Siamese Network

Gold definitionUpdated Apr 2, 2026

A Siamese Network is a type of neural network architecture characterized by having two or more identical subnetworks (or 'towers') that share the same weights and architecture. Its primary purpose is to learn a similarity metric between two inputs, rather than performing direct classification. The core mechanism involves feeding two distinct inputs into the twin networks, which then produce corresponding feature vectors in an embedding space. A distance metric (e.g., Euclidean distance) or a contrastive loss function (like triplet loss or contrastive loss) is applied to these feature vectors to minimize the distance between similar pairs and maximize it for dissimilar pairs. This architecture is crucial for tasks where the number of classes is very large or changes frequently, or when only a few examples per class are available (few-shot learning). It is widely used in research areas like computer vision for face recognition, signature verification, and object tracking, as well as in natural language processing for semantic similarity and in graph learning for node classification, as seen in methods like SNGCL.

Core Architecture of Siamese Networks

Twin Subnetworks: The fundamental design involves two or more identical neural networks, each processing one input. These subnetworks share all their weights and biases, ensuring that they perform the same transformation on their respective inputs, mapping them into a comparable embedding space.
Embedding Space Learning: The primary goal of a Siamese Network is to learn a robust embedding function. This function transforms input data into a lower-dimensional vector space where the distance between vectors directly corresponds to the similarity of the original inputs.
Distance Metric and Loss Functions: After generating embeddings, a distance metric (e.g., L1, L2) is computed between the output vectors. Training typically employs specialized loss functions like contrastive loss or triplet loss, which push embeddings of similar pairs closer and dissimilar pairs further apart, as exemplified by the 'improved triple recombination loss function' in SNGCL for graph learning.

Applications of Siamese Networks in Research

Graph Node Classification: Siamese networks are applied in graph learning for tasks like node classification, especially when labels are scarce. Methods like SNGCL utilize Siamese networks by passing global and local feature smoothing matrices into its target and online networks to improve classification efficiency and address challenges of negative sample reliance.

At a glance

Executive summary

Siamese Networks are neural networks with two identical branches that learn to compare inputs and determine their similarity. They are particularly useful for tasks where you need to tell if two items are the same or different, even with limited examples, by mapping them to a space where similar items are close together.

TL;DR

A Siamese Network uses two identical neural networks to learn how similar two things are, making it great for tasks like face recognition or comparing data points.

Key points

Comprises two or more identical subnetworks sharing weights to learn a similarity function.
Solves problems requiring similarity comparison, few-shot learning, and verification tasks.
Used in computer vision (face recognition), NLP (semantic similarity), and graph learning (node classification).
Unlike traditional classifiers that output class probabilities, Siamese networks output a distance/similarity score.
Research trend focuses on improving contrastive loss functions and robust negative sample generation for better embeddings.

Use cases

Face verification systems for unlocking smartphones or access control.
Signature verification in banking or document processing to detect forgeries.
One-shot image classification where a model identifies new objects from a single example.
Duplicate detection for products in e-commerce or documents in information retrieval.
Node classification in graph neural networks, identifying similar nodes in complex networks.

Also known as

Twin Network, Siamese Neural Network, Pseudo-Siamese Network, Triplet Network