layer_trainable

requires_grad

grad

train

Applies k-means clustering

Performs dimensionality reduction

Trains a CNN for image classification

 Tokenizes and processes text using BERT

train_on_batch

fit

compile

evaluate

Computes class probabilities using softmax

Creates a convolutional neural network ResNet18

Trains the model on random data

Applies Grad-CAM for prediction explanation

get_shape()

dim

shape

output_shape

[0, 0, 0, 1, 2]

[1, 1, 1, 1, 1]

[-2, -1, 0, 1, 2]

[2, 1, 0, -1, -2]

It will be very large (exploding gradients)

It will be zero (ReLU dead neurons)

It will match torch.ones_like(output)

It will be near zero due to sigmoid saturation

[3, 4]

[0.1, 0.1]

[9, 16]

[6, 8]

To tune the learning rate

To verify correctness of backpropagation via numerical gradient

To update parameters using gradient descent

To visualize the loss surface

Resize

RandomCrop

Standardize

Normalize