cognitive trajectory modeling

Existing research on human-AI co-creativity struggles to capture the temporal evolution of higher-order interaction dynamics. This work proposes a Cognitive Trajectory Modeling (CTM) framework that, for the first time, generalizes cognitive trajectories beyond task-specific encodings. Grounded in theories of generative creativity and meaning-making, the framework formalizes trajectory principles imbued with directional cognitive significance. It integrates attractor-based dynamical systems with meaning-making curves to construct a three-layer dynamic architecture—cognition, interaction, and domain—thereby distinguishing cognitive trajectories from mere interaction traces. The resulting paradigm offers an interpretable, theory-driven approach to temporal modeling, establishing a novel foundation for quantitative analysis and evolutionary modeling in human-AI co-creative processes.

A theoretical gap persists between the fine-scale structure of microcircuit neural dynamics and the mechanisms underlying high-level cognitive functions. Method: We propose a brain-inspired computational framework based on state space models (SSMs), leveraging their neuron-like dynamical properties to model flexible learning behaviors—including temporal perception and event counting. Crucially, we employ a diagonalized S5 architecture coupled with reinforcement learning for training and analysis on temporally resolved tasks. Contribution/Results: We discover that rotational dynamics of hidden states in the complex plane unify the emergence of time cells, ramping activity, and traveling waves—phenomena widely observed in neurophysiological experiments. Moreover, this mechanism naturally generalizes to abstract cognitive tasks such as event counting. The model successfully recapitulates diverse neural dynamics reported in biological experiments while demonstrating robust generalization beyond basic timing tasks. These results substantiate SSMs as a unifying computational paradigm bridging neural mechanisms and cognitive function—both theoretically grounded and empirically validated.

This work addresses the insufficient coupling between system-level collective behavior and individual dynamics in multi-agent collaborative time-series modeling. We propose a hierarchical recursive switching state model, featuring a two-layer hidden Markov–recurrent coupled architecture that enables context-aware bottom-up and structure-driven top-down latent-state interactions—marking the first explicit characterization of topological influence of group dynamics on individual trajectories. Learning is performed via variational coordinate ascent, ensuring linear scalability in unsupervised training. Empirically, our model matches the predictive accuracy of large neural networks on basketball and military coordination datasets, while employing orders-of-magnitude fewer parameters and exhibiting linear training cost growth with respect to agent count. Furthermore, it successfully uncovers phased collaborative patterns in a synthetic 64-agent task, demonstrating interpretability and scalability in complex multi-agent dynamics.

Existing neural dynamical modeling approaches rely on single-dataset training and struggle with statistical heterogeneity across recordings. To address this, we propose the Meta Dynamical State-Space Model (Meta-DSSM), the first framework to integrate meta-learning into neural dynamical modeling. Meta-DSSM parameterizes task-specific dynamical families via a low-dimensional manifold and learns a shared dynamical solution space across multi-task neural activities. It unifies variational inference, deep state-space modeling, and meta-learning to enable rapid adaptation, reconstruction, and long-horizon prediction of latent dynamics from few-shot data. Evaluated on synthetic dynamical systems and multi-arm reaching datasets from primate motor cortex, Meta-DSSM significantly improves few-shot reconstruction accuracy and trajectory prediction stability. It establishes a generalizable modeling paradigm for cross-subject and cross-session neural decoding, advancing robustness and transferability in neural dynamical inference.

Existing methods struggle to efficiently model large-scale temporal neural data and capture dynamic inter-regional brain communication patterns. This paper introduces the Markovian Gaussian Process (Markovian GP), a universal state-space representation that transforms any stationary Gaussian process—single- or multi-output—into a linear dynamical system (LDS), bypassing the restrictive separability assumption on kernels for the first time. Our approach combines spectral density decomposition with discretization of stochastic differential equations, integrating state-space modeling and Bayesian inference. Experiments demonstrate that Markovian GP achieves high accuracy in covariance approximation, regression, and neuroscience applications, while scaling linearly in time complexity—O(N)—and yielding significantly lower error than existing GP-to-LDS conversion methods. By unifying theoretical rigor with computational scalability, it establishes a novel, principled framework for large-scale dynamic brain network modeling.

