Effects of blocked versus interleaved training on relative value learning.

In reinforcement learning tasks, people learn the values of options relative to other options in the local context. Prior research suggests that relative value learning is enhanced when choice contexts are temporally clustered in a blocked sequence compared to a randomly interleaved sequence. The present study was aimed at further investigating the effects of blocked versus interleaved training using a choice task that distinguishes among different contextual encoding models. Our results showed that the presentation format in which contexts are experienced can lead to qualitatively distinct forms of relative value learning. This conclusion was supported by a combination of model-free and model-based analyses. In the blocked condition, choice behavior was most consistent with a reference point model in which outcomes are encoded relative to a dynamic estimate of the contextual average reward. In contrast, the interleaved condition was best described by a range-frequency encoding model. We propose that blocked training makes it easier to track contextual outcome statistics, such as the average reward, which may then be used to relativize the values of experienced outcomes. When contexts are interleaved, range-frequency encoding may serve as a more efficient means of storing option values in memory for later retrieval.

Testing models of context-dependent outcome encoding in reinforcement learning.

Previous studies of reinforcement learning (RL) have established that choice outcomes are encoded in a context-dependent fashion. Several computational models have been proposed to explain context-dependent encoding, including reference point centering and range adaptation models. The former assumes that outcomes are centered around a running estimate of the average reward in each choice context, while the latter assumes that outcomes are compared to the minimum reward and then scaled by an estimate of the range of outcomes in each choice context. However, there are other computational mechanisms that can explain context dependence in RL. In the present study, a frequency encoding model is introduced that assumes outcomes are evaluated based on their proportional rank within a sample of recently experienced outcomes from the local context. A range-frequency model is also considered that combines the range adaptation and frequency encoding mechanisms. We conducted two fully incentivized behavioral experiments using choice tasks for which the candidate models make divergent predictions. The results were most consistent with models that incorporate frequency or rank-based encoding. The findings from these experiments deepen our understanding of the underlying computational processes mediating context-dependent outcome encoding in human RL.

Reinforcement learning in and out of context: The effects of attentional focus.

In reinforcement learning (RL) tasks, decision makers learn the values of actions in a context-dependent fashion. Although context dependence has many advantages, it can lead to suboptimal preferences when choice options are extrapolated beyond their original encoding contexts. Here, we tested whether we could manipulate context dependence in RL by introducing a secondary task designed to bias attention toward either absolute or relative outcomes. Participants completed a learning phase that involved choices between two (Experiment 1; n = 111) or three (Experiment 2; n = 90) options per trial with complete feedback. Choice options were grouped in stable contexts so that only a small set of the possible combinations were encountered. One group of participants rated how they felt about particular options (Feelings condition), and another group reported how much they expected to win from particular options (Outcomes condition) at occasional points throughout the learning phase. A third group (Control condition) made no ratings. In the subsequent transfer test, participants chose between all possible pairs of options without feedback. The experimental manipulation had no effect on learning phase performance but a significant effect on transfer, with the Feelings and Control conditions exhibiting greater context dependence than the Outcomes condition. Further, rated feelings reflected relative valuation whereas expected outcomes were more sensitive to absolute option values. Hierarchical Bayesian modeling was used to summarize the findings from both experiments. Our results suggest that attending to affective reactions versus expected outcomes moderates the effects of encoding context on subsequent choices. (PsycInfo Database Record (c) 2023 APA, all rights reserved).

Context effects on choice under cognitive load.

The context-dependent nature of choice is well illustrated by decoy effects, in which adding an alternative to a choice set can change the preference relations among the other alternatives. The current within-subjects study tested whether manipulating cognitive load affects the magnitude of attraction and compromise decoy effects. Participants (n = 96) made simulated online grocery shopping choices from three options described by price and quality for each grocery item they encountered. On half the 96 trials, they had to memorize a telephone number prior to encountering the choice set, after which they recalled the number. The choice task was rated significantly more difficult under load, providing some face validity for the load manipulation. Across decoy types, context effects were large and unaffected by the load manipulation. Bayesian analysis provided substantial evidence in favor of this null effect, with the study powered at better than .95 to detect a moderate effect. Across individuals, the magnitude of decoy effects was positively correlated with perception of the greater difficulty of the task under load, with this relationship fully mediated by increases in response times. These results are consistent with the idea that compromise and attraction decoy effects can operate relatively automatically and require minimal effortful processing.

