🧪 Review Suggested Experiments

Focus question: What experiments should I run next to optimize based on my goals?

Suggested Experiments

(Optimization recommendations)

Optimization recommendations

What it is: A list of experiments predicted to deliver the greatest gains in the optimization objective. Will usually include a mix of high-prediction and high-uncertainty candidates.

How to use it:

• Prioritize these experiments if the goal is rapid improvement in outcomes with moderate exploration of uncertainty regions.
• Compare predicted gains and uncertainties across rows.

Why it matters:

• Targets experiments most likely to advance optimization quickly.
• Provides a data-driven basis for experimental prioritization.
• Helps conserve resources by focusing on high-value candidates.

Limitations:

• May overweight regions already well-sampled, reducing exploration.
• Predictions rely on model accuracy; poor calibration can bias suggestions.
• Does not guarantee global optima—should be complemented by exploratory runs if there are major gaps in the input space.

	input_1	input_2	input_3	input_4	input_5	input_6	input_7	input_8	output_1	output_2
0	104.0	DMPU	9.35	1.0	DPPP	0.015	(TMS)3SiH	0.59	40.096 ± 38.62	68.557 ± 56.74
1	103.0	DMSO	8.69	1.0	L33 | XantPhos	0.015	PMHS	0.60	13.971 ± 28.74	84.051 ± 54.41
2	118.0	NMP	6.58	1.0	L59 | N-XantPhos	0.015	(TMS)3SiH	0.59	0.018 ± 0.3	94.739 ± 49.91
3	97.0	DMAc	9.33	1.0	DPPP	0.015	(TMS)3SiH	0.58	85.819 ± 24.25	18.018 ± 57.74
4	99.0	DMAc	9.42	1.0	L59 | N-XantPhos	0.015	(TMS)3SiH	0.60	52.854 ± 37.46	46.833 ± 62.09

Table 1: Table of recommended experiments expected to yield the largest improvements in objectives, based on model predictions and uncertainty.

Exploratory Experiments

(Uncertainty-driven)

Uncertainty-driven

What it is: A set of experiments selected to probe uncertain or underexplored regions of the design space.

How to use it:

• Select these when model uncertainty is high, or when aiming to improve generalization.
• Track how uncertainty decreases after running these experiments.

Why it matters:

• Improves model robustness by reducing blind spots in the input and predictive space.
• Essential for long-term optimization if beginning from a biased or imbalanced dataset.
• Supports discovery of novel solutions outside previously tested regions.

Limitations:

• Exploratory runs may yield suboptimal output optimization in the short term.
• Resource-intensive if performed too often without balancing with exploitative suggestions.

	input_1	input_2	input_3	input_4	input_5	input_6	input_7	input_8	output_1	output_2
0	103.0	DMPU	7.11	0.8	DPPP	0.037	(TMS)3SiH	0.13	66.856 ± 56.88	2.857 ± 46.75
1	110.0	DMSO	6.26	1.0	DPPP	0.016	PMHS	0.58	14.157 ± 47.48	41.178 ± 69.18
2	102.0	NMP	7.63	0.8	L33 | XantPhos	0.037	(TMS)3SiH	0.11	33.826 ± 63.89	2.006 ± 47.56
3	118.0	NMP	6.44	0.8	L33 | XantPhos	0.037	PMHS	0.10	19.471 ± 58.4	1.212 ± 38.41
4	101.0	DMAc	6.44	0.8	L29 | DPPF	0.037	(TMS)3SiH	0.11	25.118 ± 62.01	4.14 ± 48.07

Table 2: Table of exploratory experiments proposed to expand coverage of the predictive space, focusing on high-uncertainty or under-sampled regions.