Research

Put broadly, I am currently interested in knowing what information we can extract from complex, networked, dynamical systems simply by observing their behavior over time. For example, if we record the firing of the neurons of an animal's brain, is it possible to know how they are wired? If we record the temperatures of several geysers in the Yellowstone National Park over a long period of time, can we predict how the geysers are influencing each other's eruptions? Like these examples, a direct view of the inner workings of many such complex systems is impossible. It is hard to dissect small, delicate brains of lab animals and observe all the neuronal wirings there in microspic details. It is impossible to go underground to explore how the Yellowstone geysers are hydraulically connected. Rather, we can only passively observe how these systems behave over time. Thus, a common question is what information do these dynamical behaviors reveal about different causal interactions among the components of these complex systems? This is why I am developing new computer codes based on artificial intelligence and machine learning that tell us how the inner workings of complex systems can be backtraced from their dynamics.

I am applying special types of machine learning architechtures called ``reservoir computers" to decipher short-term and long-term (that is, time-delayed) causal interactions among the components of a networked system. These reservoir computers have the special ability to mimic and predict the dynamics of any network they are exposed to. In doing so, they are able to build an in silico model of the network inside them and it is that model which I use as a proxy of the actual network to predict how parts of the network are connected in reality. Click on the individual projects above to learn more about how my method is applicable to a wide variety of systems, ranging from worm brains and Yellowstone geysers to opto-electronic feedback networks in the lab.