Skip to content
English
Contact us
  • There are no suggestions because the search field is empty.

Understanding Optimized Suggestions, Exploratory Suggestions, and Risk-Averse Suggestions in a finished Campaign

When to run each of these and how these data points are most likely to impact your model.

When you configure a campaign iteration in SuntheticsML, the platform can generate three types of suggested experiments: Optimized, Exploratory, and Risk-Averse — each serving a distinct purpose in guiding your model toward optimal conditions.
 
Understanding the role each suggestion type plays helps you make informed decisions about which experiments to prioritize at each stage of your campaign.
 

Note: Risk-Averse suggestions are only available with the Lithium ML algorithm, not Helium or older algorithms.

 

Suggestion types

1. Optimized Suggestions

Default suggested: 5 experiments
 
Available with these Campaign Goals: Balanced Optimize, Optimize
 
Where are they are located: Experiment Planning -> Suggested Experiments
 

Optimized suggestions are designed to drive toward your system's optimal region as quickly as possible. A slight degree of exploration is built into these experiments, which means they balance exploitation of what the model already knows with just enough exploration to avoid getting stuck in local optima.

Collecting data for these suggested experiments will help your model converge on optimized conditions for your outputs with the fewest iterations needed.

Primary Goal

  • Converge on output-optimal conditions quickly

Model impact

  • Accelerates convergence with each iteration

Best used when

  • You want to make rapid progress toward campaign targets

Exploration built in?

  • Yes — slight degree of exploration is included

Optimized Suggestion Experiments Table - Lithium

 

Tip: The optimized suggestions are ranked, so #1 is considered a better recommendation than #2, etc.

If experimental runs are limited, prioritize the top suggested experiments first as these will be most beneficial for the model's active learning in your next iteration.

 

2. Exploratory Suggestions

Default suggested: 1 experiment
 
Available with these Campaign Goals: Balanced Optimize, Explore
 
Where are they are located: Experiment Planning -> Suggested Experiments
 
Exploratory suggestions target gaps in your training data — regions of the parameter space where the model has little or no existing data to train on. Think of them as experiments designed to fill in the "holes" in your data. Collecting data for exploratory suggestions reduces model uncertainty and improves uncertainty and error metrics in subsequent iterations, making the overall model more reliable across the full input space.
 

Primary Goal

  • Fill gaps in the training data parameter space

Model impact

  • Reduces uncertainty; improves error metrics in future iterations

Best used when

  • Model uncertainty is high or error metrics are poor

Exploration built in?

  • Yes — exploration is the primary purpose
 Exploratory Experiments Table - Lithium
 

 

3. Risk-Averse Suggestions

Default suggested: 0 experiments (must be increased manually)
Available with these Campaign Goals: Balanced Optimize, Risk-Averse
Where are they are located: Experiment Planning -> Suggested Experiments
 

Tip: In order to do Risk-Averse Optimization, the campaign requires at least 1 numerical input.

Risk-averse suggestions target regions of the input space where the predicted output is stable across nearby conditions. Rather than chasing the absolute peak of model performance, these experiments avoid "edges" — areas where a small change in inputs could cause a large, unexpected drop in output performance. The result is robust, reproducible conditions that mitigate the risk of performance drop-offs in practice.

Primary Goal

  • Find stable, robust operating conditions

Model impact

  • Identifies input regions with low sensitivity to small changes

Best used when

  • Reproducibility matters more than peak performance

Requirement

  • At least one numerical input variable in the campaign

Risk-Averse Suggestion Experiments Table - Lithium

Tip: In most campaigns, running all five optimized suggestions each iteration is the fastest path to convergence. Add exploratory experiments when model uncertainty remains high after multiple iterations, and consider risk-averse suggestions when process robustness is a priority — for example, when conditions will be transferred to manufacturing or repeated across sites.

At a glance

SUGGESTION TYPE WHAT IT TARGETS METRIC IT IMPROVES WHEN TO PRIORITIZE

1. Optimized

 Predicted optimal region of output space Output performance; convergence speed Most iterations — especially early-to-mid campaign
2. Exploratory Under-sampled regions of input space Model uncertainty; error metrics When uncertainty is high or coverage is sparse
3. Risk-Averse Stable, low-sensitivity regions of input space Robustness; reproducibility When transferring to scale-up or requiring consistency