BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Combining data-driven and physics-based approaches to predict\, un
 derstand\, and control active matter dynamics - Michael Hagan (Brandeis Un
 iversity)
DTSTART:20231107T150000Z
DTEND:20231107T160000Z
UID:TALK208135@talks.cam.ac.uk
DESCRIPTION:Active matter is composed of particles that generate forces\, 
 which leads to spectacular emergent dynamics resembling the lifelike prope
 rties of biological organisms. Yet\, active materials exhibit diverse dyna
 mical states\, most of which have chaotic dynamics that do not produce wor
 k or other functions. Designing or controlling an active material to selec
 t a state corresponding to a desirable function requires accurate dynamica
 l models. However\, developing models using traditional statistical physic
 s approaches is challenging because active materials lack the scale separa
 tion characteristic of equilibrium systems.\nIn this presentation\, I will
  discuss efforts to use data-driven techniques to address this challenge i
 n the context of a model active material\, microtubule-based active nemati
 cs. I will describe two complementary approaches. In the first\, we have a
 dapted a method to discover optimal continuum models directly from spatiot
 emporal data\, using sparse regression. We have identified several approac
 hes to mitigate measurement errors in the data. We find that the method ca
 n reveal the relative contributions of different physical mechanisms\, and
  quantitatively estimates key experimental parameters. In the second appro
 ach\, we have used deep learning to automatically learn and forecast activ
 e nematics dynamics\, using data from particle simulations and experiments
 . We find that the method can predict spatiotemporal dynamics including th
 e spontaneous creation and annihilation of defects\, but that inaccuracies
  arise from measurement errors in this complex system. Further\, I will di
 scuss reduced-dimensional representations of the forecaster\, which reduce
  training time and may facilitate human interpretation. In the second appr
 oach\, we have adapted a method to discover optimal continuum models direc
 tly from spatiotemporal data\, using sparse regression. We have identified
  several approaches to mitigate measurement errors in the data. We find th
 at the method can reveal the relative contributions of different physical 
 mechanisms\, and quantitatively estimates key experimental parameters\, e.
 g. how the &lsquo\;activity&rsquo\; depends on ATP concentration. I will a
 lso describe a tensor-based formulation of the method for 3D systems\, and
  its application to 3D simulations of dry active nematics.\nTime permittin
 g\, I will discuss how models discovered by these approaches can be combin
 ed with control theory to drive active materials toward particular emergen
 t behaviors.\n&nbsp\;
LOCATION:Seminar Room 2\, Newton Institute
END:VEVENT
END:VCALENDAR