Computer Vision Interview 20 essential Q&A Updated 2026
Autoencoder

Autoencoders: 20 Essential Q&A

Compress to a bottleneck and reconstruct—unsupervised representation learning and the path to VAEs.

~11 min read 20 questions Advanced
encoderbottleneckVAEreconstruction
Quick Navigation
1. What is an autoencoder?
2. Encoder role
3. Decoder role
4. Bottleneck
5. Reconstruction loss
6. Under vs over-complete
7. Denoising AE
8. Sparse autoencoder
9. VAE vs plain AE
10. KL term in VAE
11. Reparameterization trick
12. Anomaly detection
13. Relation to PCA
14. Disentanglement
15. AE vs GAN
16. Convolutional AE
17. Super-resolution / inpaint
18. Embedding for retrieval
19. Training tips
20. Limitations
1 What is an autoencoder? ⚡ easy
Answer: Neural net trained to copy input to output through a bottleneck: encoder maps x→z, decoder maps z→x̂—forces compact representation.
2 Role of the encoder? ⚡ easy
Answer: Maps high-dimensional input (e.g. image) to a lower-dimensional latent code z—extracts salient factors.
3 Role of the decoder? ⚡ easy
Answer: Maps latent z back to output space—should reconstruct structure lost only if bottleneck truly limits capacity.
4 Why a bottleneck? 📊 medium
Answer: Constrains information flow so the model must learn a compressed code—similar inputs map to nearby latents if the AE is well regularized.
5 Common reconstruction loss? 📊 medium
Answer: MSE (L2) per pixel for continuous images; BCE if outputs are probabilities; perceptual losses use a pretrained net’s features. 
# loss = F.mse_loss(recon, x)  # vanilla AE
6 Under-complete vs over-complete? 🔥 hard
Answer: Under-complete: dim(z) < dim(x)—true compression. Over-complete: dim(z) larger—needs regularization (sparse, denoising, VAE) or trivial identity.
7 What is a denoising autoencoder? 📊 medium
Answer: Train on corrupted inputs (noise, masking) to reconstruct clean x—learns robust features instead of copying noise.
8 Sparse autoencoder? 📊 medium
Answer: Penalize activations (e.g. KL on firing rates) so few units active per example—encourages meaningful distributed codes when over-complete.
9 VAE vs deterministic AE? 📊 medium
Answer: VAE encodes a distribution q(z|x); sample z for decoder—adds KL to prior p(z) for a generative model with smooth latent space.
10 What does the KL term do? 🔥 hard
Answer: Pulls approximate posterior toward prior (often N(0,I))—balances reconstruction vs regularization; enables sampling new z ~ p(z).
11 Reparameterization trick? 🔥 hard
Answer: Write z = μ(x) + σ(x)⊙ε with ε~N(0,1) so gradients flow through μ,σ—needed to backprop through stochastic sampling.
12 Use for anomaly detection? 📊 medium
Answer: Train on normal data; high reconstruction error on test indicates out-of-distribution—used in defect and fraud pipelines.
13 Link to PCA? 🔥 hard
Answer: Linear AE with MSE and tied weights can recover PCA subspace—deep nonlinear AE generalizes with stronger representational power.
14 Disentangled representations? 🔥 hard
Answer: Ideal latents align with generative factors; plain AE does not guarantee this—β-VAE and supervision help.
15 AE vs GAN for generation? 📊 medium
Answer: AE/VAE optimize likelihood-like objectives; GAN uses adversarial realism—GANs often sharper; VAEs more stable latent geometry.
16 Convolutional autoencoder? ⚡ easy
Answer: Encoder stacks conv+pool/downsample; decoder uses upsample/transpose conv—standard for images.
17 AE for super-resolution? 📊 medium
Answer: Condition decoder on low-res input or use skip connections (U-Net style)—AE ideas plus perceptual loss improve texture.
18 Embeddings for search? 📊 medium
Answer: Use encoder output as vector; nearest neighbors in latent space for similar images—may need contrastive training for metric quality.
19 Training tips? ⚡ easy
Answer: Normalize inputs; watch for posterior collapse in VAE; use skip connections if reconstruction is blurry from pure bottleneck.
20 Limitations? 📊 medium
Answer: Reconstructions can be blurry (MSE averages); latent may be entangled; vanilla AE is not a sharp generative model without VAE/GAN hybrids.

Autoencoder Cheat Sheet

Core
  • Encoder → z → decoder
Loss
  • MSE / BCE
  • Denoise / sparse
VAE
  • KL + sample z

💡 Pro tip: Bottleneck forces compression; regularize if over-complete.

Full tutorial track

Go deeper with the matching tutorial chapter and code examples.

Autoencoder Tutorial
Previous Next