PyTorch torch.nn.Bilinear Function

PyTorch torch.nn 参考手册PyTorch torch.nn Reference Manual

torch.nn.BilinearIt is a bilinear layer in PyTorch.

It performs a bilinear transformation on two inputs, commonly used for feature fusion.

Function Definition

torch.nn.Bilinear(in1_features, out_features, out_features, bias=True)

Formula

y = x1 * W * x2 + b

Usage Example

Example 1: Basic Usage

Example

import torch
import torch.nn as nn

# Bilinear layer: two 100-dimensional inputs -> 50-dimensional output
bilinear = nn.Bilinear(100, 100, 50)

# Two inputs
x1 = torch.randn(4, 100)
x2 = torch.randn(4, 100)

output = bilinear(x1, x2)

print("Input 1:", x1.shape)
print("Input 2:", x2.shape)
print("Output:", output.shape)

Example 2: Feature Fusion

Example

import torch
import torch.nn as nn

# Bilinear feature fusion (similar to Skip-gram)
class FusionNet(nn.Module):
    def __init__(self, dim=128):
        super(FusionNet, self).__init__()
        self.bilinear = nn.Bilinear(dim, dim, dim)

    def forward(self, feat1, feat2):
        return self.bilinear(feat1, feat2)

model = FusionNet()
f1 = torch.randn(4, 128)
f2 = torch.randn(4, 128)

output = model(f1, f2)
print("Fusion output:", output.shape)

Example 3: Parameter Count

Example

import torch
import torch.nn as nn

# Bilinear layer parameters
bilinear = nn.Bilinear(100, 100, 50)

print("Number of parameters:", sum(p.numel() for p in bilinear.parameters()))
print("Weight shape:", bilinear.weight.shape)  # (50, 100, 100)

Use Cases

  • Feature fusion: Multimodal
  • Attention mechanism
  • Interaction modeling

Note: The bilinear layer has a large number of parameters; use with caution.

PyTorch torch.nn 参考手册PyTorch torch.nn Reference Manual

Other Extensions