Evolution and Analytics: An Interview with Dr. Gary Fogel

Earlier this year I had the opportunity to sit down with Gary Fogel, CEO of Natural Selection, Inc. to discuss a data analytics approach that his firm specializes in called evolutionary computation.

Evolutionary computation encapsulates the process of natural selection into a computer by following a process which 1) generates a population of solutions 2) scores and ranks these solutions 3) retains and modifies the solutions which perform well and finally 4) iterates this process thousands of times to find the predictive models with the best performance.

I found our chat to be highly informative and thought provoking, particularly with regard to some of the types of applications this approach is best suited for.  Excerpts from our talk are featured in the video below.

 

Some core insights from the talk include:

• Evolutionary computation is valid not only for comparing learning methodologies, but also for optimizing within them.  Gary provides insight on how the approach can be used in cluster analysis (2:54), decision tress (3:49), and neural networks (4:17).

• A primary advantage of the approach is that you can look data with large feature sets using a broad set of solutions and search them in ways that people have not necessarily considered before.  The methodology can evolve its own answers without relying on pre-programmed knowledge.  This flexibility allows researchers to find patterns and relationships in data that might not have been previously realized or inferred (7:32).

• The methodology is highly dynamic allowing for it to engage in continuous learning as environmental conditions change (9:23).

• Evolutionary computation works through a flow of generating solutions, fitness scoring, ranking/selection, variation, termination and then iteration (10:06).

• Approach can provide insight and adapt even when importance of data features and importance of combinations of these features changes over time (12:16).

• NSI has successfully applied evolutionary computation methods for homeland security applications such as risk management for container ship traffic (14:33) and safety of food imports (16:25), life science projects in drug design, cancer diagnostics and personalized medicine (19:20) and in routing and logistics optimization projects for both military and industry (20:32).