value

Basics Attention

A way of deciding how much weight to give each input

LLM Basics Autoencoder

A structure that learns by shrinking data down and rebuilding it

Generative Basics Autoregressive

Rereads what it just wrote, then adds one more piece

Advanced Backpropagation

Tracing error backward to split blame across the network

Training Basics Bias Term

The default amount that's added no matter what the input is

Basics Black Box

An AI's decision process when the inside can't be seen

Ethics Intro Browser AI

AI that runs inside a web page with nothing installed

Tools Intro CNNConvolutional Neural Network

A neural network built to spot patterns in a photo, layer by layer

Vision Basics Compute

The total amount of calculation it takes to train or run a model

Basics Convolution

Sliding a small tile across a photo or other data to score each spot

Vision Basics Deep Learning

Machine learning that stacks processing steps many layers deep

Basics Intro Depth vs Width

How many layers to stack versus how wide to make each one

Basics DQNDeep Q-Network

A neural network standing in for a table of expected scores

Training Advanced Dropout

Randomly resting part of a model during training

Training Basics Encoder-Decoder

A two-part design that reads with one half, writes with the other

LLM Basics Environmental Cost

The electricity and resources it takes to build and run AI

Ethics Intro Feature Map

A picture that lays scan results out in their original spots

Vision Advanced Feed-Forward

A neural network structure where signals only flow forward

Basics Filter / Kernel

A small value tile that carries what to look for

Vision Basics GPU

A device that processes the same calculation many times at once

Basics Basics Gradient

The signal that says which way, and how far, to reduce error

Basics Hidden Layer

The middle layers between input and output that build features

The time between sending a request and getting an answer back

Tools Intro Layer

A basic computing stage of neurons that share the same input

Intro Layer Normalization

A device that recalibrates the numbers' scale at every layer

Advanced Local LLM

Running an AI model directly on your own computer

Tools Intro Logits

The raw score each candidate gets before becoming a probability

LLM Advanced LSTM

A recurrent network that uses gates to choose what to keep

Advanced Matrix Multiplication

Pairing up rows and columns, multiplying each pair, then summing

Basics Mixture of Experts

A model structure that switches on only the branches it needs

LLM Advanced Model Architecture

The blueprint that says how a model's layers stack and connect

Intro Model Serving

Putting a finished model to work answering everyone's requests

Tools Basics Model Size

How many numbers a model is carrying

Intro Multi-Head Attention

The same scene watched by several viewpoints at once

LLM Advanced Multilayer Perceptron

A basic neural network made of simple deciders stacked in layers

Basics Neural Network

Simple calculating units chained together in layers

Basics Intro Neuron

A tiny computing unit that takes in many values and outputs one

Intro On-Device AI

AI that runs directly on the device itself

Tools Intro Open Weights

A model that hands out only the finished trained numbers

Tools Basics Parameter

A tuning dial inside a model whose value training decides

Intro Perceptron

An early neural network that weighs inputs and fixes its own weights

Shrinking a picture by keeping one representative value per region

Vision Basics Positional Encoding

A way of tagging each spot with its place in line

LLM Advanced Pruning

Cutting connections that add little to make a model lighter

Advanced Quantization

Coarsening the scale numbers are written on to make a model lighter

Blocks negative inputs at zero, passes positive ones through

Basics Residual Connection

A shortcut skipping a layer, carrying the original value through

Advanced RNNRecurrent Neural Network

A network that carries forward a running summary, word by word

Basics Self-Attention

Words in the same sentence checking in with each other

LLM Advanced Sequence-to-Sequence

Listens to the end, then rebuilds it at whatever length it needs

Advanced Sigmoid

Squeezes any value into a smooth range between zero and one

A calculation that turns several scores into shares totaling 100%

LLM Basics Tensor

A bundle of numbers organized along several axes

Basics Transformer

An AI design where every word in a sentence looks at all the others

LLM Basics U-Net

Shrinks an image down, then grows it back while recovering position

Advanced Vanishing Gradient

When the fix-it signal almost fades out layer by layer, going back

Advanced Variational Autoencoder

An autoencoder that remembers things as a range, not a point

Generative Advanced Vector

A set of numbers for several items, written in a fixed order

LLM Basics Vision Transformer

Cutting a photo into pieces and treating it like text

Vision Advanced VRAM

The workspace where a graphics card lays out its numbers to compute

The learned multiplier on each input, deciding how much it counts