Extent to which transient-dynamics-based computation can support learning

Determine the extent to which computing with transients—i.e., transient trajectory-based computation away from steady-state attractors—can incorporate learning, particularly on-the-fly (online) learning and life-long learning, in biological and artificial systems.

Background

The paper argues that attractor-based computational frameworks are insufficient to explain real-time, flexible, and robust information processing in living systems. Instead, it advocates computation based on transient dynamics away from attractors, highlighting mechanisms such as ghost states at criticality that can provide quasi-stable working memory and responsiveness to time-varying inputs.

Building on this perspective, the authors note that a comprehensive framework for natural computation must also encompass learning—especially online and life-long learning. While they cite evidence suggesting quasi-stable transient structures (e.g., slow points/ghosts) may support learning in artificial and natural neural systems, they emphasize that the general extent and mechanisms by which transient-based computation enables learning remain unresolved.

References

To which extent computing with transients can incorporate learning, and in particular on-the-fly and life-long learning, is another exciting open question.

— Biological computations: limitations of attractor-based formalisms and the need for transients  (2404.10369 - Koch et al., 2024) in Section 'Computations at criticality as a possible road forward' (final paragraph)

An important direction for future work is to extend this framework to learning rules that modify the recurrent connectivity itself, rather than only the readout and feedback pathways considered here. Such plasticity would allow the network to develop persistent internal structure and long-term memory, raising the question of how previously acquired dynamical states influence subsequent learning dynamics and the ability of the network to acquire new tasks.

— Learning-Induced Dynamical Transition in Recurrent Neural Networks  (2609.19288 - Vaidya, 16 Sep 2026) in Conclusion