SECAI Core · phase 2 of 15 · noun

Convolutional neural network

A deep learning method designed to process and analyze visual data by mimicking the way the human brain processes images; especially effective for tasks involving image recognition, object detection, and other image processing analyses.

The Explain card

Plain English
A convolutional neural network (CNN) is a deep learning design for images. It scans small patches, a bit like the visual cortex, to spot edges, shapes and finally whole objects.
Example
A CNN in a building's camera system identifies faces and license plates. A researcher prints a patterned sticker on a cap and the camera no longer recognizes a face at all.
Why it matters
CNNs are sensitive to pixel-level tweaks humans cannot see, which makes adversarial patches and perturbations effective against recognition systems used for access control and safety.
Hook
It sees the world in patches, so a patch can blind it.

Word knowledge

How the word is built, where it came from, and what it sits beside in memory.

In a sentence

The access camera passes each face crop through a convolutional neural network that must match the enrolled badge photo before the turnstile unlocks.

Why these words
  • convolutional Latin convolvere, to roll together, from volvere, to roll qualifies the network by the folding operation that gives the phrase its first word
  • neural Greek neuron, nerve or sinew, via New Latin neuralis imports the nerve-cell metaphor this family of models already carries
  • network English net plus work, a mesh of knots supplies the head noun, the lattice the other two words specify
Where it came from
Origin
English computing compound, from Latin convolvere (to roll together, from volvere, to roll), Greek neuron (nerve or sinew), and English network (net plus work)
Entered the language
1980s
What changed
Convolution had meant a coiling since the 1500s, then a fold of the brain, then the mathematical folding of two functions. Neural had meant of nerves. Network had meant a mesh of threads, then any lattice of links. Labs in the 1980s joined the three to name a model that slides a small filter across an input, and the phrase became the ordinary label for that architecture.
How it is spelled
Pattern
Latin volvere keeps v or u in English, never w
Pattern
-tion nouns sit beside -tional adjectives
Pattern
Greek eu in neuron is written eu, not u
Pattern
network is one closed compound, not net-work
Breaks the pattern
the noun phrase stays three words; a hyphen appears only when the stack modifies another noun
Breaks the pattern
convolution keeps Latin volut, while the verb convolve, from the same convolvere, writes v

Spelled like

  • involve
  • evolution
  • conventional
  • neurology
Broken into chunks
  • con
    • connect
    • confer
    • contain
  • volution
    • evolution
    • revolution
    • involution
  • al
    • cultural
    • original
    • typical
  • neur
    • neuron
    • neurology
    • neuralgia
  • net
    • internet
    • netting
    • dragnet
  • work
    • homework
    • workshop
    • artwork
What it sits beside

Same subject

Same shape

  • recurrent neural network
  • graph neural network
  • artificial neural network

Neural network architectures

Neural-network family

  • recurrent neural network
  • graph neural network
  • artificial neural network

Convolution family

  • convolution
  • deconvolution
  • convolutional layer

Where it sits in the deck

Phase 2: How Models Are Built: Architectures and Learning Mechanics

With the paradigms named, zoom in on the architectural building blocks that turn data into trained artifacts.

Neighbours in this phase