CS5720 - Week 2
Slide 33 of 40

Bias-Variance Tradeoff (Simplified)

Understanding the Components

🎯 Bias
How far off your model's average prediction is from the true answer. High bias = systematic errors.
📊 Variance
How much your model's predictions vary when trained on different datasets. High variance = inconsistent.
⚖️ The Tradeoff
Reducing bias often increases variance, and vice versa. The art is finding the right balance.

The Archery Analogy

🏹 Think of Training a Model Like Archery
Each arrow = a model trained on different data
Bullseye = the true answer we want to hit
Pattern of arrows = our model's behavior
Perfect archer: Hits bullseye every time
(Low bias, low variance)

Systematic miss: Consistently hits same wrong spot
(High bias, low variance)

Random scatter: All over the target
(Low bias, high variance)

Terrible archer: Scattered and off-target
(High bias, high variance)

Interactive Bias-Variance Visualization

Low Bias, Low Variance
Ideal: Accurate and consistent predictions. This is what we aim for!
High Bias, Low Variance
Underfitting: Consistently wrong but predictably wrong. Model too simple.
Low Bias, High Variance
Overfitting: On average correct but inconsistent. Model too complex.
High Bias, High Variance
Worst case: Wrong and inconsistent. Usually means fundamental problems.

Adjust Model Complexity

Simple Model
Complex Model
Complexity Level: Medium
Bias²
25
Variance
25
Noise
10
Total Error = 60
Prepared by Dr. Gorkem Kar