Crossmodal negativity bias in semantic processing.

Negative events have greater influence on cognitive processing compared with positive events, a phenomenon referred to as the negativity bias. Previous studies have shown that reaction times (RTs) to negatively valenced items are slower in semantic tasks. According to the automatic vigilance hypothesis, these effects are caused by preferential attention to negative stimuli or features diverting cognitive resources away from semantic processing. However, it is still unclear whether the negativity bias can be modulated by affective context in a crossmodal setting and how that occurs. Experiment 1 examined individually presented pictures and words and established that participants were slower to judge negatively valenced picture and word targets in a semantic task. Experiments 2 and 3 probed the crossmodal influences of valence on subsequent semantic processing by using short music clips as primes and valenced pictures or words as targets. Both experiments demonstrated that priming negative versus positive music slowed RTs in a semantic task, irrespective of target valence. Hierarchical Bayesian drift diffusion model analyses suggest that the slow-down effects for negative conditions are mainly attributed to reduced drift rates. Together, these experiments demonstrate that negative auditory valence can impair subsequent semantic processing of visual targets in an additive fashion. These results provide further support for the automatic vigilance hypothesis. (PsycInfo Database Record (c) 2022 APA, all rights reserved).

Evaluating non-affective cross-modal congruence effects on emotion perception.

Although numerous studies have shown that people are more likely to integrate consistent visual and auditory signals, the role of non-affective congruence in emotion perception is unclear. This registered report examined the influence of non-affective cross-modal congruence on emotion perception. In Experiment 1, non-affective congruence was manipulated by matching or mismatching gender between visual and auditory modalities. Participants were instructed to attend to emotion information from only one modality while ignoring the other modality. Experiment 2 tested the inverse effectiveness rule by including both noise and noiseless conditions. Across two experiments, we found the effects of task-irrelevant emotional signals from one modality on emotional perception in the other modality, reflected in affective congruence, facilitation, and affective incongruence effects. The effects were stronger for the attend-auditory compared to the attend-visual condition, supporting a visual dominance effect. The effects were stronger for the noise compared to the noiseless condition, consistent with the inverse effectiveness rule. We did not find evidence for the effects of non-affective congruence on audiovisual integration of emotion across two experiments, suggesting that audiovisual integration of emotion may not require automatic integration of non-affective congruence information.

Evoked and induced power oscillations linked to audiovisual integration of affect.

Our affective experiences are influenced by combined multisensory information. Although the enhanced effects of congruent audiovisual information on our affective experiences have been well documented, the role of neural oscillations in the audiovisual integration of affective signals remains unclear. First, it is unclear whether oscillatory activity changes as a function of valence. Second, the function of phase-locked and non-phase-locked power changes in audiovisual integration of affect has not yet been clearly distinguished. To fill this gap, the present study performed time-frequency analyses on EEG data acquired while participants perceived positive, neutral and negative naturalistic video and music clips. A comparison between the congruent audiovisual condition and the sum of unimodal conditions was used to identify supra-additive (Audiovisual > Visual + Auditory) or sub-additive (Audiovisual < Visual + Auditory) integration effects. The results showed that early evoked sub-additive theta and sustained induced supra-additive delta and beta activities are linked to audiovisual integration of affect regardless of affective content.

Autonomic responses to choice outcomes: Links to task performance and reinforcement-learning parameters.

Previous work has shown that autonomic responses to choice feedback can predict subsequent decision-making. In this study, we tested whether skin conductance responses (SCRs) and heart rate (HR) decelerations following the presentation of choice outcomes predict Iowa Gambling Task performance in nonclinical participants (n = 64). We also examined how these signals related to parameters of a reinforcement-learning (RL) model. Feedback SCRs and HR decelerations were greater following outcomes that included losses and choices from the bad decks defined by their negative expected value. In addition, SCRs predicted task performance. A hierarchical Bayesian RL model indicated that greater feedback SCR for the bad decks compared to good decks was associated with stronger loss aversion and a lower learning rate, both of which predicted higher performance. These results suggest that feedback-related SCRs are linked to individual differences in outcome evaluation and learning processes that guide reinforcement-learning.

