Flatten layer in PyTorch - Cheat Sheet & Quick Revision

Recall & Review

beginner

What is the purpose of a Flatten layer in a neural network?

A Flatten layer converts a multi-dimensional input (like an image) into a one-dimensional vector so it can be fed into a fully connected layer.

beginner

How does the Flatten layer affect the shape of the input tensor?

It reshapes the input tensor from shape (batch_size, channels, height, width) to (batch_size, channels * height * width), keeping the batch size the same.

beginner

Show a simple PyTorch code snippet to add a Flatten layer in a model.

import torch.nn as nn

model = nn.Sequential(
    nn.Flatten(),
    nn.Linear(28*28, 10)
)

beginner

Why do we need to flatten data before feeding it to a fully connected layer?

Fully connected layers expect 1D input vectors. Flattening changes multi-dimensional data into a 1D vector so the layer can process it.

beginner

Can the Flatten layer change the batch size of the input?

No, the Flatten layer keeps the batch size unchanged. It only reshapes the other dimensions into one dimension.

What does the Flatten layer do to the input tensor?

A Normalizes the data

B Changes the batch size

C Adds more dimensions

D Converts it to a 1D vector per sample

The Flatten layer reshapes each sample into a 1D vector without changing batch size.

In PyTorch, which class is used to add a Flatten layer?

A nn.Flat

B nn.Reshape

C nn.Flatten

D nn.Vectorize

PyTorch provides nn.Flatten to flatten input tensors.

Why is flattening necessary before a fully connected layer?

A Because fully connected layers require 1D input vectors

B To reduce batch size

C To increase the number of channels

D To normalize the input

Fully connected layers expect 1D vectors, so flattening prepares the data.

If input shape is (batch_size, 3, 32, 32), what will be the shape after Flatten?

A (batch_size, 3*32*32)

B (3, 32, 32)

C (batch_size, 32, 32)

D (batch_size, 3, 32)

Flatten combines all dimensions except batch into one dimension.

Does Flatten layer change the batch size dimension?

A Yes, it doubles the batch size

B No, batch size stays the same

C Yes, it halves the batch size

D Yes, it removes the batch size

Flatten only reshapes other dimensions, batch size remains unchanged.

Explain in your own words what a Flatten layer does and why it is used in neural networks.

Think about how images are prepared before classification.

Write a simple PyTorch model snippet that includes a Flatten layer followed by a linear layer.

Use nn.Sequential for simplicity.