FMTP: A unifying computational framework of temporal preparation across time scales.

Abstract

Temporal preparation is the cognitive function that takes place when anticipating future events. This is commonly considered to involve a process that maximizes preparation at time points that yield a high hazard. However, despite their prominence in the literature, hazard-based theories fail to explain the full range of empirical preparation phenomena. Here, we present the formalized multiple trace theory of temporal preparation (fMTP), an integrative model which develops the alternative perspective that temporal preparation results from associative learning. fMTP builds on established computational principles from the domains of interval timing, motor planning, and associative memory. In fMTP, temporal preparation results from associative learning between a representation of time on the one hand and inhibitory and activating motor units on the other hand. Simulations demonstrate that fMTP can explain phenomena across a range of time scales, from sequential effects operating on a time scale of seconds to long-term memory effects occurring over weeks. We contrast fMTP with models that rely on the hazard function and show that fMTP’s learning mechanisms are essential to capture the full range of empirical effects. In a critical experiment using a Gaussian distribution of foreperiods, we show the data to be consistent with fMTP’s predictions and to deviate from the hazard function. Additionally, we demonstrate how changing fMTP’s parameters can account for participant-to-participant variations in preparation. In sum, with fMTP we put forward a unifying computational framework that explains a family of phenomena in temporal preparation that cannot be jointly explained by conventional theoretical frameworks.

Simple Human Response Times are Governed by Dual Anticipatory Processes with Distinct and Distributed Neural Signatures

“…Additionally, we identified clusters with activity patterns that appeared to be relevant for anticipatory information processing, including several clusters with activity modulations related to the short- versus long-delay conditions (e.g., clusters 2, 3, 8, 11–13, and 19). Some of these modulations affect ramp-like signals that are roughly consistent with certain theoretical models of how a neural population in aggregate could represent represent elapsed time from a sensory event (Salet et al, 2022; Shankar & Howard, 2012). Thus, these clustering analyses showed a much more widespread anatomical distribution than just premotor areas that have been the targets of many previous studies and a much greater diversity of activity patterns than the kinds of joint modulations by delay condition and stochastic RT variability expected from simple versions of the rise-to-bound model (Fig.…”

“…We included two rise-to-bound processes to capture predictive and reactive components of RT variability (including premature false alarms) within the same model and help bridge prior findings. The pre-stimulus process, which accounted for relatively high rates of false alarms in some participants, is a simplified mechanism to capture certain aspects of predictive information processing that allow individuals to generate responses based on expectations formed based on prior experience (Los et al, 2014; Los & Van Den Heuvel, 2001; Salet et al, 2022), possibly via associative learning mechanisms (Rescorla and Wagner, 1972; Montague et al, 1996; Friston, 2010). We also wanted to account for the shape of RT distributions, which are often not accounted for by such models (Salet et al, 2022).…”

“…The pre-stimulus process, which accounted for relatively high rates of false alarms in some participants, is a simplified mechanism to capture certain aspects of predictive information processing that allow individuals to generate responses based on expectations formed based on prior experience (Los et al, 2014; Los & Van Den Heuvel, 2001; Salet et al, 2022), possibly via associative learning mechanisms (Rescorla and Wagner, 1972; Montague et al, 1996; Friston, 2010). We also wanted to account for the shape of RT distributions, which are often not accounted for by such models (Salet et al, 2022). The post-stimulus process, which resulted in faster RTs without a corresponding increase in false alarms, is akin to other reactive models that that emphasize mechanisms that allow rapid fluctuations in temporal expectation (within trial) to alter behavior, for example, via selective attention that prioritizes incoming sensory information) (Summerfield & Egner, 2009).…”

“…Predictive network dynamics were influenced strongly by neural populations with attenuated responses during settings of relatively increased temporal expectation, consistent with a representation related to prediction errors (Montague et al, 1996; Nobre & Van Ede, 2018; Rescorla & Wagner, 1972). These signals tended to have relatively high prevalence in the medial temporal lobe, where memory-related functions and representations of time (Eichenbaum, 2014; Howard et al, 2015; MacDonald et al, 2011; Shankar & Howard, 2012) have been suggested to contribute to predictive information processing (Salet et al, 2022; Stachenfeld et al, 2017). In contrast, reactive network dynamics were influenced strongly by neural populations that showed task-related activity modulations near the time of stimulus presentation, consistent with a role in selective attention (Summerfield & Egner, 2009).…”

“…The first major challenge is conceptual: there is a disconnect between two broad views on anticipatory processing of sensory-driven behaviors that have dominated the field, particularly when expectations are related to the timing and probability of occurrence of an impending event. A “predictive” view emphasizes anticipatory mechanisms that prepare behavioral responses that are associated with expected sensory inputs (Friston, 2010; Los et al, 2014; Montague et al, 1996; Rescorla & Wagner, 1972; Salet et al, 2022). In contrast, a “reactive” view emphasizes anticipatory mechanisms that prioritize (e.g., via selective attention) how expected sensory inputs are processed once they arrive (Summerfield & Egner, 2009).…”