The Elusive Laws of Synaptic Plasticity

This piece, by Onno Berkan, was published on 03/04/25. The original text, by Confavreux et al., was presented in the 2023 NeurIPS Conference.

This joint Institute of Science and Technology Austria and University of Tübingen study focuses on understanding how connections between neurons (i.e., synapses) change and adapt, in a synaptic plasticity process crucial for learning and memory. While this topic has been studied for decades, the exact laws governing elasticity remain unclear. Due to the difficulty in studying synapses in vivo (ie, in living beings), many current methods rely on theoretical predictions.

The researchers developed a new systematic approach to study plasticity rules in complex neural networks. Their novel method, called filter simulation-based inference (fSBI), can handle more complicated scenarios and is more efficient. This method is meant to narrow the pool of potential laws into a small percentage of candidates.

The study created artificial neural networks that included excitatory and inhibitory neurons connected randomly with a 10% connectivity rate. These networks also received input from external neurons firing at random rates between 5 and 15 times per second. This setup helped create a more realistic model of how actual brain networks might function. The study’s main improvement over current literature in this field is its lack of reliance on intuition to narrow the possibility pool down, instead relying on accuracy alone.

One of the most interesting findings was that the researchers discovered new types of plasticity rules that were quite different from what has been observed in traditional laboratory experiments. These rules showed stable activity patterns and weight dynamics, demonstrating that there might be more ways for neural networks to achieve stable functioning than previously thought.

The researchers used a systematic filtering process to identify which plasticity rules would work effectively. They started with broad criteria and progressively applied more restrictive metrics to ensure they found rules that maintained stable network activity while allowing for proper learning and memory formation. This methodical approach helped them identify rules that maintained appropriate firing rates for excitatory and inhibitory neurons while stabilizing the network's overall activity.

The study represents a significant step in understanding brain plasticity, introducing a more principled and comprehensive approach to studying how neural networks might learn and adapt. By providing a new framework for studying plasticity rules, this research could help bridge the gap between theoretical models and actual brain function, potentially leading to better understanding of learning and memory processes in the brain.

Want to submit a piece? Or trying to write a piece and struggling? Check out the guides here!

Thank you for reading. Reminder: Byte Sized is open to everyone! Feel free to submit your piece. Please read the guides first though.

All submissions to berkan@usc.edu with the header “Byte Sized Submission” in Word Doc format please. Thank you!