求问：是否存在以两个样本为输入、标签为{同类,异类}的神经网络实例？

Absolutely, these models exist—they’re commonly referred to as Siamese Networks (or contrastive learning networks) and they’re built specifically for tasks where you need to compare two samples and classify them as same class (同类) or different class (异类).

How they work #

The core idea is straightforward:

• You feed two separate samples (e.g., two images, two text sequences, two sensor readings) into a shared feature extractor network. Sharing weights ensures that both samples are encoded using the same set of learned features, which is key for a fair, consistent comparison.
• After extracting features for each sample, you compute a similarity metric between the two feature vectors (like L1 distance, cosine similarity, or a learned similarity function).
• A final classification head takes this similarity score and outputs a binary label: 1 for "same class" and 0 for "different class".

Quick code example (PyTorch) #

Here’s a minimal, runnable implementation to illustrate the structure:

import torch
import torch.nn as nn

class SiameseNetwork(nn.Module):
    def __init__(self, input_dim):
        super().__init__()
        # Shared feature extractor (same weights for both inputs)
        self.feature_extractor = nn.Sequential(
            nn.Linear(input_dim, 128),
            nn.ReLU(),
            nn.Linear(128, 64),
            nn.ReLU(),
            nn.Linear(64, 32)
        )
        # Classification head for same/different prediction
        self.classifier = nn.Sequential(
            nn.Linear(32, 16),
            nn.ReLU(),
            nn.Linear(16, 1),
            nn.Sigmoid()
        )

    def forward(self, x1, x2):
        # Extract features for both input samples
        feat1 = self.feature_extractor(x1)
        feat2 = self.feature_extractor(x2)
        # Use L1 distance as the similarity measure (cosine similarity works too)
        similarity = torch.abs(feat1 - feat2)
        # Output probability of being the same class
        output = self.classifier(similarity)
        return output

# Test the model
model = SiameseNetwork(input_dim=100)
batch_sample1 = torch.randn(32, 100)  # Batch of 32 input samples
batch_sample2 = torch.randn(32, 100)
predictions = model(batch_sample1, batch_sample2)

Why you might have missed them #

Chances are you were using the wrong search terms. Instead of generic phrases like "two-sample input neural network", try targeted keywords like:

• Siamese network
• Contrastive classification network
• Pairwise similarity classification model

These models are widely used in real-world applications:

• Face verification (is this the same person as the ID photo?)
• Text semantic matching (do these two sentences mean the same thing?)
• E-commerce product matching (are these two listings for the same item?)

内容的提问来源于stack exchange，提问作者Will Bolden