CS5720 - Week 2
Slide 21 of 40

The Learning Problem

What is Learning in Neural Networks?

Learning in neural networks is the process of automatically adjusting the network's parameters (weights and biases) to improve its performance on a given task.
The Core Challenge:

• We have data (inputs and desired outputs)
• We have a network (with random initial weights)
• We need to find the right weights that map inputs to outputs
• This is fundamentally an optimization problem
💡 Key Insight
Learning is not memorization! The goal is to find patterns in the training data that generalize to new, unseen examples.

The Learning Process

  • 1️⃣
    Initialize
    Start with random weights (the network knows nothing)
  • 2️⃣
    Predict
    Feed input through the network to get predictions
  • 3️⃣
    Measure Error
    Compare predictions with actual outputs
  • 4️⃣
    Update Weights
    Adjust weights to reduce error
  • 5️⃣
    Repeat
    Continue until performance is satisfactory

The Learning Problem Visualized

Training Data
Input
Neural Network
Weights?
Need to learn!
Desired Output
Target
The Challenge: Find weights that transform inputs into correct outputs for ALL training examples
Prepared by Dr. Gorkem Kar