Current large-scale neuroscience data remain highly fragmented, lacking a unified framework to elucidate the closed-loop coupling mechanisms among brain, body, and environment in behavior generation. This work proposes an integrative neuro-cybernetic modeling paradigm that conceptualizes the brain as a controller pursuing latent goals. By synthesizing multi-scale neural recordings, behavioral measurements, perturbations, and connectomic anatomical constraints, it constructs interpretable nonlinear state-space models. The approach incorporates meta-dynamics extensions, hybrid open- and closed-loop training, and knowledge distillation strategies to identify shared dynamical structures across experiments and individuals, disentangle individual variability, infer behavioral objectives, and achieve few-shot generalization. This framework establishes a mechanistic and generalizable computational foundation for uncovering the organizing principles of neural-behavioral systems.

Existing methods struggle to effectively model large-scale, overlapping recurrent interactions in high-dimensional systems, particularly within recursive networks of biological and neural systems. This work proposes a variational framework that, for the first time, represents directed cyclic interactions as edge flows on simplicial complexes embedded in a Hilbert space, enabling projection, averaging, and population-level statistical inference without explicitly enumerating individual loops. By leveraging energy-minimizing dynamical systems, the approach disentangles transient from persistent harmonic flows, constructing a low-dimensional cyclic subspace that captures stable recursive structures. The method substantially outperforms current techniques in densely recurrent systems and successfully recovers large-scale cyclic organization from resting-state fMRI data across 400 human subjects.

This study investigates how predictive artificial intelligence prematurely stabilizes decision trajectories before humans complete autonomous exploration, thereby suppressing cognitive exploration and the development of representational structures. By constructing a geometric dynamical framework, the work characterizes the evolution of attention in policy space as jointly driven by stable drift, endogenous exploratory perturbations, and response-gated learning, modeling predictive assistance as an exogenous mechanism that compresses exploration. The research uncovers three key mechanisms: predictive assistance reduces exploratory responsiveness, asymmetric accumulation of policy-space curvature induces hysteresis in recovery, and early intervention severely constrains subsequent exploration breadth. It further proposes a testable exploration entropy metric and predictions for premature convergence. Simulations demonstrate that sustained prediction attenuates endogenous perturbation effects, delays the restoration of exploratory capacity upon withdrawal, and that intervention timing critically shapes cognitive developmental trajectories.

This study addresses the challenge of decoding latent dynamic structures underlying large-scale neuronal population activity by proposing a unified latent variable modeling framework that, for the first time, jointly integrates three core tasks: single-region dynamics modeling, inter-regional communication analysis, and behavioral alignment. The approach combines classical state-space models with cutting-edge deep generative architectures—including Transformers, diffusion models, and neural ordinary differential equations—to systematically construct a taxonomy and establish clear evaluation benchmarks. Emphasizing critical challenges such as causal inference and directional connectivity, this work provides both theoretical foundations and methodological tools for interpretable brain dynamics analysis and robust neural decoding.

This study addresses the challenge of constructing a universal representational space for brain activity that generalizes across diverse cognitive states and individuals to elucidate mechanisms of cognitive transitions and individual variability. To this end, we propose the Universal Brain Dynamics (UBD) framework, which integrates spatial connectivity and temporal dynamics from fMRI data and leverages the Jacobian matrix derived from dynamical systems theory to quantify neural dynamics. Applying this approach to data from 963 participants in the Human Connectome Project across eight cognitive states, we achieve high-accuracy prediction of fMRI signals (Pearson correlation > 0.9). This work establishes, for the first time, a unified cross-state representation at scale, clarifies the dynamical basis of structure–function coupling, and highlights the critical role of ultra-low-frequency fluctuations in organizing brain activity.