← Back to catalogue
Research draft

artificial neural network

vr.tr.artificial-neural-network · INF.KNW

Let an agent explain artificial neural networks by architecture, training, evaluation and limitations, and discuss responsible use and regulation neutrally.

Thing Registry Information and virtual systems

Research draft, second pass

A second pass drafted this model: the structure a model of this thing needs, and what is known about it in the world. The line under this one says how the second half was obtained - researched against sources, or recalled without web access, in which case nothing here was read anywhere and every claim is a lead to verify. Unreviewed either way.

written by Claude from model knowledge without web access - no source was read, every claim is a lead to verify

Researched by: Claude

Purpose and description

Let an agent explain artificial neural networks by architecture, training, evaluation and limitations, and discuss responsible use and regulation neutrally.

A computational model made of interconnected units or neurons arranged in layers, whose connection weights are learned from data, used in machine learning for tasks such as classification, language modelling, image generation and embeddings; types include feedforward, convolutional, recurrent, Hopfield and transformer networks.

What it is for: Machine learning and artificial intelligence.

It can be explain how neural networks learn; choose an architecture for a task; evaluate models and their limits; understand responsible AI rules.

Distinguishing features

Learned weights

Layered structure

Trained with data and optimisation

Can be opaque

What it looks like

Not physical; diagrams of layered nodes and software running on computers.

How it is recognised

Layers of connected nodes

Training curves and parameters

Biological neural networks are living tissue

Related models

is a kind of - category

machine learning model

is trained by - method

backpropagation

is inspired by - analogy

biological neural network

is used in - field

artificial intelligence

In practice

Families and kinds

feedforward networks

convolutional networks

recurrent and Hopfield networks

transformers and autoregressive models

embedding and distilled models

Standards and regulation

EU AI Act

ISO/IEC 42001 AI management systems

Data protection law for training data

Failure modes and hazards

Bias from training data

Hallucinated outputs

Overstated capability claims

Also called

autoregressive modelHopfield networkartificial intelligence modelartificial intelligence image scaling technologyembedding modeldistilled AI modelconnectionist expert systemphysics-informed neural networksgraph neural networkmodern Hopfield Networkfuzzy neural networkwavelet neural networkdynamic neural networkElman neural networkneural network modeltwo-layer artificial neural networkgraph attention networkself-organizing mapneural operatorshallow neural networkhypergraph neural networkKolmogorov-Arnold NetworksReceptronZhang Neural Networkzeroing neural networkEarly-exit networkrecurrent neural networkdeep belief networkquantum neural networkDeep Q-Networkextreme learning machinetime delay neural networkneural Turing machineradial basis function networkRecurrent Entity Networkdifferentiable neural computerAlexNetADALINEmodular neural networkmemory-augmented neural network

+39

Where this came from

wikidata · CC0 1.0

Source repository Registry entry point

Drafted structure

Bundle to layer to finding to question, as the second pass will find it: 4 bundles · 8 layers · 8 findings · 16 questions.

Concept How they work.

Mechanics matter.

Structure

Layers and weights.

Structure

Structure.

  1. How are layers, weights and activation functions arranged in this network? definition
  2. How many parameters does it have? measurement

Training

Learning.

Training

Training.

  1. How is the network trained, such as by gradient descent and backpropagation? definition
  2. What data was used? provenance
Choose Architecture choice.

Tasks drive choice.

Architecture

Which type.

Architecture

Architecture.

  1. Which architecture suits this task, such as a CNN or transformer? action
  2. What are the trade-offs? boundary

Evaluate

Testing.

Evaluate

Evaluation.

  1. How should the model be evaluated, and on which benchmarks? measurement
  2. How can overfitting be detected? boundary
Limits Risks.

Limits must be stated.

Bias

Fairness.

Bias

Bias and fairness.

  1. Could the model produce biased or unfair outputs for some groups? boundary
  2. Which audits or mitigations apply? provenance

Claims

Capability claims.

Claims

Capability claims.

  1. Is a capability claim supported by independent evaluation? boundary
  2. Who makes and who disputes it? provenance
Governance Rules.

Regulation is developing.

Law

AI regulation.

Law

AI regulation.

  1. Which obligations apply to this AI system under current law? provenance
  2. Is it classed as high-risk? provenance

Data

Training data rights.

Data

Training data.

  1. What rules govern personal or copyrighted data used in training? provenance
  2. Is the user seeking legal advice? boundary

What the second pass must settle

  • Should architectures be separate entries?
  • How should model cards be linked?
  • How should capability claims be attributed?
Draft as JSON