Siamese network
In plain wordsA matching machine built from twin copies of one neural network: each copy turns one of two inputs, say two signatures or two face photos, into a summary, and the two summaries are compared.
An architecture in which two or more identical subnetworks with shared weights (weight sharingweight sharing, n.: Reusing the same learned numbers in many places in a network, for example one edge detector applied at every spot in an image, so the network needs fewer parameters and learns from less data.) map different inputs into a common embeddingembedding, n.: A list of numbers that stands for a word, picture or other item, arranged so that similar items get similar lists. It lets a computer measure similarity of meaning as distance. space, where a distance or similarity between the outputs decides whether the inputs match. Because the same function encodes both inputs, the comparison is symmetric and the network can verify pairs from classes never seen in training, suiting one-shot and few-shotfew-shot learning, n.: Getting an AI system to handle a new task or recognize a new kind of thing from just a handful of examples, the way a person can learn a new word after hearing it once or twice. recognition, signature and face verification, duplicate detection and object tracking. Siamese networks are trained with pairwise contrastivecontrastive learning, n.: A way to learn without labels by comparison: the model learns to pull together two versions of the same thing, such as two crops of one photo, and push apart unrelated things. losses or, with three branches (anchor, positive, negative), with the triplet loss of metric learningmetric learning, n.: Teaching a model to measure similarity: it learns to place items so that ones that should count as alike end up close together and different ones far apart.. The same twin-encoder structure, with an asymmetric predictor added, underlies self-supervised methods such as BYOL and SimSiam, and the bi-encoderbi-encoder, n.: An AI design that turns a question and a document into number lists separately, so documents can be processed once in advance and compared with any new question in a flash. models used for retrieval.
The Siamese network embeds both signatures and accepts the pair if their distance is below 0.4.
Origin
Introduced by Jane Bromley, Isabelle Guyon, Yann LeCun, Eduard Säckinger and Roopak Shah of Bell Labs in 'Signature Verification Using a Siamese Time Delay Neural Network' (NIPS 1993). Sumit Chopra, Raia Hadsell and LeCun applied the idea to face verification in 2005.
Usage note
The name alludes to Chang and Eng Bunker, the 19th-century conjoined twins born in Siam; some writers now prefer 'twin network'.
Related
- contrastive learningcontrastive learning, n.: A way to learn without labels by comparison: the model learns to pull together two versions of the same thing, such as two crops of one photo, and push apart unrelated things.
- metric learningmetric learning, n.: Teaching a model to measure similarity: it learns to place items so that ones that should count as alike end up close together and different ones far apart.
- weight sharingweight sharing, n.: Reusing the same learned numbers in many places in a network, for example one edge detector applied at every spot in an image, so the network needs fewer parameters and learns from less data.
- bi-encoderbi-encoder, n.: An AI design that turns a question and a document into number lists separately, so documents can be processed once in advance and compared with any new question in a flash.
- few-shot learningfew-shot learning, n.: Getting an AI system to handle a new task or recognize a new kind of thing from just a handful of examples, the way a person can learn a new word after hearing it once or twice.
- facial recognitionfacial recognition, n.: Technology that identifies or checks who someone is by comparing a picture of their face with stored photos, used by phones, passport gates and police databases.
Sources
- Bromley, J., Guyon, I., LeCun, Y., Säckinger, E. & Shah, R. (1993). Signature Verification Using a 'Siamese' Time Delay Neural Network. Advances in Neural Information Processing Systems 6.
Mentioned in
bi-encoderbi-encoder, n.: An AI design that turns a question and a document into number lists separately, so documents can be processed once in advance and compared with any new question in a flash., contrastive learningcontrastive learning, n.: A way to learn without labels by comparison: the model learns to pull together two versions of the same thing, such as two crops of one photo, and push apart unrelated things., Sentence-BERTSentence-BERT, prop. n.: A way of turning a whole sentence into a list of numbers so that sentences with similar meanings get similar numbers, which makes it fast to search millions of texts by meaning rather than by exact words., weight sharingweight sharing, n.: Reusing the same learned numbers in many places in a network, for example one edge detector applied at every spot in an image, so the network needs fewer parameters and learns from less data., metric learningmetric learning, n.: Teaching a model to measure similarity: it learns to place items so that ones that should count as alike end up close together and different ones far apart., time delay neural networktime delay neural network, n.: An early network for speech that applied the same small pattern detector at every moment of a sound, so it could recognize a sound wherever it occurred in time.