Training Time vs Accuracy Trade-off

Scatter plot of training time (x) vs validation accuracy (y) with Pareto frontier highlighted.

Prompt

A scatter plot showing training time vs validation accuracy for 8 model variants.

X-axis: Training time (hours), log scale, range 1 to 32.
Y-axis: Validation accuracy (%), range 85 to 97.

Eight data points (each labeled with model name):
- BaseSmall: 1.5h, 88.2%
- BaseMedium: 3.2h, 90.5%
- BaseLarge: 7.5h, 92.7%
- ProSmall: 2.1h, 89.8%
- ProMedium: 4.8h, 92.0%
- ProLarge: 11.0h, 94.5%
- UltraSmall: 3.5h, 91.5%
- UltraLarge: 24.0h, 96.3%

Highlight the Pareto frontier with a connected line through ProSmall, ProLarge, UltraLarge (the time-accuracy efficient set).

Use distinct shapes per model family (Base = circle, Pro = triangle, Ultra = square) and per-family colors.

Style: clean academic scatter, gridlines, log-scale x-axis labeled (1, 2, 4, 8, 16, 32), sans-serif, white background. Place Pareto label inside the figure with a small arrow.

Use in Generator

When to use

For systems-ML papers comparing model variants on the time-accuracy frontier.

Variations

With error bars

Add vertical error bars on each point representing the standard deviation across 3 random seeds. Make Pareto-frontier points slightly larger.

Tips

Use log scale for time. Linear time axes crush small models against the y-axis.
Highlight the Pareto frontier. Without it the figure is just a cloud.
Use distinct shapes per family — color alone fails for color-blind readers.

FAQ

Can I add a budget constraint line?

Add a vertical dashed line at your training-time budget (e.g., 8h) with a "budget" label. Points to the right are infeasible.