A study in affect: Predicting valence from fMRI data.

Previous studies have shown that affective valence states induced by brief stimulus presentations are identifiable from whole brain activation patterns observed with functional magnetic resonance imaging (fMRI). However, it is unclear whether those results will generalize to identification of continuous changes in affective valence states under naturalistic settings, such as watching a movie. We examined neural representations of signed (positive versus negative) and unsigned (valenced versus non-valenced) valence on previously collected fMRI data from 17 participants who watched a TV show episode in a passive viewing task in the scanner (Chen et al., 2017). These data were correlated with behavioral valence ratings from a separate group of 125 participants. We spatially localized both signed and unsigned valence representations and were able to predict valence ratings for most participants based on the signed valence model in a cross-participant cross-validation procedure. These findings extend previous results from controlled experimental studies to naturalistic settings, demonstrating the ecological validity of prior findings.

An fMRI Study of Affective Congruence across Visual and Auditory Modalities.

Evaluating multisensory emotional content is a part of normal day-to-day interactions. We used fMRI to examine brain areas sensitive to congruence of audiovisual valence and their overlap with areas sensitive to valence. Twenty-one participants watched audiovisual clips with either congruent or incongruent valence across visual and auditory modalities. We showed that affective congruence versus incongruence across visual and auditory modalities is identifiable on a trial-by-trial basis across participants. Representations of affective congruence were widely distributed with some overlap with the areas sensitive to valence. Regions of overlap included bilateral superior temporal cortex and right pregenual anterior cingulate. The overlap between the regions identified here and in the emotion congruence literature lends support to the idea that valence may be a key determinant of affective congruence processing across a variety of discrete emotions.

Correction to: Context effects on reproduced magnitudes from short-term and long-term memory.

The citation of Hellström (1985) in the body and the reference section of this article is incorrectly printed as Helström.

Context effects on reproduced magnitudes from short-term and long-term memory.

Extant research has demonstrated strong contextual dependencies in reproducing magnitudes of perceptual stimuli from short-term memory. Two experiments examined how context as defined by (a) the mean of the distribution, (b) stimulus ranks, (c) values of anchor stimuli used in the reproduction task, and (d) values from the most recent trial operate on estimates of square size. Experiment 1 demonstrated distributional contrast effects on ratings of squares and distributional assimilation effects on reproduction of squares from short-term memory for the same participants. The fit of a modified version of the category adjustment model demonstrated reliable effects of the running mean, start anchors, and previous stimulus on reproduction bias. In Experiment 2, participants first learned to associate labels with squares, then reproduced square sizes based on the label cues, a long-term memory task, followed by a reproduction from short-term memory task as in Experiment 1. Results for the short-term memory task were largely consistent with Experiment 1. Results for the long-term memory task showed a very different pattern of effects, with larger reproduced sizes when squares were drawn from positively skewed rather than negatively skewed distributions. This contrast effect was explained by a modified range-frequency model as the result of rank encoding of square values along with displacement away from the running mean and shifts towards the prior response and start anchors. The combined results identify multiple sources of context effects in estimation that depend critically on memory retrieval factors and show how they can be incorporated into existing models.

Audiovisual Representations of Valence: a Cross-study Perspective.

