Carlos Floyd: How biological systems can achieve control without global information

Abstract: Controlling non-equilibrium dynamical systems is a key challenge in both engineering and biology. Typical engineering approaches require extensive computation and near-complete knowledge of the system's state and dynamics, which biological controllers such as cells generally lack. How then can biological networks achieve robust control while relying only on imperfect, local information? In this talk, I will show that effective control of complex non-equilibrium systems does not require global, top-down strategies or precise system models. Just as approximate gradients are useful for optimization provided they remain descent directions, spatiotemporally local learning algorithms can succeed with imperfect feedback provided they are correlated enough with the true objective. I will demonstrate this across a several physical systems, including active fluid models of tissue dynamics where model-free reinforcement learning using only coarse, low-dimensional observables enables precise defect control without accurate parameterization. I will then identify response constraints that can explain why such simple strategies work: for broad classes of perturbations, non-equilibrium steady-state responses have fixed signs across all driving strengths, and a graph-theoretic property called "coherence" ensures monotonicity even when inputs modulate multiple rates. These constraints guarantee that simple linear feedback on small subsets of kinetic rates achieves globally stable control without fine-tuning. Together, these results illustrate a broader principle: approximate solutions to complex control problems can succeed by exploiting constrained responses and lower-dimensional descriptions, allowing biological systems to bypass the need for global information. Bio: Carlos Floyd is an AI + Science postdoctoral fellow of the Data Science Institute at the University of Chicago, working with Suri Vaikuntanathan and Aaron Dinner.  He is an incoming Assistant Professor in the Physics department at Carnegie Mellon University. He received his Ph.D. in Biophysics from the University of Maryland, College Park, where he worked with Christopher Jarzynski and Garegin Papoian studying nonequilibrium thermodynamics and self-organization in cytoskeletal networks. His current research uses theory, simulation, and machine learning to understand how disordered biomolecular networks can perform computation and control.