arXiv subjects

Steffen Rulands

Publications and source records attributed to Steffen Rulands.

14 recordsLinked to original sources

arXiv2026

Emergent interactions lead to collective frustration in robotic matter

Artificial intelligence and robotic systems are increasingly deployed as interacting collectives of learning agents. This raises the question of whether robotic matter, where many learning agents interact, shows the emergence of collective behaviour. Here we study a paradigmatic model of robotic matter and show the emergence of a range of complex, collective behaviours. Specifically, we study systems composed of stochastic interacting particles, each endowed with a deep neural network that optimises transitions based on its environment. In a one-dimensional system, we show that robotic matter exhibits complex phenomena arising from emergent interactions, including self-organisation into distinct temporal learning regimes, particle species, and long-lived frustrated states with suboptimal reward. We further identify an abrupt, density-dependent change in collective behaviour. Active matter theory suggests that this phenomenon reflects a phase transition with signatures of criticality. Our results establish robotic matter as a platform for novel non-equilibrium physics.

cond-mat.soft↗

arXiv2026

Model-Aware Schedules Improve Generation via Fiberwise Optimal Transport

Diffusion and flow-matching schedules control the signal and noise coefficients that mix data and noise along affine probability paths. Minimizing a kinetic action defined on coefficient paths, motivated by optimal transport, helps explain strong baselines but remains model-agnostic and ignores prediction error. Here we introduce a model-aware schedule construction based on fiberwise optimal transport. At a fixed time and state on the probability path, compatible signal/noise decompositions form an affine fiber. We define a fiberwise prediction risk by averaging optimal-transport costs between the true and predictor-induced decompositions within these fibers. On a fixed coefficient curve, combining this risk with coefficient-path kinetic action yields a closed-form optimal time allocation. This construction extends to general linear prediction targets, and the risk profile can be estimated from an early baseline checkpoint. We evaluate DDPMs and flow matching across prediction targets, training configurations, risk-estimation checkpoints, datasets, and architectures. Our model-aware schedules consistently outperform strong baselines, including a 38.6% relative FID reduction for flow matching on CIFAR-10 at 16 function evaluations. Each model-agnostic kinetic baseline determines its own kinetic reference coordinate. In these coordinates, fiberwise-risk profiles from independently trained models in different settings align closely after normalization to unit area. The resulting schedule deformations used in training also align, suggesting empirical universality across the evaluated models and settings. Pretrained-checkpoint diagnostics extend this normalized-risk agreement to larger conditional latent diffusion and 2-RF models. A frozen analytic allocation template retains most of the model-aware improvement without further risk estimation or model-specific fitting.

cs.LG↗

arXiv2025

Spatio-temporal patterns of active epigenetic turnover

DNA methylation is a primary layer of epigenetic modification that plays a pivotal role in the regulation of development, aging, and cancer. The concurrent activity of opposing enzymes that mediate DNA methylation and demethylation gives rise to a biochemical cycle and active turnover of DNA methylation. While the ensuing biochemical oscillations have been implicated in the regulation of cell differentiation, their functional role and spatio-temporal dynamics are, however, unknown. In this work, we demonstrate that chromatin-mediated coupling between these local biochemical cycles can lead to the emergence of phase-locked domains, regions of locally synchronized turnover activity, whose coarsening is arrested by genomic heterogeneity. We introduce a minimal model based on stochastic oscillators with constrained long-range and non-reciprocal interactions, shaped by the local chromatin organization. Through a combination of analytical theory and stochastic simulations, we predict both the degree of synchronization and the typical size of emergent phase-locked domains. We qualitatively test these predictions using single-cell sequencing data. Our results show that DNA methylation turnover exhibits surprisingly rich spatio-temporal patterns which may be used by cells to control cell differentiation.

physics.bio-ph↗

arXiv2025

Emergent weight morphologies in deep neural networks

Whether deep neural networks can exhibit emergent behaviour is not only relevant for understanding how deep learning works, it is also pivotal for estimating potential security risks of increasingly capable artificial intelligence systems. Here, we show that training deep neural networks gives rise to emergent weight morphologies independent of the training data. Specifically, in analogy to condensed matter physics, we derive a theory that predict that the homogeneous state of deep neural networks is unstable in a way that leads to the emergence of periodic channel structures. We verified these structures by performing numerical experiments on a variety of data sets. Our work demonstrates emergence in the training of deep neural networks, which impacts the achievable performance of deep neural networks.

