Sensory and action neural tuning explains how priors guide human visual decisions

Sensory and action neural tuning explains how priors guide human visual decisions | bioRxiv /* */ /* */ <!-- <!-- /*! * yepnope1.5.4 * (c) WTFPL, GPLv2 */ (function(a,b,c){function d(a){return"[object Function]"==o.call(a)}function e(a){return"string"==typeof a}function f(){}function g(a){return!a||"loaded"==a||"complete"==a||"uninitialized"==a}function h(){var a=p.shift();q=1,a?a.t?m(function(){("c"==a.t?B.injectCss:B.injectJs)(a.s,0,a.a,a.x,a.e,1)},0):(a(),h()):q=0}function i(a,c,d,e,f,i,j){function k(b){if(!o&&g(l.readyState)&&(u.r=o=1,!q&&h(),l.onload=l.onreadystatechange=null,b)){"img"!=a&&m(function(){t.removeChild(l)},50);for(var d in y[c])y[c].hasOwnProperty(d)&&y[c][d].onload()}}var j=j||B.errorTimeout,l=b.createElement(a),o=0,r=0,u={t:d,s:c,e:f,a:i,x:j};1===y[c]&&(r=1,y[c]=[]),"object"==a?l.data=c:(l.src=c,l.type=a),l.width=l.height="0",l.onerror=l.onload=l.onreadystatechange=function(){k.call(this,r)},p.splice(e,0,u),"img"!=a&&(r||2===y[c]?(t.insertBefore(l,s?null:n),m(k,j)):y[c].push(l))}function j(a,b,c,d,f){return q=0,b=b||"j",e(a)?i("c"==b?v:u,a,b,this.i++,c,d,f):(p.splice(this.i++,0,a),1==p.length&&h()),this}function k(){var a=B;return a.loader={load:j,i:0},a}var l=b.documentElement,m=a.setTimeout,n=b.getElementsByTagName("script")[0],o={}.toString,p=[],q=0,r="MozAppearance"in l.style,s=r&&!!b.createRange().compareNode,t=s?l:n.parentNode,l=a.opera&&"[object Opera]"==o.call(a.opera),l=!!b.attachEvent&&!l,u=r?"object":l?"script":"img",v=l?"script":u,w=Array.isArray||function(a){return"[object Array]"==o.call(a)},x=[],y={},z={timeout:function(a,b){return b.length&&(a.timeout=b[0]),a}},A,B;B=function(a){function b(a){var a=a.split("!"),b=x.length,c=a.pop(),d=a.length,c={url:c,origUrl:c,prefixes:a},e,f,g;for(f=0;f<d;f++)g=a[f].split("="),(e=z[g.shift()])&&(c=e(c,g));for(f=0;f<b;f++)c=x[f](c);return c}function g(a,e,f,g,h){var i=b(a),j=i.autoCallback;i.url.split(".").pop().split("?").shift(),i.bypass||(e&&(e=d(e)?e:e[a]||e[g]||e[a.split("/").pop().split("?")[0]]),i.instead?i.instead(a,e,f,g,h):(y[i.url]?i.noexec=!0:y[i.url]=1,f.load(i.url,i.forceCSS||!i.forceJS&&"css"==i.url.split(".").pop().split("?").shift()?"c":c,i.noexec,i.attrs,i.timeout),(d(e)||d(j))&&f.load(function(){k(),e&&e(i.origUrl,h,g),j&&j(i.origUrl,h,g),y[i.url]=2})))}function h(a,b){function c(a,c){if(a){if(e(a))c||(j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}),g(a,j,b,0,h);else if(Object(a)===a)for(n in m=function(){var b=0,c;for(c in a)a.hasOwnProperty(c)&&b++;return b}(),a)a.hasOwnProperty(n)&&(!c&&!--m&&(d(j)?j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}:j[n]=function(a){return function(){var b=[].slice.call(arguments);a&&a.apply(this,b),l()}}(k[n])),g(a[n],j,b,n,h))}else!c&&l()}var h=!!a.test,i=a.load||a.both,j=a.callback||f,k=j,l=a.complete||f,m,n;c(h?a.yep:a.nope,!!i),i&&c(i)}var i,j,l=this.yepnope.loader;if(e(a))g(a,0,l,0);else if(w(a))for(i=0;i (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0];var j=d.createElement(s);var dl=l!='dataLayer'?'&l='+l:'';j.src='//www.googletagmanager.com/gtm.js?id='+i+dl;j.type='text/javascript';j.async=true;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-M677548'); Skip to main content Home About Submit ALERTS / RSS Search for this keyword Advanced Search New Results Sensory and action neural tuning explains how priors guide human visual decisions View ORCID Profile Gabriela Iwama , View ORCID Profile Randolph F. Helfrich doi: https://doi.org/10.1101/2025.03.26.645167 Gabriela Iwama 1 Hertie Institute for Clinical Brain Research, Center for Neurology, University Medical Center Tübingen , Tübingen, Germany 2 University of Tübingen, Graduate Training Center of Neuroscience , Tübingen, Germany 3 International Max Planck Research School for the Mechanisms of Mental Function and Dysfunction, Max Planck for Biological Cybernetics , Tübingen, Germany Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Gabriela Iwama For correspondence: gabriela.iwama{at}uni-tuebingen.de randolph.helfrich{at}gmail.com Randolph F. Helfrich 1 Hertie Institute for Clinical Brain Research, Center for Neurology, University Medical Center Tübingen , Tübingen, Germany Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Randolph F. Helfrich For correspondence: gabriela.iwama{at}uni-tuebingen.de randolph.helfrich{at}gmail.com Abstract Full Text Info/History Metrics Supplementary material Preview PDF Abstract Prior expectations bias how we perceive the world. Despite well-characterized behavioral effects of priors, such as confirmation bias, their neural mechanisms remain unclear. Contemporary theories postulate conflicting predictions: does the brain enhance expected sensory information (sharpening the expected representation) or rather prioritize unexpected information (dampening the expected representations)? Here, we combined reversal learning with a noisy motion discrimination task to investigate how priors impact sensory and action information processing. Using behavioral modeling to infer participants’ latent priors and EEG to track neural dynamics, we demonstrated that priors differentially impact sensory and action representations over time. While priors introduced long-lasting biases on action coding irrespective of their validity, sensory information was selectively enhanced with confirmatory evidence. Critically, dampening of action representations predicted confirmation biases, whereas sensory tuning dynamics tracked speed-accuracy trade-offs. These