The research, led by Yuhang Wu, Shi Gu, and colleagues at Zhejiang University, the University of Electronic Science and Technology of China, New York University, and the University of Pennsylvania, including network neuroscientist Dani S. Bassett, took a deliberately controlled approach. Rather than embedding their networks in physical space and imposing wiring-cost constraints, the team trained recurrent neural networks (RNNs) on batteries of cognitive tasks of the kind long used in systems neuroscience: working memory, perceptual decision-making, context-dependent categorization, and other tasks that probe the computational repertoire of prefrontal and parietal cortex. The question was simple but profound: if you strip away all spatial and metabolic pressure, does modularity still appear when a network must learn many demanding tasks at once?

The answer was a resounding yes. Networks trained under multitask learning paradigms developed significantly higher modularity than networks trained on a single task, and the effect grew stronger as the task load pushed against the network’s capacity. When the number of simultaneous tasks strained what the fixed pool of units could compute, the networks responded by reorganizing their internal connectivity into functionally segregated communities. This is a striking result because nothing in the training objective rewarded modularity directly. The networks were optimized only for task performance, yet the pressure of limited capacity and diverse demands was sufficient to carve the connectivity matrix into modules—much as the pressure of diverse cognitive demands may have shaped the brain’s own architecture.

The study went further by comparing different training regimes. Networks trained with incremental multitask learning—in which tasks were introduced sequentially and the network had to integrate each new demand into an already functioning system—developed the highest degree of modularity of all, while also maintaining superior performance across the full task set. This detail matters because it mirrors the developmental trajectory of biological brains, which do not acquire all cognitive abilities simultaneously but build them progressively over years of experience. The finding suggests that the order and pacing of task acquisition, not merely the total computational load, shapes the topology that emerges. Modularity, in this view, is not a static design feature but an adaptive response to the sequential introduction of complex problems.

Technically, the team quantified modularity using established network-science measures, including community-detection methods of the kind pioneered by Leicht and Newman for directed networks, applied to the learned weight matrices of the RNNs. They tracked how modular structure unfolded over the course of training, revealing that community boundaries sharpened as learning progressed and as additional tasks accumulated. They also examined the incremental addition of connections during learning, drawing an intriguing parallel to the lottery ticket hypothesis from deep learning research—the idea that sparse, trainable subnetworks exist within larger networks and are the components that effectively carry the computational load. In the task-trained RNNs, sparse modular substructures appeared to play an analogous role, suggesting a possible computational rationale for why both artificial and biological learning systems might favor segregated, sparsely interconnected architectures.

Perhaps the most consequential finding came when the researchers compared their task-induced networks against biological data. Using structural connectivity data from the Human Connectome Project, covering 84 cortical areas, they evaluated how closely the artificial networks matched the brain’s own organization. The task-trained networks exhibited structural properties that more closely resembled biological brain networks than models based solely on spatial constraints such as wiring-cost minimization. In other words, functional demand—the need to compute—appears to be a stronger organizing principle for brain-like topology than physical economy alone. This does not mean spatial constraints are irrelevant; the brain is certainly shaped by the geometry of the skull and the metabolic cost of axons. But the new results demonstrate that spatial models alone cannot fully explain the functional organization of brain networks, and that computational pressure fills a substantial part of that explanatory gap.

The work builds on a rich lineage of research at the intersection of machine learning and neuroscience. Previous studies had shown that RNNs trained on many cognitive tasks develop mixed selectivity and shared dynamical motifs that support flexible behavior, and that spatially embedded RNNs recapitulate numerous structural and functional findings from neuroscience. Other work demonstrated that brain-like functional specialization can emerge spontaneously in deep networks trained on naturalistic tasks. The new study adds a crucial piece: a controlled computational demonstration that modularity itself—arguably the signature feature of brain network organization—can be induced purely by the functional demands of multitask learning under capacity constraints. It thereby offers a causal, mechanistic account where earlier work offered correlations or spatial explanations.

The implications run in both directions. For neuroscience, the study provides a testable framework: if modularity is an adaptive response to cumulative cognitive demands, then developmental changes in brain network segregation should track the acquisition of complex abilities, a hypothesis consistent with prior findings that modular segregation of structural brain networks supports the development of executive function in youth. For artificial intelligence, the results hint at design principles for more adaptable machines. Modular deep learning has become a vibrant field precisely because modular systems can learn new skills without catastrophically forgetting old ones, and this study suggests that simply structuring the training curriculum—introducing tasks incrementally under realistic capacity limits—can coax modularity into existence without hand-engineered architectural constraints. That could inform how researchers build continual-learning systems, from robotics to large multimodal models, where flexibility and stability must coexist.

