A novel theoretical framework unravels how processes in complex systems that occur at different timescales are coupled together at the functional level by sharing information.

An analytical framework describing ecosystems in which species interactions drive large population fluctuations provides a way to address fundamental questions about this dynamical state.

Achieving realistic subthreshold variability in a biophysical neuronal model requires low-level synchrony in its synaptic input drive, a finding that challenges current theories to explain spiking activity in cortical neurons.

Disease contagion is suppressed when different social groups have a large overlap in membership.

A new model reproduces both the dynamical and steady-state behavior of a group of living organisms, a first for such systems.

Nonequilibrium activity may provide a surprisingly general way to improve the ability of a system to store and retrieve memory.

Models of systems in physics usually start with elementary processes. New work with a neural network shows how models can also be built by observing the system as a whole and deducing the underlying interactions.

A simplified model of a family of machine-learning architectures offers a way to explore a major form of training failure—mode collapse—that is not well understood.

A mathematical model shows how increased intricacy of cognitive tasks can break the mirror symmetry of the brain’s neural network.

A simple twist on a mainstay tool for analyzing the dynamics of disordered systems provides a way to describe out-of-equilibrium properties, which are traditionally much harder to obtain.

A wide class of physical systems could be turned into learning machines, thanks to a new general approach to training them based entirely on physical dynamics combined with a time-reversal operation.

A new framework for analyzing forces and currents in nonequilibrium systems generalizes existing graph-theoretical tools to now encompass interacting reaction networks and time-dependent properties.

By describing network topology using an underlying geometric space, spatial Turing patterns can be found in the geometric embeddings of real networks.

The “six degrees of separation” are the property of the equilibrium state of any network where individuals weigh their aspiration to improve their centrality against the costs incurred in forming or maintaining connections.

A deep neural network learns feedback-control protocols that convert information obtained from measuring a fluctuating nanosystem into heat or work.

A new methodology for analyzing the 3D distribution of galaxies borrows techniques from the study of colloids and other disordered materials.

A new approach to applying a power-law model to describe extreme events avoids traditional pitfalls and offers a more robust approach to predicting and mitigating risk.

Methods from dynamical systems and manifold learning offer a simpler, physically sound framework for evaluating and improving upon climate models.

When the heart starts to beat irregularly, mechanical contraction of the heart muscle does not simply follow electrical excitation waves but exhibits more complex disorganization.

A metric for neural networks’ similarity, based on synaptic differences, provides accurate predictions of how novel networks function, thus identifying a key relation between structure and function.

Graph theory provides universal algorithms that can be used to control stochastic biological systems at any scale, from single proteins to the evolution of whole populations of organisms.

Random walkers (such as particles, cells, or animals) that interact attractively or repulsively with their own paths exhibit memories that aid the exploration of their domains.

An analysis of neural network models with biophysically realistic descriptions of synaptic connections shows that irregular neuronal activity—observed in experiments—emerges naturally from interactions between cells.

Gradually adding metadata to a complex network causes a crossover in the ability to infer properties and make predictions about that network.

Quantum information processing can provide a significant competitive advantage for any system that must adapt to its environment, an enhancement that scales without bound.