Regensburg 2022 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
BP: Fachverband Biologische Physik
BP 2: Computational Biophysics and Neuroscience
BP 2.9: Vortrag
Montag, 5. September 2022, 12:00–12:15, H13
Available processing time regulates optimal balance between sensitivity and precision — Sahel Azizpour1, •Johannes Zierenberg2, Viola Priesemann2, and Anna Levina3 — 1Donders Institute for Brain, Cognition and Behavior, Nijmegen, Netherlands — 2Max Planck Institute for Dynamics and Self-Organization, Göttingen Germany — 3Eberhard Karls University of Tübingen, Tübingen, Germany
Solving everyday tasks naturally leads to a trade-off between the time spent on processing some input and the accuracy of the outcome. In particular, fast decisions have to rely on uncertain information about inputs. However, standard estimates of information processing capabilities, such as the dynamic range, are defined based on infinite-time averages that do not incorporate noise effects from finite processing times. Here, we develop estimates of processing capability that explicitly account for noisy outputs. We use these measures to show that limiting the processing time in recurrent neural networks can drastically affect the sensitivity and precision of outcomes. This way, optimal dynamical states shift away from the conventionally expected critical point toward subcritical states for finite processing times. Our results thus highlight the necessity to explicitly account for processing times in future estimates of information processing capabilities.