The study also speaks to a long-standing debate about the economy of brain network organization. The brain has often been described as a compromise between wiring cost and topological value, with small-world architecture emerging from that trade-off. The new findings suggest the ledger has more entries than previously appreciated: computational value, capacity limits, and the temporal sequence of learning demands all leave structural fingerprints. Modularity may be less a consequence of saving wire and more a consequence of solving problems—a functional adaptation that spatial economy then refines rather than creates. As the authors put it in their abstract, modularization emerges as an adaptive response to the sequential introduction of complex tasks, a framing that reframes the brain’s architecture as the product of a computational curriculum written by evolution and experience.

The team has released its code and processed connectivity data publicly via GitHub and Zenodo, allowing other researchers to reproduce the simulations and extend the approach to new task sets, architectures, and species comparisons. Cross-species network comparisons, representational similarity analyses, and generative models of the connectome are natural next steps, and the framework could eventually be applied to clinical questions, since many neuropsychiatric conditions involve disruptions of modular brain organization. For now, the study stands as a vivid example of a growing trend in modern science: artificial neural networks are no longer just engineering tools but instruments for asking why questions about the brain—and, increasingly, they are answering with structures that look strikingly familiar. When a machine learns the way a mind must, it begins, it seems, to build itself the way a brain is built.

Subject of Research: Emergence of task-driven modular network organization in recurrent neural networks and its alignment with human brain architecture

Article Title: Task-structured modularity emerges in artificial networks and aligns with brain architecture

Article References: Wu, Y., Deng, S., Du, K., Mattar, M. G., Wu, Y., Bassett, D. S., Tang, H., Pan, G., & Gu, S. (2026). Task-structured modularity emerges in artificial networks and aligns with brain architecture. Nature Machine Intelligence. https://doi.org/10.1038/s42256-026-01306-9

Image Credits: AI Generated

DOI: 10.1038/s42256-026-01306-9

Keywords: recurrent neural networks, multitask learning, modularity, brain networks, connectome, cognitive tasks, network neuroscience, Human Connectome Project, continual learning, wiring cost, lottery ticket hypothesis, Nature Machine Intelligence

To explore the question, Hellwig constructed model networks of pyramidal neurons arranged in monolayers of 101 by 101 cells, a total of 10,201 neurons each. Connectivity in these networks followed distance-dependent, Gaussian connectivity profiles, meaning that the likelihood of a synapse between two neurons peaked when they were adjacent and decayed smoothly as their separation grew. Crucially, the model parameters were anchored in experimental data rather than chosen arbitrarily. In one anatomical scenario, calibrated from histological measurements of connectivity in the cortex, the probability that two neighboring neurons were connected was set at 0.8. In a second, electrophysiological scenario, based on recordings of synaptic connections between nearby pyramidal cells, connection probabilities for adjacent neurons ranged from 0.08 to 0.23, reflecting the much sparser functional connectivity often observed in slice experiments.

The key methodological move of the study was to compare these distance-dependent networks against a rigorous benchmark: the configuration model. This well-established construct from network science takes a given network and randomly rewires its connections while preserving each neuron’s degree, the number of connections it maintains. Because the configuration model retains the degree distribution but destroys any spatial structure, differences between the two types of network can be attributed specifically to the distance-dependent wiring rule rather than to how many connections neurons happen to have. This makes the comparison a powerful test of what geometry alone contributes to cortical circuit architecture.

The analysis drew on a standard toolkit of network science. Hellwig computed average degrees and degree distributions to characterize how connectivity was spread across the population, local clustering coefficients to measure how densely interconnected a neuron’s neighbors were, and graph distances to quantify how many synaptic steps separated any two neurons. He also examined cliques, groups of neurons in which every member connects to every other, recording their numbers, sizes and spatial dimensions. Finally, he estimated the cost of connectivity, a measure of the total wiring length the network requires, which is widely assumed to be a major evolutionary constraint on brain architecture because axons and dendrites occupy physical space and consume metabolic resources.

