CS5720 - Week 1
Slide 11 of 20

Layers, Nodes, and Connections

Network Components

🏗️ Layers
Organized groups of neurons that process information at different abstraction levels
🔮 Nodes (Neurons)
Basic computational units that receive, process, and transmit signals
🔗 Connections (Weights)
Learnable parameters that determine signal strength between neurons
⚖️ Biases
Threshold adjustments that shift activation functions

Understanding Depth vs Width

Neural network architecture is defined by two key dimensions:
Depth: Number of layers (how deep)
More layers = More abstract representations
Width: Number of neurons per layer (how wide)
More neurons = More features per layer
Architecture Trade-offs:
  • Deep & Narrow: Efficient but harder to train
  • Shallow & Wide: Easy to train but less efficient
  • Optimal: Problem-dependent balance

Interactive Network Explorer

Network Topology Comparison

Shallow Network

⚪→⚪⚪⚪⚪⚪→⚪
2
Layers
30
Parameters

Deep Network

⚪→⚪⚪→⚪⚪→⚪⚪→⚪
5
Layers
24
Parameters

Wide Network

⚪⚪→⚪⚪⚪⚪⚪⚪⚪⚪→⚪⚪
3
Layers
40
Parameters
Prepared by Dr. Gorkem Kar