PRX Life

Active persistent dynamics are shown to mimic some aspects of Hebbian unlearning — suggesting that nonequilibrium dynamics can provide an alternative way to improve associative recall in neural network models.

A cytoskeleton-inspired mechanical network with mechanosensitive proteins and motors performs contrastive learning from environmental cues via strain-rate dependent reactions and active forces, while maintaining learned responses despite continuous molecular turnover.

A study of bacterial movement across a range of both unconfined and highly confined environments establishes a quantitative link between behavioral rules and environmental context and reveals how E. coli can navigate complex habitats.

Resonant frequencies prioritize multi-looped structures or higher-level loops close to the driving source, while the network is prone to vessel shunting away from resonances — underscoring the importance of short-term pulsatility on long-term structures in adapting, periodically driven elastic flow networks like mammalian vasculature.

This study proposes that olfactory systems can adapt to rapidly fluctuating odor backgrounds using a manifold learning mechanism and demonstrates how this mechanism can outperform traditional predictive filtering methods.

This review article proposes that Marr’s three levels of analysis — the computational problem, the algorithms used, and their molecular implementation — can be a useful framework for studying developmental processes.

By examining collective patterning in a tractable model, this study reveals a trade-off between speed and accuracy and identifies counterintuitive properties of collective behavior, for instance optimized strategies do not always maximize information transfer.

A recurrent neural network model containing neurons with intrinsic plasticity measures how neuronal excitability varies with plasticity under different synaptic inputs — illuminating how these dynamics influence heterogeneity in neural responses.

This study details a data-driven implementation of Fisher’s geometric model — providing an empirically grounded approach for understanding evolutionary dynamics using deep learning methods.

Inspired by Crick’s molecular memory proposal, this study analyzes a nonequilibrium protein model and identifies molecular automata that can be exploited for computation, including in molecular stopwatches and error-tolerant memory.

An enhanced form of mechanical proofreading during dynamic cell-cell contact helps reconcile robust clonal selection with tunable specificity in adaptive immune responses.

E. coli integrate environmental history through scale-free memory, performing recurrent, neural-like computation to balance growth and survival in fluctuating nutrient conditions.

Experiments in a unicellular relative of animals validate the predictions of a model which treats organisms as Markovian Bayesian agents that learn from environmental states, providing insights into the link between within-lifetime learning and generational adaptation.

Mapping evolutionary dynamics to Bayesian learning shows that organismal complexity tends to match environmental complexity, but rapid environmental changes may favor simpler forms.

Using disordered solids as a model, this work demonstrates how cyclic environmental change can produce robust system memory.

A simulated cortical network rapidly establishes functional circuitry through autonomous learning from mechanically coupled sensors.

Tuning elastic networks for bidirectional allostery improves cooperative binding, while a modified tuning process ensures robustness against thermal fluctuations by raising the crossover temperature.

By mimicking biological brain noise, this work shows that activity-dependent fluctuations drive the emergence of modular architectures in artificial neural networks, improving robustness and generalization.