Optimisation under uncertainty

Making the right decisions can be difficult when there are only a few variables and possible outcomes. Making complex decisions in the face of multiple moving parts, integrated systems, and a variety of time-based considerations is incredibly complex, and the solutions that can help make those decisions are sometimes equally as difficult to get a handle on. Furthermore optimization and decision-making under uncertainty involves other challenges, such as large numbers of scenarios and complex mathematical models.

ut logo The IBM Research Uncertainty Toolkit was developed to help businesses resolve all of those challenges. The toolkit is a suite of user-friendly tools for optimization under uncertainty, which allow one to analyze the tradeoff between robustness and cost-effectiveness of a planning decision, and to create, evaluate and select alternative solutions, including solutions hedged against uncertainty through robust and stochastic optimization. The toolkit has been developed as a plugin to the IBM Decision Optimization Center and as independent libraries.

Susara van den Heever Martin Mevissen Rudi Verago Marco Laumanns Bissan Ghaddar Chungmok Lee
Nicole Taheri Radu Marinescu Ban Kawas Ali Koc


Here are the five steps to make resilient decisions in the face of uncertainty, and how the Uncertainty Toolkit can support that process.

  1. Define the decision model: Outline the objective as well as the decisions that need to be made and any constraints you may have. You can create many different scenarios, and compare the outcomes of those scenarios.
  2. Characterize uncertainty: Define the parts in your scenario that are in flux. This includes rising and falling prices, shifting economic forecasts, and any other data that will change. Using the Uncertainty Toolkit, you can select uncertain data, determine whether it is a scenario or a range, define the decision stages, and choose the risk measures – for example, the worst-case outcome.
  3. Generate uncertain models: Once you have entered all the data, the Uncertainty Toolkit automatically generates the uncertainty models based on your choices. You can choose robust optimization, which hedges against the worst-case outcomes, stochastic, which optimizes for expected outcomes, or a model that allows a comparison of each, so you can visualize the tradeoffs.
  4. Generate plans: The Uncertainty Toolkit generates multiple solutions based on your choices, and automatically performs solution-scenario cross-comparison, so you can see the impact of a change on each plan.
  5. Analyze trade-offs: Using these plans, you can make your decisions. There is not one right plan, but many—you can drill down into the data to compare them all. Use built-in visual tools to run trade-off analyses across many different plans and scenarios and find ways to reduce risks and improve KPIs.

UT steps


The toolkit involves two types of users who work closely together:


Creating an user-friendly toolkit involves many challenges in terms of architecture the system must deal with: a large amount of data and scenarios, complex optimization models, many different types of uncertainty characterizations and different methods of dealing with them and parallelizable process.

The figure below shows the recommended architecture for the toolkit, designed to address the combined challenges of usability, automation and scale.

Components and tools

Use cases
  • Increase solution stability & value
  • Easy integration into an existing application
  • Commission your own applications
Related links