Fig. 3: Effects of cost and correlation on method performance.
From: A multi-fidelity machine learning approach to high throughput materials screening
The x-axis shows the effect of varying the magnitude of the discount of the low fidelity measurement relative to a measurement of the actual target function, while the y-axis shows the effect of varying the correlation of the lower fidelity to the target function. Each cell in the heatmap is shaded to reflect the difference between the number of epochs taken by the computational funnel and TVR-EI. Positive (blue shading) indicates that the TVR-EI algorithm was more efficient, whilst negative (red shading) indicates that the computational funnel was more efficient.
