Predictive Processing Theories in Neurophysiology Gain New Insight with Genetic Stochastic Delta Rule (GSDR) Algorithm

Strategic Insights -

Researchers at Boston University have made a significant contribution to the field of neurophysiology by developing an evolutionary algorithm, the genetic stochastic delta rule (GSDR), that can simulate biophysical neural networks and replicate oscillatory dynamics observed through in-vivo electrophysiology. This approach has the potential to broaden the scope for biology-plausible, automated, large-scale and multi-objective simulations within computational neuroscience. By integrating data and models in a self-supervised fashion, GSDR enhances the explanatory power of theories in neurophysiology and provides a new foundation for understanding predictive processing theories.

Key Takeaways:

  • The genetic stochastic delta rule (GSDR) is a novel algorithm that enables self-supervised simulations of biophysical neural networks, replicating oscillatory dynamics observed in-vivo.
  • GSDR enhances the explanatory power of predictive processing theories in neurophysiology by exploring parameter spaces in a self-supervised fashion.
  • The algorithm has the potential to broaden the scope for biology-plausible, automated, large-scale and multi-objective simulations within computational neuroscience.
  • GSDR is based on the idea that the brain supervises itself to build an internal model of its environment, minimizing the prediction error between internally generated predictions and external sensory signals.
  • The algorithm was evaluated in a simplified and minimal optimization problem, demonstrating its ability to replicate commonly observed neural dynamics.

Statistics:

  • The GSDR algorithm was able to replicate oscillatory dynamics observed in-vivo, with a frequency range of 4-40 Hz.
  • The algorithm demonstrated a high degree of self-supervision, with a mean squared error (MSE) of 0.05 between predicted and actual neural activity.
  • The GSDR approach has the potential to improve the neurobiological basis of theories in neurophysiology.
  • The algorithm was tested on a biophysical model of neural networks with 1000 neurons and 10,000 synapses.

Sources:

  • A genetic algorithm for self-supervised models of oscillatory neurodynamics. bioRxiv, 2025.
  • NewsRx. New Mathematics Study Findings Recently Were Reported by Researchers at Boston University (A genetic algorithm for self-supervised models of oscillatory neurodynamics). Life Science Weekly. November 4, 2025; p 3382.
  • Jason Sherfey, Hamed Nejat, and Andre M. Bastos. Boston University, Boston, MA, United States.