Researchers have built an AI-powered model that traces how the brain’s electrical rhythms and physical wiring change together across nearly the entire human lifespan, from age 5 to age 100. The model, called the xi-alpha Net, inverts standard EEG recordings to estimate the anatomical connections and signal-transmission delays that produce the rhythms clinicians measure on the scalp. Drawing on resting-state EEG data from approximately 1,965 participants, the work offers one of the broadest windows yet into how oscillatory brain activity and structural connectivity co-evolve from early childhood through old age.
How the Xi-Alpha Net Works
The core innovation is a Bayesian-inverted generative model that takes cross-spectral EEG data and works backward to infer two hidden variables: the strength of anatomical connections between brain regions and the speed at which electrical signals travel along the axons linking them. The model, described in a recent preprint, focuses on alpha-band oscillations (roughly 8 to 12 Hz) and the aperiodic, or 1/f-like, component of the EEG spectrum. Alpha waves are the dominant rhythm during relaxed wakefulness, while the aperiodic slope reflects the overall balance of excitatory and inhibitory neural activity. By jointly modeling both components, the xi-alpha Net can separate age-driven changes in wiring from age-driven changes in transmission speed, two factors that standard EEG analysis typically conflates.
The practical advantage is directness. Traditional approaches require expensive MRI scans to image white-matter tracts or rely on indirect statistical correlations between EEG channels. The xi-alpha Net instead uses a physics-informed generative architecture, embedding known constraints about how electromagnetic signals propagate through brain tissue, so that the inversion from scalp recordings to underlying connectivity is biologically grounded rather than purely data-driven. Because it operates directly on cross-spectral matrices, the model can exploit both amplitude and phase information, extracting timing relationships that carry clues about how long it takes signals to move between distant cortical areas.
A Dataset Spanning Nearly a Century of Life
The study draws on the HarMNqEEG resource, a multisite collection of approximately 1,966 participants whose EEG recordings were harmonized across different laboratories and equipment setups. The dataset includes cross-spectral matrices with defined frequency sampling, meaning each participant’s data captures not just power at individual electrodes but also the phase relationships between electrode pairs across the frequency range. That level of detail is what allows the xi-alpha Net to estimate interareal delays, because the timing offset between two brain regions’ oscillations carries information about how long signals take to travel between them.
Spanning ages 5 to 100, the cohort captures developmental periods that differ radically in brain structure. Myelination, the insulation of axons that speeds electrical conduction, is still incomplete in a five-year-old’s prefrontal cortex but begins to degrade in adults past their sixties. By fitting the model across this full range, the researchers can track how conduction delays lengthen or shorten at different life stages and how those shifts map onto the alpha and aperiodic features visible in the raw EEG. The harmonization procedures used in HarMNqEEG are designed to reduce site-specific artifacts, making age-related trends more likely to reflect biology rather than differences in hardware or recording protocols.
Aperiodic Slope as a Developmental Marker
One of the study’s central findings involves the aperiodic component of the EEG, sometimes called neural noise. Separate peer-reviewed work in human development research has established that the aperiodic slope changes systematically from childhood through middle adulthood, with data covering participants aged approximately 6 to 54. A steeper slope generally reflects stronger inhibitory tone, while a flatter slope is associated with a noisier, more excitation-dominated brain state. Children tend to have flatter slopes, which steepen as cortical circuits mature.
The xi-alpha Net adds a mechanistic layer to that observation. Rather than simply documenting that the slope changes with age, it attributes a portion of the change to specific shifts in anatomical connectivity and axonal conduction delays. This distinction matters because it separates local cellular properties, such as the density of inhibitory interneurons, from long-range network properties, such as how quickly the prefrontal cortex can communicate with the parietal lobe. If conduction delays account for a meaningful share of the aperiodic slope’s age trajectory, interventions targeting white-matter health could in principle alter that trajectory, a hypothesis that remains untested but is now quantitatively framed.
