Variational Autoencoder in PyTorch - Model Metrics & Evaluation

Metrics & Evaluation - Variational Autoencoder

Which metric matters for Variational Autoencoder and WHY

For Variational Autoencoders (VAEs), the key metric is the Evidence Lower Bound (ELBO). ELBO combines two parts: the reconstruction loss and the Kullback-Leibler (KL) divergence.

The reconstruction loss measures how well the model can recreate the input data from its compressed form. The KL divergence measures how close the learned latent space is to a prior distribution, typically a standard normal distribution.

We want to minimize the total ELBO loss because it means the model is good at compressing and generating data that looks like the original.

Confusion matrix or equivalent visualization

VAEs are unsupervised models, so they don't use confusion matrices like classifiers.

Instead, we look at the loss curve during training:

Epoch | Reconstruction Loss | KL Divergence | Total ELBO Loss
-----------------------------------------------------------
  1   |       150.3         |     12.5      |     162.8
  2   |       120.7         |     10.2      |     130.9
  3   |       110.1         |      9.8      |     119.9
  ... 
    

Lower total ELBO loss over epochs means the model is learning better.

Precision vs Recall tradeoff (or equivalent) with concrete examples

VAEs balance two goals:

• Reconstruction quality: How close the output is to the input.
• Latent space regularization: How well the latent space follows a prior distribution, usually a normal distribution.

If we focus too much on reconstruction, the latent space may not be smooth or useful for generating new data.

If we focus too much on the KL divergence, the model may produce blurry or poor reconstructions.

Example: In image generation, a good VAE produces images that look like real digits (MNIST) but also can create new digits by sampling the latent space.

What "good" vs "bad" metric values look like for Variational Autoencoder

Good:

• ELBO loss steadily decreases during training.
• Reconstruction loss is low, meaning outputs look like inputs.
• KL divergence is not zero but balanced, indicating a meaningful latent space.
• Generated samples from latent space look realistic and diverse.

Bad:

• ELBO loss stays high or fluctuates wildly.
• Reconstruction loss is very high, outputs are blurry or wrong.
• KL divergence is near zero (posterior collapse), meaning latent space is ignored.
• Generated samples look random or all the same.

Metrics pitfalls

• Posterior collapse: KL divergence goes to zero, model ignores latent space, reconstruction may still be okay but generation fails.
• Overfitting: Very low reconstruction loss on training but poor generation or high loss on new data.
• Ignoring KL term: If KL weight is too low, latent space won't be regularized, hurting generation quality.
• Misinterpreting loss: ELBO loss is negative log likelihood plus KL, so lower is better, but absolute values depend on data scale.

Self-check question

Your VAE model has a reconstruction loss of 100 and KL divergence of 0.1 after training. Is this good?

Answer: The KL divergence is very low, close to zero, which suggests the model might be ignoring the latent space (posterior collapse). Even if reconstruction loss looks okay, the model may not learn a useful latent representation. You should try increasing the KL weight or adjusting the model to fix this.

Key Result

ELBO loss (reconstruction + KL divergence) is the key metric to evaluate Variational Autoencoders.

Practice

1. What is the main purpose of the encoder in a Variational Autoencoder (VAE)?

easy

A. To calculate the loss function

B. To produce the mean and variance for the latent space distribution

C. To generate new data samples from noise

D. To directly reconstruct the input data

Solution

1. Step 1: Understand the role of the encoder in VAE

   The encoder processes input data and outputs parameters of a probability distribution, typically mean and variance, representing the latent space.

2. Step 2: Differentiate encoder from decoder and loss

   The decoder reconstructs data, and loss measures reconstruction and latent space closeness. Encoder specifically outputs mean and variance.

3. Final Answer:

   To produce the mean and variance for the latent space distribution -> Option B

4. Quick Check:

   Encoder outputs mean and variance = D [OK]

Hint: Encoder outputs distribution parameters, not data directly [OK]

Common Mistakes:

• Confusing encoder with decoder
• Thinking encoder reconstructs data
• Assuming encoder calculates loss

2. Which PyTorch function is typically used to sample from the latent space in a VAE using the reparameterization trick?

easy

A. torch.randn()

B. torch.sigmoid()

C. torch.softmax()

D. torch.exp()

Solution

1. Step 1: Identify sampling method in VAE

   The reparameterization trick samples from a normal distribution using random noise, typically generated by torch.randn() in PyTorch.

