ICore Research Directions: The Contemporary Landscape

1.1 Neural Oscillations & Synchrony: From Correlation to Causality

Research on neural oscillations has undergone a paradigm shift — from observational correlation toward causal mechanism. Barack and Bhatt (2024, Trends in Cognitive Sciences) proposed a framework distinguishing "oscillations as measurement" from "oscillations as process," arguing that oscillations actively organise neuronal activity through temporal, spatial, and frequency "grammar."

Gamma Oscillations (γ, 30–120 Hz)

Bhatt et al. (2024, eLife) characterised the gamma rhythm as a "guardian of brain health" — not merely an epiphenomenon of cognition but an active driver of glymphatic amyloid-β clearance via VIP+ interneuron-mediated neuropeptide release. MIT Picower Institute's GENUS therapy is the landmark achievement: Murdock et al. (2024, Nature) elucidated the VIP-mediated glymphatic clearance mechanism of 40 Hz audiovisual stimulation; Chan et al. (2025, Alzheimer's & Dementia) reported MMSE scores in late-onset Alzheimer's patients significantly superior to matched controls after ~two years of daily stimulation. Wei et al. (2024, PNAS) identified DDL-920, a negative allosteric modulator selectively targeting PV+ interneuron GABA-A receptors that pharmacologically enhances gamma power and improves memory in AD mouse models.

Theta–Gamma Coupling

Ursino & Pirazzini (2024, Current Opinion in Behavioral Sciences) demonstrated that theta–gamma coupling extends across the whole brain, participating in working memory, episodic memory, attention, dreaming, and imagination — exhibiting abnormalities in schizophrenia, epilepsy, and AD. Diedrich et al. (2025, GeroScience) confirmed in a triple-blind RCT of 77 older adults that repeated peak-coupled theta–gamma tACS improves working memory.

Alpha Oscillations (8–13 Hz)

Jensen (2024, Communications Psychology) proposed a revised mechanism: alpha oscillations reflect inhibitory processes indirectly controlled by goal-relevant information load, not direct top-down inhibition. Palva et al. (2025, Journal of Neuroscience) revealed that posterior alpha comprises functionally distinct sources — occipital alpha (10–14 Hz) linked to eye movements, parietal alpha (7–10 Hz) to attentional deployment.

Beta Oscillations (13–30 Hz) & Inter-Brain Synchrony

Mendoza-Halliday et al. (2024, Nature Neuroscience) discovered a universal "spectral laminar gradient": high-frequency power (γ/β) distributed in a gradient from deep to superficial cortical layers — a fundamental organisational principle of cortical oscillations. Multiple 2025 hyperscanning studies (EEG + fNIRS) demonstrated that empathy and shared identity drive cross-brain phase synchrony during social interactions.

1.2 Neural Network Dynamics Modelling: From Classical to Data-Driven

Modern Extensions of Hodgkin–Huxley

A 2025 eLife study introduced a mean-field model retaining full ion exchange mechanisms (Na⁺, K⁺, Cl⁻, ion pumps, astrocytic buffering) capable of reproducing activity patterns from fast spiking to slowly modulated ionic concentration changes, validated in vitro. Di Volo et al. (2024, Journal of Neurophysiology) developed a mean-field model predicting collective dynamics from adaptive exponential integrate-and-fire neurons through to full Hodgkin–Huxley models.

Attractor Networks

Khona & Fiete (2022, Nature Reviews Neuroscience, MIT) provided a landmark review affirming continuous attractor dynamics in head-direction cells, grid cells, and place cells. Beer & Barak (2024, PLOS Computational Biology, Technion) discovered a counterintuitive finding: electrical stimulation of specific attractors in cultured cortical networks paradoxically eliminates them from spontaneous activity — explained by Hebbian strengthening and homeostatic weakening. Spisak et al. (2025, eLife) proposed connectome-based Hopfield Neural Networks (fcHNNs) linking attractor states to global Bayesian priors within Friston's free-energy framework.

Spiking Neural Networks (SNNs)

Surrogate gradient methods enabled SNN training at deep-network scale. Yao et al. (2023) achieved a 87.2× computational energy efficiency improvement via spike-driven self-attention. Stanojevic et al. (2024, Nature Communications) demonstrated high-performance deep SNNs requiring only 0.3 spikes per neuron. A 2025 Nature Communications study showed SNNs achieve twice the adversarial robustness of ANNs on CIFAR-10.

Dynamical System Reconstruction (DSR)

Durstewitz, Koppe & Thurm (2023, Nature Reviews Neuroscience) published a landmark paper proposing RNNs trained directly on neural/behavioural data as formal surrogate models of biological systems — preserving attractor structure, vector field topology, and temporal geometric properties. Hierarchical DSR with cross-domain transfer learning was presented at ICLR 2025.

Wilson–Cowan & Neural Mass Models

Marino et al. (2024, Neural Computation) transformed the Wilson–Cowan metapopulation model into a learning algorithm achieving high classification accuracy on MNIST and CIFAR-10. Byrne et al. developed the next-generation Wilson–Cowan model based on exact mean-field reduction of quadratic integrate-and-fire neurons, with population firing rate derived from the Kuramoto order parameter.

1.3 Brain Rhythms & Higher Cognitive Function

A 2025 Nature Communications MEG+computational modelling study identified four distinct functional brain states in theta and alpha bands: the encoding state dominated by posterior theta, the maintenance state by dorsal alpha — with optimal state-transition rates correlated with better cognitive performance.

