Convolutional Neural Network
/ˌkɒnvəˈluːʃənəl ˈnjʊərəl ˈnɛtwɜːk/con·vo·lu·tion·al neu·ral net·worknoun
Definition
1.[in deep learning] an architecture that slides a small set of learned filters across grid-structured input, so the same local pattern is recognised wherever it appears and far fewer weights are needed than a fully connected design.
A convolutional neural network reached human-level accuracy on the digit set with a fraction of the parameters.
Formal statement
(f * g)[n] = sum_m f[m] g[n - m]The discrete convolution each filter applies at every position.
Etymology
From convolution, Latin convolvere 'to roll together', naming the integral operation the layers perform, plus neural network. The architecture descends from Fukushima's neocognitron of 1980 and was given its modern trainable form by LeCun and colleagues in 1989.
Synonyms
- ConvNetsense 1 · Absolute
Antonyms
- fully connected networksense 1 · Relational
See also
References
- LeCun, Y. et al. (1989). Backpropagation Applied to Handwritten Zip Code Recognition.Neural Computation, 1(4), 541–551.
- Fukushima, K. (1980). Neocognitron: A self-organizing neural network model for a mechanism of pattern recognition.Biological Cybernetics, 36, 193–202.