From time to time we produce some fun videos. Below is a recent sample.

The knowledge gradient policy (click here)

The knowledge gradient policy is a way of taking a series of measurements to learn a function (such as, the total costs estimated by a business simulator as a function of the price or number of pieces of equipment). This video shows how measuring one part of the function allows us to update other parts of the function (that is the easy part). What is less obvious is that we use information about these relationships to determine where to measure. So, we might measure a point that we do not think is optimal, but because we will learn the most about the function (more precisely, we are going to maximize the expected value of the function as a result of the measurement). The result is a much faster rate of convergence when searching for the best parameter or policy.

Energy policy modeling (click here)

We are developing an optimization/simulation model of energy resource management. In this animation, we show the flows of different types of energy resources at the bottom (coal, nuclear, wind, solar, hydro, natural gas, oil, ...) through different nodes representing storage and conversion, until it meets various types of demands at the top. The model steps forward in hourly increments, but the model is designed to run over multiple decades. This video is fairly short, primarily reflecting hourly wind variations, along with solar patterns and rainfall.

Fleet management (click here)

This animation is derived from one of our major transportation contracts. When viewed this way, we are reminded that these are large, dynamic systems, which exhibit their own physics. One of our challenges is identifying how these dynamic systems respond to changes in inputs.