Hedonic valence describes the pleasantness or unpleasantness of psychological states elicited by stimuli and is conceived as a fundamental building block of emotional experience. Multivariate pattern analysis approaches contribute to the study of valence representation by allowing identification of valence from distributed patterns of activity. However, the issue of construct validity arises in that there is always a possibility that classification results from a single study are driven by factors other than valence, such as the idiosyncrasies of the stimuli. In this work, we identify valence across participants from six different fMRI studies that used auditory, visual, or audiovisual stimuli, thus increasing the likelihood that classification is driven by valence and not by the specifics of the experimental paradigm of a particular study. The studies included a total of 93 participants and differed on stimuli, task, trial duration, number of participants, and scanner parameters. In a leave-one-study-out cross-validation procedure, we trained the classifiers on fMRI data from five studies and predicted valence, positive or negative, for each of the participants in the left-out study. Whole-brain classification demonstrated a reliable distinction between positive and negative valence states (72% accuracy). In a searchlight analysis, the representation of valence was localized to the right postcentral and supramarginal gyri, left superior frontal and middle frontal cortices, and right pregenual anterior cingulate and superior medial frontal cortices. The demonstrated cross-study classification of valence enhances the construct validity and generalizability of the findings from the combined studies.

Context effects in reproduction and perception of song tempo.

Studies of tempo perception suggest that exposure to a distribution of predominantly faster or slower versions of a song can shift one's memory for the original tempo toward the contextual tempos. Three experiments were conducted to examine whether similar assimilation effects would occur when participants are asked to reproduce the tempo of a song from memory. In Experiments 1 and 2, participants listened to a tempo-altered version of a pop song prior to tapping out the original tempo from memory on each trial. Reproduced tempos assimilated toward the immediately preceding tempo, but there was no evidence of global assimilation toward the mean of the distribution of tempos. However, Experiment 2 demonstrated a partial dissociation between perception and production, with the same participants showing large assimilation effects derived from comparative judgments but not from tempo reproduction. In Experiment 3, participants listened to and then tapped out the beat of a tempo-altered version before reproducing the original from memory on each trial, which resulted in a global assimilation effect in reproduction. The results of these experiments highlight that contextual bias in memory for tempo depends on the match between the context and the task, with differential effects for perceptual and motor contexts. (PsycINFO Database Record (c) 2020 APA, all rights reserved).

Distinguishing abstract from concrete concepts in supramodal brain regions.

Concrete words have been shown to have a processing advantage over abstract words, yet theoretical accounts and neural correlates underlying the distinction between concrete and abstract concepts are still unresolved. In an fMRI study, participants performed a property verification task on abstract and concrete concepts. Property comparisons of concrete concepts were predominantly based on either visual or haptic features. Multivariate pattern analysis successfully distinguished between abstract and concrete stimulus comparisons at the whole brain level. Multivariate searchlight analyses showed that posterior and middle cingulate cortices contained information that distinguished abstract from concrete concepts regardless of feature dominance. These results support the view that supramodal convergence zones play an important role in representation of concrete and abstract concepts.

Temporal dynamics of audiovisual affective processing.

This study used event-related potentials (ERPs) to investigate the time course of auditory, visual, and audiovisual affective processing. Stimuli consisted of naturalistic silent videos, instrumental music clips, or combination of the two, with valence varied at three levels for each modality and arousal matched across valence conditions. Affective ratings of the unimodal and multimodal stimuli showed evidence of visual dominance, congruency, and negativity dominance effects. ERP results for unimodal presentations revealed valence effects in early components for both modalities, but only for the visual condition in a late positive potential. The ERP results for multimodal presentations showed effects for both visual valence and auditory valence in three components, early N200, P300 and LPP. A modeling analysis of the N200 component suggested its role in the visual dominance effect, which was further supported by a correlation between behavioral visual dominance scores and the early ERP components. Significant congruency comparisons were also found for N200 amplitudes, suggesting that congruency effects may occur early. Consistent differences between negative and positive valence were found for both visual and auditory modalities in the P300 at anterior electrode clusters, suggesting a potential source for the negativity dominance effect observed behaviorally. The separation between negative and positive valence also occurred at LPP for the visual modality. Significant auditory valence modulation was found for the LPP, implying an integration effect in which valence sensitivity of the LPP emerged for the audiovisual condition. These results provide a basis for mapping out the temporal dynamics of audiovisual affective processing.

Identification of task sets within and across stimulus modalities.

