Psychometric Evaluation of a Thai Version of the Cardiac Distress Inventory.

The brain performs functions such as perception, decision-making, memory, planning, and navigation thanks to the coordinated activity of groups of neurons. This coordinated activity of neural groups presents a low-dimensional continuous structure that can carry complex perceptual, motor, and cognitive task information in the form of trajectories, which is called a neural manifold. For different cognitive tasks, neural activity trajectories are typically confined to specific low-dimensional subspaces, along which the system evolves to maintain stable information processing and predictive functions—allowing organisms to perceive and make decisions efficiently in complex environments. This type of trajectory not only reflects the dynamic evolution of neural states over time, but also inherently encodes the ability to predict environmental conditions and future events, forming the basis for the brain’s information encoding, decision-making, and anticipatory perception. Therefore, constructing a brain-inspired neural manifold computing architecture is a key pathway to achieving efficient and intelligent perception. In recent years, researchers have been striving to emulate neural functions through hardware implementations. As a promising solution, artificial neuron devices—such as Mott-transition memristors—have attracted widespread attention due to their structural simplicity, fast spiking response, and low power consumption. These devices also offer the potential to preserve complex neural dynamics and simplify circuit architectures. However, most existing studies focus on achieving static input-output mappings for tuning curves, reproducing spiking-like behaviors, and scaling up arrays for computational capacity. Constructing spike-based neural manifolds from a device perspective to replicate brain-like perception remains largely unexplored. Although various methods, such as long short-term memory (LSTM) networks and reservoir computing (RC), have been deployed in neuromorphic hardware for prediction, they typically require large training datasets and exhibit limited generalization to unseen scenarios. Therefore, developing a hardware manifold architecture capable of achieving efficient, accurate, and robust prediction with minimal samples is desirable. Building upon this perspective, a fundamental question concerns how the brain leverages neural manifolds to infer and predict future trajectories from partial and noisy sensory inputs. This further raises the question of whether future trajectories can be directly predicted solely from experimental data. The spatiotemporal information (STI) equation, is derived from randomly distribution embedding based on Takens’ theorem, and can convert high-dimensional data in experiments into future trajectories, effectively expanding the sample size and alleviating the small sample size problem. Recently, driven by the ever-increasing training resources and the importance of the short-term characteristics of dynamic systems, STI has gained increasing attention for short-term prediction tasks, including applications in typhoon, traffic flow, earthquake, and wireless data. However, existing studies primarily focus on emergency application predictions. It remains unclear how STI can be implemented in a physically grounded, energy-efficient neuromorphic hardware with manifold structures to emulate the brain’s short-term predictive capabilities.