findings reveal a dissociable and temporally dynamic influence of priors in visual decisions, reconciling competing theories of predictive perception. Significance statement Prior expectations, abstract rules, and recent decisions impact perception and ensuing actions. While priors are beneficial in most circumstances, they might be detrimental in other scenarios, as exemplified by confirmation biases, serial response biases, or switch costs. Albeit well-characterized at the behavioral level, the neural mechanisms underlying these biases remain undetermined. By tracking participants’ latent priors and neural dynamics with high-density EEG, the authors demonstrate that priors differentially impact sensory and action information, uncovering the neural mechanisms underlying confirmation biases and speed-accuracy trade-offs in predictive perception. Introduction In a dynamic and uncertain world, sensory evidence is often limited. Thus, the ability to flexibly make perceptual decisions based on available prior and current information is fundamental for survival. An inherent speed-accuracy trade-off arises for almost every perceptual decision: should one rely on prior knowledge or gather more evidence before committing to a choice? For example, prior information on which trees are likely to bear fruit can save time and energy if correct, but relying on it may incur delays and opportunity costs if the prior proves wrong, leading to confirmation and choice history biases 1 – 3 . While the effect of priors on behavior is well-characterized, it remains unclear how the brain flexibly integrates prior information with ongoing sensory input to guide adaptive behavior. There are two main theories of how prior information is integrated with sensory input: Bayesian and cancellation theories. According to Bayesian theories 4 – 6 , sensory information is biased relative to its prior probability. Therefore, they predict a relative enhancement of expected sensory input (sensory sharpening). In contrast, cancellation theories predict a prioritization of informative signals given the limited capacity of the brain 7 , 8 . Since unexpected sensory inputs are more informative than expected ones, cancellation theories predict a relative dampening of expected sensory inputs. Therefore, these theories have conflicting predictions of how sensory processing is modulated by prior information. What are the possible solutions to these contradictions? One possibility is that action and sensory processes are modulated by distinct regimes. Bayesian explanations are highly prevalent within the perceptual domain 5 , 9 , 10 , whereas cancellation theories are mostly prevalent within the action literature 7 , 11 . Another explanation is that the modulation of sensory information is a dynamic process that depends on the validity of the prior as evidence is accumulated over time. As proposed by the Opponent Process Theory 8 , sensory information is initially sharpened by prior expectations and only dampened if a prediction error is signaled by a bottom-up sensory process. While both hypotheses have received empirical support, both have limited power to explain the balance between speed and accuracy in visual decisions. To resolve this debate, it is fundamental to investigate both action and sensory processing when priors are confirmed or disconfirmed by sensory information during rapid decision-making. Here, we investigated how prior expectations modulate action and sensory information during a cued motion discrimination task. Critically, the cue validity changed across trials, and participants had to adaptively use the cue information, the prior, across trials. Behavioral modeling was employed to estimate participants’ prior strength, while multivariate decoding was used to assess high-density electroencephalography (EEG) to capture sensory and action information. Our results show that prior expectations distinctively modulate action and sensory processing while driving a dynamic sensory tuning that supports flexible decision-making. Results To assess how prior expectations modulate sensory information and ensuing actions, participants completed a cued motion discrimination task ( Fig. 1a ). The cues indicated which direction of motion was more likely to appear in the upcoming stimulus. Cue validity was unknown to the participants. Each trial began with participants selecting one of two cues (orange or blue) and rating their confidence in its validity (low or high). Next, the cue was presented indicating a direction (left or right), followed by a Random Dot Kinematogram (RDK) with titrated left/right motion coherence to achieve 75% accuracy. Participants reported the perceived motion direction (left/right) on a response box. As stimuli were titrated to perceptual threshold and included high noise levels, participants were instructed to rely on the cue information (i.e., the prior) to guide their motion choices. Download figure Open in new tab Fig. 1 Experimental design, modeling, and validation. a, On every trial, participants selected either the orange or the blue cue, rated their confidence (low or high) that the selected cue would be valid, and were subsequently presented a cue (colored arrow) indicating the direction of the Random Dot Kinematogram (RDK; left or right). After a delay, participants were presented with the RDK at their individual discrimination level (threshold at 75% accuracy) and made a decision about the motion direction (left or right). At the end of each