Previous studies have shown that task sets can be identified from functional magnetic resonance imaging (fMRI) data. However, these results may be partially confounded by differences in stimulus features associated with the different tasks. We disentangle stimulus modality and task features by presenting the same stimulus while varying task, and conversely, presenting different stimuli using the same task. Analyses were conducted on fMRI data previously collected on twenty participants who made either affective or semantic judgements of the same music pieces or the same silent video clips (Kim et al., 2017). Holding stimuli constant, task set was identified from fMRI data across individuals from both task activation data and functional connectivity data. Thus, we were able to identify whether participants made affective or semantic judgments when exposed to identical stimuli based on the task activation and functional connectivity data from other participants. Moreover, task set was successfully identified for cross-modal prediction in which stimuli in the training set bore no resemblance to those in the test set (e.g., using videos data to predict task for music data). Brain regions that were sensitive to tasks irrespective of sensory modality were identified by univariate and searchlight analyses of fMRI data. Consistent with a frontal-parietal network, middle frontal gyrus, inferior parietal gyrus, mid-cingulate cortex, and superior temporal sulcus were found to be key regions distinguishing the two task sets.

Modelling audiovisual integration of affect from videos and music.

Two experiments examined how affective values from visual and auditory modalities are integrated. Experiment 1 paired music and videos drawn from three levels of valence while holding arousal constant. Experiment 2 included a parallel combination of three levels of arousal while holding valence constant. In each experiment, participants rated their affective states after unimodal and multimodal presentations. Experiment 1 revealed a congruency effect in which stimulus combinations of the same extreme valence resulted in more extreme state ratings than component stimuli presented in isolation. An interaction between music and video valence reflected the greater influence of negative affect. Video valence was found to have a significantly greater effect on combined ratings than music valence. The pattern of data was explained by a five parameter differential weight averaging model that attributed greater weight to the visual modality and increased weight with decreasing values of valence. Experiment 2 revealed a congruency effect only for high arousal combinations and no interaction effects. This pattern was explained by a three parameter constant weight averaging model with greater weight for the auditory modality and a very low arousal value for the initial state. These results demonstrate key differences in audiovisual integration between valence and arousal.

Representations of modality-general valence for videos and music derived from fMRI data.

This study tested for neural representations of valence that are shared across visual and auditory modalities referred to as modality-general representations. On a given trial participants made either affective or semantic judgments of short silent videos or music samples. For each modality valence was manipulated at three levels, positive, neutral, and negative, while controlling for the level of arousal. Whole-brain crossmodal identification of affect indicated the presence of modality-general valence representations that distinguished 1) positive from negative trials (signed valence) and 2) valenced from non-valenced trials (unsigned valence). These results generalized across the two tasks. Brain regions that were sensitive to valence states in the same way for both modalities were identified by searchlight analysis of fMRI data by comparing the correlation of voxel responses to the same and different valence conditions across the two modalities. These analyses identified seven clusters that distinguished signed valence, unsigned valence or both. Signed valence was represented in the precuneus, unsigned valence in the bilateral medial prefrontal cortex, superior temporal sulcus (STS)/postcentral, and middle frontal gyrus (MFG) and both types were represented in the STS/MFG and thalamus. These results support the idea that modality general valence is represented in a network of several locations throughout the brain.

Identifying Core Affect in Individuals from fMRI Responses to Dynamic Naturalistic Audiovisual Stimuli.

Recent research has demonstrated that affective states elicited by viewing pictures varying in valence and arousal are identifiable from whole brain activation patterns observed with functional magnetic resonance imaging (fMRI). Identification of affective states from more naturalistic stimuli has clinical relevance, but the feasibility of identifying these states on an individual trial basis from fMRI data elicited by dynamic multimodal stimuli is unclear. The goal of this study was to determine whether affective states can be similarly identified when participants view dynamic naturalistic audiovisual stimuli. Eleven participants viewed 5s audiovisual clips in a passive viewing task in the scanner. Valence and arousal for individual trials were identified both within and across participants based on distributed patterns of activity in areas selectively responsive to audiovisual naturalistic stimuli while controlling for lower level features of the stimuli. In addition, the brain regions identified by searchlight analyses to represent valence and arousal were consistent with previously identified regions associated with emotion processing. These findings extend previous results on the distributed representation of affect to multimodal dynamic stimuli.