2. Step 2: Understand other functions' roles

   torch.sigmoid() and torch.softmax() are activation functions, torch.exp() computes exponentials, none sample noise.

3. Final Answer:

   torch.randn() -> Option A

4. Quick Check:

   Sampling noise uses torch.randn() = A [OK]

Hint: Sampling noise uses torch.randn(), not activations [OK]

Common Mistakes:

• Using torch.sigmoid() to sample noise
• Confusing softmax with sampling
• Using torch.exp() incorrectly for sampling

3. Given the following PyTorch code snippet for the reparameterization trick in a VAE, what is the shape of the output z if mu and log_var have shape (batch_size, latent_dim)?

std = torch.exp(0.5 * log_var)
epsilon = torch.randn_like(std)
z = mu + epsilon * std

medium

A. (batch_size, latent_dim)

B. (latent_dim, batch_size)

C. (batch_size, 1)

D. (1, latent_dim)

Solution

1. Step 1: Analyze shapes of mu and log_var

   Both mu and log_var have shape (batch_size, latent_dim), so std and epsilon will have the same shape due to torch.exp and torch.randn_like.

2. Step 2: Calculate shape of z

   Since z = mu + epsilon * std, all operands have shape (batch_size, latent_dim), so z also has shape (batch_size, latent_dim).

3. Final Answer:

   (batch_size, latent_dim) -> Option A

4. Quick Check:

   Output shape matches input batch and latent dims = A [OK]

Hint: Output shape matches mu and log_var shapes [OK]

Common Mistakes:

• Swapping batch_size and latent_dim
• Assuming output is reduced to (batch_size, 1)
• Confusing shape broadcasting rules

4. Identify the error in this VAE loss function implementation in PyTorch:

def vae_loss(recon_x, x, mu, log_var):
    BCE = nn.functional.binary_cross_entropy(recon_x, x)
    KLD = -0.5 * torch.sum(1 + log_var - mu.pow(2) - log_var.exp())
    return BCE + KLD

medium

A. Incorrect sign in KLD calculation

B. Using torch.sum instead of torch.mean for KLD

C. Missing reduction='sum' in binary_cross_entropy causing scale mismatch

D. Not detaching mu and log_var before loss

Solution

1. Step 1: Check binary_cross_entropy usage

   By default, binary_cross_entropy returns mean loss per element. Summing KLD but not BCE causes scale mismatch.

2. Step 2: Fix BCE reduction

   Adding reduction='sum' to BCE matches scale with KLD, ensuring proper combined loss.

3. Final Answer:

   Missing reduction='sum' in binary_cross_entropy causing scale mismatch -> Option C

4. Quick Check:

   BCE needs reduction='sum' to match KLD scale = B [OK]

Hint: Match loss scales by using reduction='sum' in BCE [OK]

Common Mistakes:

• Ignoring BCE reduction parameter
• Changing KLD sign incorrectly
• Detaching mu/log_var unnecessarily

5. You want to train a VAE on images of size 28x28 pixels with a latent dimension of 20. Which of the following model architectures is most appropriate for the encoder part in PyTorch?

hard

A. A recurrent neural network processing images as sequences, outputting 20 values

B. A single linear layer flattening input and outputting 20 values directly

C. A fully connected network with no nonlinearities outputting 28x28 values

D. A convolutional network with Conv2d layers reducing spatial size, ending with linear layers outputting mean and log variance vectors of size 20

Solution

1. Step 1: Consider input data type and size

   Images of 28x28 pixels are spatial data; convolutional layers efficiently extract features and reduce spatial dimensions.

2. Step 2: Match output to latent dimension

   The encoder must output two vectors (mean and log variance) each of size 20, so linear layers after convolutions are suitable.

3. Step 3: Evaluate other options

   Single linear layer ignores spatial structure; RNNs are not typical for images; fully connected without nonlinearities won't learn complex features.

4. Final Answer:

   A convolutional network with Conv2d layers reducing spatial size, ending with linear layers outputting mean and log variance vectors of size 20 -> Option D

5. Quick Check:

   Conv layers + linear output for mean/log_var = C [OK]

Hint: Use conv layers for images, linear for latent params [OK]

Common Mistakes:

• Using only linear layers ignoring image structure
• Choosing RNNs for image encoding
• Omitting nonlinearities in encoder