Highlights and Insights From GECCO 2016

There is perhaps no conference with as much density of expertise in genetic and evolutionary computation as last week’s GECCO event in Denver. We were impressed and inspired by the five days we spent there and want to share a few salient themes that ran through the conference.

Here are a few of our key takeaways:

– Neuroevolution continues to expand: There were significant advancements in this field, both in terms of methodology and practical, real-world use-cases. For the former, combining neuroevolution with deep learning architectures is a new trend. Examples included evolving Compositional Pattern Producing Networks (CPPNs) to design convolutional nets, evolving Neural Turing Machines to express memory, and extreme mini-batching to scale up to large datasets and network. Additionally, we heard about substantive successes in using neuroevolution for real-world problems like protein folding and power plant control.

– The growing prevalence of novelty search: Are we witnessing the fall of pure objective-based evolutionary computation? We saw novelty search spreading across nearly every GECCO subfield with, again, some positive real-world applications. Related techniques that did not explicitly reference novelty but placed an enhanced focus on harnessing rich phenotypic information were also prevalent.

– Algorithm optimization with evolutionary computation: Another trend we noticed was the use of evolutionary computation to optimize, configure, and correct algorithms. This trend was evident in particular through the keynote lectures of Hoos and Forrest, a tutorial by Stützle, and the Automated Design of Algorithms Workshop. The idea is that humans can take care of approximate, high-level algorithm design, while the details are better left for automated optimization.

And that was just the tip of the iceberg. There was a lot of interest in taking evolutionary computation out of the lab and into the real world, as well as how exactly research in industry and academia could leverage each other.

Suffice it to say, it was a really interesting five days in Denver. We’d like to send a hearty thanks to the folks we met at GECCO and all the great speakers and presenters. If you’d like to see what we were doing there, below, find a video that highlights some of the research we presented at GECCO as well as some other recent research by our Sentient Labs team.

If you’re in the field and want to leverage the world’s largest distributed compute network to solve real-world, practical problems, we’re always looking for great talent.

And, no matter what, we’ll see you next year.