Cognitron
No estimate
No hardware requirements for this model
The weights for this model have not been published, so it cannot be downloaded or run on your own hardware at any size. It is reachable only through its provider, and no graphics card changes that.
On record
Full specification
Everything on record for this model. Most of it describes how it was trained rather than how it runs — useful context for judging how much work went into it, and how it compares with models built at a different scale.
Origin
Who built this model, where, and when it was published.
- Organisation
- Biological Cybernetics
- Organisation type
- Industry
- Country
- Japan
- Published
- 1 September 1975
- Authors
- Kunihiko Fukushima
What it does
The problem areas the model was built for. A model can carry several of each.
- Domain
- Other
- Task
- Miscellaneous image analysis, Image classification
Size
How large the model is and how much data it was trained on. Parameters are the figure that decides whether it fits on a given graphics card.
- Parameters
- 21.6K
- Training data
- 5 tokens
- Epochs
- 20
4 layers, 288 neurons per layer, weights connect each neuron to only 25 neurons in the previous layer. Only 3 layers with learnable weights, the first layer is just an input representation (see Fig 5) 3*288*25 parameters
5 examples presented for (at least) 20 cycles = 100 training steps
Training compute
The arithmetic performed to train the model, measured in floating-point operations. It is a measure of what the training run cost, not of how fast the finished model answers you.
- Training compute
- 5.2 × 10⁶ FLOP
- How it was established
- Operation counting
No real backward pass as weights are sparsely updated. Assuming 20% additional compute for the weight update. Total compute estimate: 100*2*3*288*25*1.2 = 5184000
How it is classified
Labels the source dataset applies when tracking notable models, and how confident it is in the entry.
- Why it is tracked
- Historical significance
- Record confidence
- Confident
- Citations
- 791
Precursor of the Neocognitron
Sources
Where this record came from and when it was last checked.
- Reference
- Cognitron: a self-organizing multilayered neural network
- Last updated
- 28 November 2025
What the numbers mean
Background
Cognitron was published by Biological Cybernetics, in Japan, in September 1975. It comes out of industry.
It works in Other, and is recorded as doing miscellaneous image analysis, Image classification.
Its weights were never published, so it can only be reached through its provider. No graphics card changes that.
Training and provenance
Producing it required around 5.2 × 10⁶ FLOP of arithmetic, which is a statement about the training budget rather than about inference.
It was trained on about 5 tokens of text.
Its inclusion criterion is historical significance.
Answers
Cognitron — common questions
Who created Cognitron?
Cognitron was published by Biological Cybernetics, based in Japan, categorised as industry.
When was Cognitron released?
Cognitron was published in September 1975. Capability per parameter has improved considerably since, so a newer model of the same size is often the better use of the same hardware.
What is Cognitron used for?
Cognitron works in Other, and is recorded as handling miscellaneous image analysis, Image classification. Models frequently carry more than one of each, and the tags describe purpose rather than capability limits.
How much compute was used to train Cognitron?
Around 5.2 × 10⁶ FLOP. That measures what producing the model cost and says nothing about how quickly it answers once trained — inference speed comes from memory bandwidth, not from the training budget.
What GPU do I need to run Cognitron?
None. Cognitron is a closed model — its weights were never published, so it cannot be downloaded or run on your own hardware at any price. It is reachable only through its provider.
Is Cognitron open source?
The licensing for Cognitron was never recorded in our source data. We treat unstated licensing as closed, because an unrecorded licence is not one to rely on.
How many parameters does Cognitron have?
Cognitron has 21.6K parameters. 4 layers, 288 neurons per layer, weights connect each neuron to only 25 neurons in the previous layer. Only 3 layers with learnable weights, the first layer is just an input representation (see Fig 5) 3*288*25 parameters. That figure is the total, and it is what decides how much memory the model needs — roughly half a gigabyte per billion at the compression most people use.
The other direction
Looking at it from the other side?
This page starts from the model. If you already own a card and want to know everything it will run, start from the hardware instead.