Research on the neural dynamics of consciousness has ignited intense empirical debate between Integrated Information Theory (IIT) and Global Workspace Theory (GWT). The COGITATE Consortium (2023–2025) confirmed posterior gamma decoding but failed to find the predicted gamma synchronisation. A new theoretically neutral framework (2025, arXiv) decomposed consciousness-related dynamics into three attributes: Hierarchical integration (H), cross-frequency complexity (D), and metastability (M), validated across nine brain states.

1.4 Pathological Neurodynamics: From Models to Clinic

Epilepsy

The EPINOV clinical trial — led by Viktor Jirsa (scientific) and Fabrice Bartolomei (coordination) — is the world's first clinical trial using whole-brain network modelling. Enrolling 356 patients across 11 French epilepsy centres (2019–2024), it used the TVB platform to simulate seizure propagation, published as cover articles in Science Translational Medicine (2023) and The Lancet Neurology (2023). A 2025 Nature Neuroscience study demonstrated that spatiotemporally targeted closed-loop electrical stimulation eliminates aberrant cortical activity, prevents spreading, and improves long-term spatial memory.

Parkinson's Disease

Oehrn et al. (UCSF, 2024, Nature Medicine) proved in a blinded RCT that chronic adaptive DBS using subthalamic beta power as feedback is superior to conventional continuous DBS. In February 2025, Medtronic's BrainSense™ adaptive DBS system received FDA approval — the world's first approved closed-loop DBS system, using beta oscillation biomarkers for real-time parameter adjustment. Mathiopoulou et al. (2024, Charité Berlin) revealed that medication and DBS have additive but distinct effects: medication primarily suppresses low beta (13–20 Hz), DBS acts across a broader beta range.

Alzheimer's Disease

The GENUS 40 Hz gamma stimulation approach is rapidly advancing from animal experiments to clinical phases. MIT spinout Cognito Therapeutics is conducting national clinical trials. Park & Tsai (2025, PLOS Biology) documented expanding evidence beyond AD to Parkinson's disease, stroke, anxiety, epilepsy, chemotherapy cognitive side effects, and multiple sclerosis.

1.5 Cross-Scale Integration: From Single Neurons to Whole Brain

Complete Adult Drosophila Brain Connectome (October 2024)

The FlyWire Consortium (200+ researchers, 50+ labs), led by Mala Murthy and Sebastian Seung (Princeton), simultaneously published 9 papers in Nature, mapping 139,255 neurons, 50 million+ synaptic connections, and 8,453 cell types. Network analysis revealed a rich-club organisational structure. Shiu et al. (UC Berkeley / Eon Systems) simulated the entire connectome on a laptop; in 2025, Eon Systems demonstrated the world's first whole-brain simulation embodiment generating multiple behaviours — a virtual fly brain controlling a simulated body.

Mouse Whole-Cortex Microscale Simulation

Anton Arkhipov (Allen Institute) and Yamazaki Tadashi (University of Electro-Communications, Japan) reported at SC25 a biophysically detailed simulation of ~10 million neurons, 26 billion synapses, across 86 brain regions on the Fugaku supercomputer — the most biologically realistic animal brain simulation to date.

The Virtual Brain (TVB) Platform & EBRAINS

Led by Jirsa (Aix-Marseille/CNRS/Inserm), TVB has produced 100+ peer-reviewed papers, simulated ~1,000 individualised connectome-based brain models, and consumed 10 million+ CPU core-hours. TVB is the primary whole-brain simulator on EBRAINS (1,029+ datasets, 225 research software tools, 10,000+ users). The HBP 10-year assessment (2025) confirmed the Structured Flows on Manifolds (SFM) concept and the Epileptor canonical model — described by Karl Friston as "a major breakthrough in the field."

IIMajor International Institutions & Leading Scientists

IIISix Major Breakthroughs 2021–2025

IVTechnical Ecosystem of Research Methods

4.1 Theoretical Modelling Foundations

The theoretical foundations are rooted in dynamical systems theory (attractors, bifurcations, limit cycles, chaos), neural field theory (Wilson–Cowan, Amari, Nunez formalisms), and Friston's free-energy principle / active inference — treating brain dynamics as approximate Bayesian inference minimising variational free energy. Bifurcation analysis illuminates state transitions (seizure onset, sleep–wake); low-dimensional manifold analysis and information-theoretic methods (transfer entropy, mutual information) complete the toolkit.

4.2 Computational Simulation Platforms

4.3 Experimental Data Analysis

In EEG/MEG analysis, Friston's Dynamic Causal Modelling (DCM) uses neural mass/field models to infer effective connectivity. Time–frequency analysis (wavelet transforms, multitaper spectral estimation), phase–amplitude coupling, microstate analysis, and dynamic functional connectivity methods (sliding-window correlation, wavelet coherence, phase-locking values) form the complete analytical pipeline. In fMRI, Deco et al.'s LEiDA (Leading Eigenvector Dynamics Analysis) and the deep-learning-based TENET framework are representative methods.

4.4 Integration with AI / Machine Learning

Neural Ordinary Differential Equations (Chen et al., NeurIPS 2018) established continuous-time dynamical system modelling, with descendants including ControlSynth Neural ODEs (NeurIPS 2024), physics-informed Neural ODEs, and equilibrium Neural ODEs (ICLR 2025). SINDy (Sparse Identification of Nonlinear Dynamics) enables data-driven discovery of governing equations. COSYNE 2025 dedicated a workshop to "Building Foundation Models for the Brain," reflecting accelerating NeuroAI convergence. Dimensionality reduction and neural geometry analysis (manifold learning, topological data analysis) are increasingly central to deciphering neural population dynamics.

VFrontier Hotspots & Future Directions

VIPremier Journals & Conferences

Core Journals

Key Conferences