CS5720 - Week 2
Slide 37 of 40

Dropout: Random Neuron Removal

What is Dropout?

Dropout randomly sets some neurons to zero during training, forcing the network to be robust and preventing over-reliance on specific neurons.
Key Benefits:

Prevents neurons from co-adapting too much
Forces the network to learn redundant representations
Reduces overfitting dramatically
Improves generalization to new data
🎭 Think of it like...
A sports team where random players are benched each game. The team learns to work well even when key players are missing, making them more resilient overall!

Training vs Inference

🎲 Training Mode
Randomly drop neurons
(e.g., 20-50% of them)
Forces learning redundancy
🧠 Inference Mode
Use ALL neurons
Scale outputs appropriately
Get full network prediction
Critical Point:
Dropout is ONLY active during training. During inference, all neurons participate in the final prediction.

Interactive Dropout Visualization

Input Layer
Hidden Layer 1
Hidden Layer 2
Output Layer
Dropout Rate: 30%
Mode:
🎲 Training Mode: Neurons will be randomly dropped during forward pass
Prepared by Dr. Gorkem Kar