Multi-Objective Optimization

Guide to Pareto-based optimization for accuracy-interpretability trade-offs.

Overview

Multi-objective optimization finds a set of Pareto-optimal solutions representing different trade-offs between objectives (accuracy vs interpretability).

Using the Pareto Optimizer

# Run Pareto optimization
python scripts/run_pareto_optimization.py --config configs/paper.yaml --dataset breast_cancer

Understanding Pareto Fronts

A solution is Pareto-optimal if improving one objective requires sacrificing another.

    Accuracy ↑
       |
    95%|     ● (High acc, low interp)
       |
    90%|   ●   ● (Pareto front)
       |
    85%| ●       (High interp, lower acc)
       |_____________
            Interpretability →

Example Results

The Pareto script explores different weight combinations:

Solution	Accuracy	Nodes	Depth	Weight Ratio
1	94.2%	15	5	90/10 (accuracy focused)
2	92.8%	10	4	70/30 (balanced)
3	90.1%	6	3	50/50 (interpretability focused)

Next Steps

See scripts/run_pareto_optimization.py for implementation details.