cs.LG↗

arXiv2025

Universal quasi-particle kinetics control the cell death decision

Understanding how fluctuations propagate across spatial scales is central to our understanding of inanimate matter from turbulence to critical phenomena. In contrast to physical systems, biological systems are organized into a hierarchy of processes on a discrete set of spatial scales: they are compartmentalized. Here, we show that dynamic compartmentalization of stochastic systems leads to emergent, quasi-particle-like kinetics which are used by cells to perform key biological functions. Specifically, we derive a general theory that predicts the emergence of a single degree of freedom irrespective of system specifics. We obtain equations of motion and response characterising its unique kinetic properties. We experimentally demonstrate the biological relevance of quasi-particle kinetics in the decision of cells to commit suicide (apoptosis). Using fluorescent microscopy, we show that the response of cells to apoptotic stimuli exhibits quasi-particle like kinetics which establish a low-pass filter for cellular stress signals. By highlighting that cells manipulate how noise and signals propagate across spatial scales, our work reveals a new mechanism of cell fate decision-making.

physics.bio-ph↗

arXiv2024

Controlling noise with self-organized resetting

Biological systems often consist of a small number of constituents and are therefore inherently noisy. To function effectively, these systems must employ mechanisms to constrain the accumulation of noise. Such mechanisms have been extensively studied and comprise the constraint by external forces, nonlinear interactions, or the resetting of the system to a predefined state. Here, we propose a fourth paradigm for noise constraint: self-organized resetting, where the resetting rate and position emerge from self-organization through time-discrete interactions. We study general properties of self-organized resetting systems using the paradigmatic example of cooperative resetting, where random pairs of Brownian particles are reset to their respective average. We demonstrate that such systems undergo a delocalization phase transition, separating regimes of constrained and unconstrained noise accumulation. Additionally, we show that systems with self-organized resetting can adapt to external forces and optimize search behavior for reaching target values. Self-organized resetting has various applications in nature and technology, which we demonstrate in the context of sexual interactions in fungi and spatial dispersion in shared mobility services. This work opens routes into the application of self-organized resetting across various systems in biology and technology.

cond-mat.stat-mech↗

arXiv2024

Clonal dynamics of surface-driven growing tissues

The self-organization of cells into complex tissues relies on a tight coordination of cell behavior. Identifying the cellular processes driving tissue growth is key to understanding the emergence of tissue forms and devising targeted therapies for aberrant growth, such as in cancer. Inferring the mode of tissue growth, whether it is driven by cells on the surface or cells in the bulk, is possible in cell culture experiments, but difficult in most tissues in living organisms (in vivo). Genetic tracing experiments, where a subset of cells is labeled with inheritable markers have become important experimental tools to study cell fate in vivo. Here, we show that the mode of tissue growth is reflected in the size distribution of the progeny of marked cells. To this end, we derive the clone-size distributions using analytical calculations in the limit of negligible cell migration and cell death, and we test our predictions with an agent-based stochastic sampling technique. We show that for surface-driven growth the clone-size distribution takes a characteristic power-law form with an exponent determined by fluctuations of the tissue surface. Our results show how the mode of tissue growth can be inferred from genetic tracing experiments.

q-bio.QM↗

arXiv2024

Field theory of enzyme-substrate systems with restricted long-range interactions

Enzyme-substrate kinetics form the basis of many biomolecular processes. The interplay between substrate binding and substrate geometry can give rise to long-range interactions between enzyme binding events. Here, we study a general model of enzyme-substrate kinetics with restricted long-range interactions described by an exponent $-λ$. We employ a coherent-state path integral and renormalization group approach to calculate the first moment and two-point correlation function of the enzyme-binding profile. We show that starting from an empty substrate the average occupancy follows a power law with an exponent $1/(1-λ)$ over time. The correlation function decays algebraically with two distinct spatial regimes characterized by exponents $-λ$ on short distances and $-(2/3)(2-λ)$ on long distances. The crossover between both regimes scales inversely with the average substrate occupancy. Our work allows to associate experimental measurements of bound enzyme locations with their binding kinetics and the spatial confirmation of the substrate.

physics.bio-ph↗

arXiv2023

Long-range interactions and disorder facilitate pattern formation in spatial complex systems