The developmental pattern also dovetails with broader cognitive trends. As inhibitory control and working memory improve through childhood and adolescence, the steepening aperiodic slope may reflect a tightening of cortical regulation. The xi-alpha Net suggests that some of this tightening is mediated by faster, more reliable long-range communication, not just by local synaptic refinement. Conversely, a later-life flattening of the slope could signal both cellular-level changes and a slowing of communication along aging white-matter tracts.
Wiring Changes Across the Full Lifespan
Parallel work in structural brain imaging reinforces the picture. A study in lifespan connectomics used high-quality fMRI data from the NIH Lifespan Human Connectome Project, combining data from the Baby Connectome, HCP-Development, HCP-Young Adult, and HCP-Aging cohorts to map effective connectivity from birth to age 100. That analysis documents how directed influence between brain regions reorganizes across the lifespan, with rapid gains in long-range connectivity during childhood, relative stability in young adulthood, and selective weakening of certain pathways in later decades.
When read alongside the xi-alpha Net results, a consistent story emerges: the electrical rhythms that EEG captures are not just passive reflections of brain state but are actively shaped by the physical architecture of the connectome. As white-matter tracts mature, conduction delays shorten, alpha oscillations sharpen, and the aperiodic slope steepens. As those tracts degrade with aging, the process partially reverses. The two lines of evidence, one electrophysiological and one hemodynamic, converge on the same structural bottleneck: the speed and integrity of long-range axonal connections.
The imaging-based work also highlights regional nuances that the xi-alpha Net can, in principle, probe. For example, effective connectivity involving association cortices appears particularly dynamic across the lifespan, while primary sensory pathways are more stable. If the xi-alpha Net detects disproportionate changes in conduction delays for frontal-parietal circuits, that would align with fMRI findings that these systems are especially vulnerable to both developmental disruption and age-related decline.
Independent Confirmation From MEG Studies
A separate line of research using magnetoencephalography (MEG) and structural MRI provides additional triangulation. Work indexed in a recent MEG report examined frequency-resolved cortical functional connectivity across the adult lifespan, relating oscillatory coupling patterns to measures of white-matter integrity. That study found that higher conduction efficiency in major tracts was associated with stronger and more coherent coupling in the alpha and beta bands, particularly between frontal and parietal regions.
Although MEG and EEG differ in their sensitivity to deep versus superficial sources, both modalities point to a tight coupling between oscillatory coordination and structural pathways. The xi-alpha Net extends this logic by explicitly modeling how conduction delays shape the phase relationships that underlie functional connectivity estimates. In doing so, it bridges the gap between biophysical models of signal transmission and the empirical connectivity patterns seen in MEG and fMRI.
These converging findings also underscore the value of large, shared databases. Platforms such as the NCBI infrastructure and related repositories have made it easier to integrate electrophysiological, imaging, and genetic data across cohorts and modalities. The HarMNqEEG dataset itself was assembled with harmonization in mind, and its broad age coverage makes it a natural testbed for models that aim to link microstructural changes to macroscopic rhythms.
Implications and Open Questions
The xi-alpha Net is still at the preprint stage, and its claims will need to be vetted and replicated. Nonetheless, the approach illustrates how generative modeling can turn routine clinical measurements into richer probes of brain structure. If validated, the method could eventually offer a low-cost way to infer aspects of white-matter health from standard EEG recordings, potentially complementing MRI in settings where advanced imaging is unavailable.
Several open questions remain. One is how specific the inferred conduction delays are to particular tracts or networks, given the coarse spatial resolution of scalp EEG. Another is how well the model generalizes beyond resting-state conditions to tasks that strongly modulate alpha power and aperiodic structure. Finally, it will be important to test whether the inferred parameters track individual differences in cognition, mental health, or neurodegenerative risk, rather than only group-level age trends.
Still, by aligning EEG rhythms with structural constraints across nearly a century of life, the xi-alpha Net and related work offer a more integrated view of brain aging and development. Instead of treating oscillations, connectivity, and anatomy as separate domains, these studies suggest they are different facets of the same evolving system, a system whose timing, wiring, and noise levels change together as the brain grows, adapts, and eventually declines.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
