Computer Vision Interview
20 essential Q&A
Updated 2026
AlexNet
AlexNet: 20 Essential Q&A
The architecture that popularized deep CNNs on ImageNet—ReLU, dropout, and GPU scale.
~10 min read
20 questions
Advanced
ImageNetReLUdropoutLRN
Quick Navigation
1
Why is AlexNet important?
⚡ easy
Answer: Won ImageNet 2012 by a large margin—showed deep CNNs + GPU + data could beat hand-crafted features, sparking the deep learning boom in vision.
2
What was ImageNet 2012?
📊 medium
Answer: 1.2M images, 1000 classes—AlexNet ~16% top-5 error vs previous ~26% with shallow methods—breakthrough result.
3
Rough architecture?
📊 medium
Answer: Five conv layers (some grouped across 2 GPUs) + max pooling + three large FC layers + softmax—deeper than prior CNNs for this task.
4
Why ReLU?
📊 medium
Answer: Faster training than saturating tanh/sigmoid; mitigates vanishing gradient in deep stacks; sparse activations.
5
Use of dropout?
📊 medium
Answer: Regularize huge FC layers by randomly zeroing neurons—reduces co-adaptation on training set.
6
What was LRN?
🔥 hard
Answer: Local response normalization—side inhibition across channels; later often replaced by batch norm; minor effect in hindsight.
7
Overlapping pooling?
📊 medium
Answer: Stride smaller than pool window—slightly richer downsampling vs non-overlapping; less common in newer nets.
8
Two GPUs?
⚡ easy
Answer: Model split across GPUs due to memory limits—cross-GPU connections only on certain layers (engineering constraint of the time).
9
Augmentation?
📊 medium
Answer: Random crops/flips from 256×256, PCA color jitter—reduces overfitting and increases effective data.
10
Parameters?
⚡ easy
Answer: On order of 60M—mostly FC layers; later architectures reduce FC params with GAP.
11
Training details?
📊 medium
Answer: SGD + momentum, weight decay, learning rate schedule dropping on plateaus—long schedule on two GPUs.
12
Overfitting risk?
📊 medium
Answer: Large capacity vs data—addressed by dropout, aug, and weight decay; still a concern for smaller datasets when fine-tuning.
13
vs VGG?
📊 medium
Answer: VGG uses uniform 3×3 stacks, deeper, more systematic—higher accuracy, more compute; AlexNet shallower irregular design.
14
vs ResNet?
📊 medium
Answer: ResNet adds residuals enabling much deeper nets—AlexNet depth modest by today’s standards.
15
Use AlexNet now?
⚡ easy
Answer: Mostly for teaching/history; ResNet/EfficientNet backbones dominate transfer learning—AlexNet too weak/slow vs modern alternatives.
16
Typical input?
📊 medium
Answer: 224×224 crops from 256×256 resized image—standard pipeline referenced in many papers.
17
Output layer?
⚡ easy
Answer: 1000-way softmax for ImageNet classes—cross-entropy loss during training.
18
Obsolete?
⚡ easy
Answer: For production accuracy, yes; for pedagogy and history, still the canonical “first big win” story.
19
Impact beyond vision?
⚡ easy
Answer: Validated deep learning at scale—influenced speech, NLP later wave; proved GPUs + data + depth recipe.
20
Modern small nets?
📊 medium
Answer: MobileNet, EfficientNet achieve better accuracy/FLOPs—mobile edge rarely uses AlexNet-sized FC heads.
AlexNet Cheat Sheet
Breakthrough
- ImageNet 2012
Ideas
- ReLU
- Dropout
Today
- Historical
- Superseded
💡 Pro tip: Name ImageNet 2012 + ReLU + dropout + GPUs.
Full tutorial track
Go deeper with the matching tutorial chapter and code examples.