Complex systems with global interactions tend to be stable if interactions between components are sufficiently homogeneous. In biological systems, which often have small copy numbers and interactions mediated by diffusing agents, noise and non-locality may affect stability. Here, we derive stability criteria for spatial complex systems with local and non-local interactions from a coarse-grained field theory with multiplicative noise. We show that long-range interactions give rise to a transition between regimes exhibiting giant density fluctuations and pattern formation. This instability is suppressed by non-reciprocity in interactions.

physics.bio-ph↗

arXiv2019

Universality of clone dynamics during tissue development

The emergence of complex organs is driven by the coordinated proliferation, migration and differentiation of precursor cells. The fate behaviour of these cells is reflected in the time evolution their progeny, termed clones, which serve as a key experimental observable. In adult tissues, where cell dynamics is constrained by the condition of homeostasis, clonal tracing studies based on transgenic animal models have advanced our understanding of cell fate behaviour and its dysregulation in disease. But what can be learned from clonal dynamics in development, where the spatial cohesiveness of clones is impaired by tissue deformations during tissue growth? Drawing on the results of clonal tracing studies, we show that, despite the complexity of organ development, clonal dynamics may converge to a critical state characterized by universal scaling behaviour of clone sizes. By mapping clonal dynamics onto a generalization of the classical theory of aerosols, we elucidate the origin and range of scaling behaviours and show how the identification of universal scaling dependences may allow lineage-specific information to be distilled from experiments. Our study shows the emergence of core concepts of statistical physics in an unexpected context, identifying cellular systems as a laboratory to study non-equilibrium statistical physics.

q-bio.TO↗

arXiv2014

Specialization and Bet Hedging in Heterogeneous Populations

Phenotypic heterogeneity is a strategy commonly used by bacteria to rapidly adapt to changing environmental conditions. Here, we study the interplay between phenotypic heterogeneity and genetic diversity in spatially extended populations. By analyzing the spatio-temporal dynamics, we show that the level of mobility and the type of competition qualitatively influence the persistence of phenotypic heterogeneity. While direct competition generally promotes persistence of phenotypic heterogeneity, specialization dominates in models with indirect competition irrespective of the degree of mobility.

q-bio.PE↗

arXiv2014

Range Expansion of Heterogeneous Populations

Risk spreading in bacterial populations is generally regarded as a strategy to maximize survival. Here, we study its role during range expansion of a genetically diverse population where growth and motility are two alternative traits. We find that during the initial expansion phase fast growing cells do have a selective advantage. By contrast, asymptotically, generalists balancing motility and reproduction are evolutionarily most successful. These findings are rationalized by a set of coupled Fisher equations complemented by stochastic simulations.

q-bio.PE↗

arXiv2013

Global attractors and extinction dynamics of cyclically competing species

Transitions to absorbing states are of fundamental importance in non-equilibrium physics as well as ecology. In ecology, absorbing states correspond to the extinction of species. We here study the spatial population dynamics of three cyclically interacting species. The interaction scheme comprises both direct competition between species as in the cyclic Lotka-Volterra model, and separated selection and reproduction processes as in the May-Leonard model. We show that the dynamic processes leading to the transient maintenance of biodiversity are closely linked to attractors of the nonlinear dynamics for the overall species' concentrations. The characteristics of these global attractors change qualitatively at certain threshold values of the mobility, and depend on the relative strength of the different types of competition between species. They give information about the scaling of extinction times with the system size and thereby the stability of biodiversity. We define an effective free energy as the negative logarithm of the probability to find the system in a specific global state before reaching one of the absorbing states. The global attractors then correspond to minima of this effective energy landscape and determine the most probable values for the species' global concentrations. As in equilibrium thermodynamics, qualitative changes in the effective free energy landscape indicate and characterize the underlying non-equilibrium phase transitions. We provide the complete phase diagrams for the population dynamics, and give a comprehensive analysis of the spatio-temporal dynamics and routes to extinction in the respective phases.

q-bio.PE↗

arXiv2013

Stability of localized wave fronts in bistable systems

Localized wave fronts are a fundamental feature of biological systems from cell biology to ecology. Here, we study a broad class of bistable models subject to self-activation, degradation and spatially inhomogeneous activating agents. We determine the conditions under which wave-front localization is possible and analyze the stability thereof with respect to extrinsic perturbations and internal noise. It is found that stability is enhanced upon regulating a positional signal and, surprisingly, also for a low degree of binding cooperativity. We further show a contrasting impact of self-activation to the stability of these two sources of destabilization.

q-bio.CB↗