An Adaptive Marked Point Process Filtering Approach for Decoding Cognitive Performance

The cognitive performance state is a hidden brain state that can be decoded from behavioral data. One of the challenges in the performance decoder design realm is specifying an informative model and decoder that can realistically reflect the performance dynamics. We employ an autoregressive conditional heteroskedasticity (ARCH) models to capture performance volatility over time and represent a performance state that evolves through a nonlinear and time-varying model. Leveraging a marked point process (MPP) framework, we decode the performance state from a sequence of correct/incorrect responses and reaction times via a Bayesian state-space approach within particle filtering. The MPP framework enables us to account for the reaction times associated with correct responses. We use both experimental and simulated data to evaluate the proposed performance decoder. Findings from experimental data reveal a reliable state estimation outcome that may preserve the environmental impact. Our simulation study depicts the reliability of the decoder. The designed decoder provides an adaptive framework to preserve the information within the process noise and decode the hidden performance state from behavioral data. The proposed performance decoder has potential applications in human-machine interaction, smart workplaces, and future educational settings.