Editorial: Emergent Neural Computation from the Interaction of Different Forms of Plasticity

Matthieu Gilson, Cristina Savin, Friedemann Zenke

Introduction

In the field of computational neuroscience, the classic Hebbian postulate—"neurons that fire together, wire together"—has long served as the foundational principle for understanding learning and memory. However, recent research suggests that Hebbian plasticity, while essential, is insufficient for maintaining a stable and functional brain. This editorial, published in Frontiers in Computational Neuroscience, synthesizes a collection of studies arguing that complex neural computation emerges not from a single rule, but from the concerted interaction of multiple, diverse plasticity mechanisms.

The Stability Challenge: Beyond Hebbian Dynamics

The primary challenge in modeling brain circuits is explaining how they remain stable despite constant synaptic changes. Pure Hebbian plasticity creates a positive-feedback loop: as connections strengthen, firing rates increase, which in turn further strengthens the connections. Without counter-mechanisms, this leads to "runaway dynamics," where neural activity eventually saturates or results in pathological, seizure-like states.

To address this, researchers are investigating how different "flavors" of plasticity—operating across various timescales and targeting distinct neuronal populations—work in tandem to create efficient, self-organizing circuits.

Mechanisms of Interaction

The research collection highlights several key mechanisms that interact with Hebbian learning to ensure functional stability:

1. The Paradox of Fast Homeostasis

Homeostatic plasticity acts as a "check and balance" system, keeping neuronal activity within a healthy range. While biological experiments often observe homeostatic processes occurring over hours or days, theoretical models suggest that stability requires much faster intervention. This "Fast Homeostasis" paradox suggests that the brain may employ rapid, yet-to-be-fully-characterized mechanisms, such as heterosynaptic plasticity or rapid threshold sliding, to prevent network crashes during intense learning phases.

2. Inhibitory Plasticity as a Functional Catalyst

While excitatory connections often receive the most attention, plasticity at inhibitory synapses (iSTDP) is increasingly recognized as a "secret sauce" for neural computation. iSTDP is crucial for maintaining the Excitation/Inhibition (E/I) balance. By sharpening neuronal responses and "decorrelating" activity, inhibitory plasticity allows networks to distinguish between highly similar input patterns, enhancing the brain's discriminatory power.

3. Metaplasticity and Error Correction

Metaplasticity—the process by which the "rules of learning" themselves change—provides a potential bridge between biological learning and the training methods used in artificial neural networks. Some models suggest that metaplasticity can adjust short-term synaptic properties to minimize overall circuit error, offering a biologically plausible mechanism for supervised-like learning.

4. Neuromodulation and Unsupervised Learning

The interaction between reward signals (such as dopamine) and homeostatic rules allows unsupervised networks to achieve performance levels comparable to supervised systems. This synergy explains how brain circuits can learn complex behaviors and decision-making tasks without an explicit external "teacher" for every input.

Computational Outcomes and Key Results

The integration of these diverse plasticity forms leads to several sophisticated computational properties:

  • Pathway Selectivity: The combined action of excitatory and inhibitory STDP allows feed-forward networks to develop highly selective signaling pathways, effectively filtering out background noise.
  • Internal Representations: Networks utilizing E/I plasticity can develop internal models of the external world, facilitating non-linear computations that are impossible for simpler, single-rule networks.
  • Stable Winner-Take-All (WTA) Dynamics: The interaction of different plasticity mechanisms enables recurrent networks to autonomously develop WTA circuits, which are fundamental for categorization and decision-making.
  • Working Memory: Models that combine dopamine-modulated STDP with homeostatic mechanisms have successfully replicated the coding properties of neurons in the prefrontal cortex, providing a theoretical framework for how working memory is maintained.

Conclusion and Future Directions

While the research presented in this editorial marks a significant step toward understanding the multi-faceted nature of brain plasticity, a "unified theory" remains elusive. Future work must address the biological plausibility of error-calculation models and the inherent mathematical complexity of multi-scale systems.

Furthermore, the mystery of "slow" homeostasis remains: if fast mechanisms are required for stability, what is the functional role of the slower processes observed in laboratory settings? One hypothesis is that they prevent "interference" between old and new memories, a question that continues to drive the next generation of computational neuroscience research.

Summary updated

Abstract

More than 60 years later, Hebb’s prophecy “neurons that fire togethe wire together” (Hebb, 1949; Shatz, 1992) prevails as one of the cornerstones of modern neuroscience. Nonetheless, it is becoming increasingly evident that there is more to neural plasticity than the strengthening of synapses between co-active neurons. Experiments have revealed a plethora of synaptic and cellular plasticity mechanisms acting simultaneously in neural circuits. How such diverse forms of plasticity collectively give rise to neural computation remains poorly understood. The present Research Topic approaches this question by bringing together recent advances in the modelling of different forms of synaptic and neuronal plasticity. Taken together, these studies argue that the concerted interaction of diverse forms of plasticity is critical for circuit formation and function. A first insight from this Research Topic underscores the importance of the time scale of homeostatic plasticity to avoid runaway dynamics of Hebbian plasticity. While known homeostatic processes act slowly, on the timescale of hours to days, existing theoretical models invariably use fast homeostasis. Yger and Gilson (2015) review a body of theoretical work arguing that rapid forms of homeostatic control are in fact critical for stable learning and thus should also exist in biological circuits. Following a similar line of thought, Chistiakova et al. (2015) review experimental and theoretical literature which suggests that th...
BibTeX
@article{gilson2015editorial,
  author = {Matthieu Gilson and Cristina Savin and Friedemann Zenke},
  title = {{Editorial: Emergent Neural Computation from the Interaction of Different Forms of Plasticity}},
  journal = {Frontiers Comput. Neurosci.},
  year = {2015},
  doi = {10.3389/fncom.2015.00145},
}
Figure