Across every structural measure, the distance-dependent networks differed profoundly from their configuration-model counterparts. The most dramatic difference appeared in local clustering. In the distance-dependent networks, a neuron’s neighbors were far more likely to be connected to one another, producing tightly interwoven neighborhoods of cells. The configuration-model networks, despite having identical degree distributions, showed much lower clustering because their rewired connections linked neurons at random across the layer, scattering each neuron’s partners across the sheet rather than concentrating them nearby.

The clique analysis reinforced this picture. Distance-dependent networks contained more numerous groups of strongly connected neurons, and these groups were spatially compact, occupying small, well-defined patches of the cortical sheet rather than being dispersed. In other words, the simple rule of connecting preferentially to nearby cells automatically produced clusters of mutually interconnected neurons. Such clusters are structurally reminiscent of the neuronal assemblies that many theories of cortical function posit as the basic units of information processing, from Hebb’s classic cell assemblies to modern models of attractor dynamics and working memory. The finding suggests that the raw material for such assemblies may emerge as an inevitable consequence of distance-dependent wiring, without requiring any additional developmental mechanism to actively group neurons together.

Wiring cost provided another important contrast. Because distance-dependent connections are short, the networks achieved their rich local structure at substantially lower wiring cost than the randomly rewired configuration-model networks, which must stretch axons across the layer to maintain the same degree distribution. This combination of high clustering and low cost is exactly the profile that efficient biological networks would be expected to show, and it helps explain why distance-dependent connectivity is such an attractive design principle for the cortex, where metabolic economy and local processing both matter.

Perhaps the most consequential finding was the sensitivity of network structure to near-neighbor connectivity. When the probability that adjacent neurons connect was high, as in the anatomical scenario with its 0.8 connection probability between neighbors, the network reliably developed tightly wired, spatially localized neuronal clusters. When near-neighbor connectivity was lower, as in the electrophysiological range of 0.08 to 0.23, the resulting structure was correspondingly less clustered. This means that the unresolved discrepancy among experimental estimates of local connection probability is not a trivial measurement problem: it translates directly into different predictions about the architecture of cortical circuits. Networks wired with high local connectivity are poised to support strong local recurrent interactions, while sparser wiring produces a more distributed pattern of connections whose functional consequences may be quite different.

The study’s conclusions point in two directions at once. Scientifically, they indicate that distance-dependent connectivity gives rise to structural features that may facilitate the emergence of functional neuronal assemblies, providing a plausible bridge between the geometry of cortical wiring and the assembly-based theories of cortical computation. Practically, Hellwig proposes a general probabilistic rule for local cortical connectivity derived from the findings, intended as a recipe for designing artificial neural networks with biologically inspired wiring principles. As neuromorphic hardware and brain-inspired machine learning architectures mature, rules that reproduce the clustering, compactness and cost efficiency of real cortical circuits could offer a principled alternative to purely random or fully connected designs.

The work also carries a caution for the field. Because network structure is so sensitive to near-neighbor connectivity, the inconsistent experimental estimates of local connection probability in the literature carry real weight for how we model the cortex. Reconciling anatomical and electrophysiological measurements, and understanding why they diverge, may be essential for building cortical network models that are structurally faithful. By showing exactly which structural features depend on which wiring parameters, and by benchmarking distance-dependent networks against degree-preserving random rewiring, the study provides a framework for testing future models against both the data and the theory. The message is elegant in its simplicity: a single rule, connect more likely to those nearby, sculpts local cortical networks into clustered, compact, low-cost architectures that look strikingly like the substrate needed for assemblies of neurons to act together.

Subject of Research: Structural organization of local cortical neuronal networks generated by distance-dependent connectivity rules

Article Title: Structural characteristics of local cortical networks wired by distance dependent connectivity rules

Article References: Hellwig, B. (2026). Structural characteristics of local cortical networks wired by distance dependent connectivity rules. BMC Neuroscience, 27(1), Article 36. https://doi.org/10.1186/s12868-026-01050-1

Image Credits: AI Generated

DOI: 10.1186/s12868-026-01050-1

Keywords: cerebral cortex, pyramidal neuron, local connectivity, connection probability, distance-dependent connectivity, neuronal assembly, network science, clustering coefficient, configuration model, wiring cost, neural circuits, network topology