trial, participants received feedback about their motion choice and the validity of the chosen cue. b, The Volatile Kalman Filter generative model 12 contains two hidden Markov chains that dynamically diffuse over time, in which x t denotes whether the blue cue is the most informative, and z t denotes the inverse volatility of x t . The outcome o t ∈ {0,1} consists of the observed cue validity, where 0 = blue invalid or orange valid and 1 = blue valid or orange invalid. The choice, y t , is given by a softmax function of x t with a free parameter, β , to model how rationally participants follow their prior estimates. The free parameters λ , σ , and v 0 indicate the (inverse) volatility update rate, the observation noise, and the initial volatility. c, Single subject example of best cue (true contingency; blue), cue choice ( y t , orange), outcome ( o t ; yellow), and estimated prior ( x t ; purple) across trials. Note that the estimated prior closely tracked reversals in the true contingency. d, Estimated latent prior of chosen cue according to confidence judgments (left), block type (middle), and cue accuracy according to prior strength during informative blocks (right). Violin plots show the median, interquartile range (IQR), and ±1.5 IQR . In this task design, we modulated the strength of prior information by varying cue validity across trials. A block design alternated between informative and non-informative trials. In informative blocks, one cue had a validity of 80%, while the other cue had a validity of 30%. In non-informative blocks, both cues had low validity (30%). Prior to the experiment, participants were informed that the validity of the cues would change but were not informed of the timing or frequency of these changes. Therefore, participants had to learn the true validity of the cues through experience. To encourage participants to balance making accurate motion choices against learning the cue validities, a dual reward system was implemented (see Methods ). Participants were rewarded for correct motion choices and for selecting a valid cue. Cue rewards were scaled by their reported confidence. If participants reported high confidence in their selected cue, they entered a high-risk, high-reward state. In contrast, reporting low confidence placed participants in a safe reward scheme with lower potential gains and substantially reduced potential losses. This design incentivized participants to report low confidence when they were uncertain about the cue validity. However, this reward system does not guarantee that participants will use the prior information during the decision. To promote a flexible reliance on prior information, we induced a speed-accuracy trade-off. Participants were penalized if they did not respond to the perceived direction of motion within 1000 ms of RDK onset. In this design, it is beneficial to rely on the prior even when no coherent motion is perceived. To optimize rewards, participants had to learn which cue was the best and adaptively use the prior information over trials. A key feature of this task is that it separates the cue choice from the cue information and the motion choice. This design disentangles the process of learning the prior (the cue validity) from forming a specific visual prior (left or right motion) and making a perceptual choice (left or right button response). As a result, it allows assessing how prior expectations modulate both sensory (motion perception) and action (overt response) information processing at different stages. Latent priors track reported confidence and learning Participants’ latent priors (i.e., their subjective belief about the cue’s validity at each trial) were estimated using the Volatile Kalman Filter 12 (VKF; Fig. 1b ), which models environmental volatility (see Methods ). The VKF was fitted to participants’ cue choices and outcomes using Laplace Approximation (see Fig. 1c for a single-subject example). To validate the estimated priors, we compared how they varied depending on participants’ reported confidence, block type, and cue accuracy on informative blocks. Across participants, prior strength was significantly higher when confidence was high ( Fig. 1d ; p < .001, t(19) = 8.22, d = 1.84) and when the block was informative (p < .001, t(19) = 12.30, d = 2.75). Moreover, participants more frequently selected the true best cue in an informative block when priors were stronger (median split; p < .001, t(17) = 10.78, d = 2.54). Together, these findings provide robust evidence supporting the validity of the priors—as estimated by the model—as a measure of participants’ latent priors. Priors increase the influence of cue information during motion decisions Having validated the estimated prior, we examined whether participants integrated the prior into their visual decisions. We performed a trial-wise regression analysis on motion accuracy and reaction time, with prior strength and cue validity interaction as a predictor. Additionally, a random intercept for each participant was included to account for individual differences in performance. The interaction effect was statistically significant for both accuracy, ( Fig. 2a ; t(6004) = 10.19; f 2 = .019) and reaction time ( Fig. 2b ; p < .001; t(6004) = -7.19; f 2 = .011). Hence, the behavioral benefit of valid cues increased when the prior was high for both accuracy ( Benefit ACC = ACC High , Valid − ACC Low , Valid ; M benefit = 4.1%, SD = 6.6%) and reaction time ( Benefit RT = RT Low , Valid − RT High , Valid ; M benefit = 16 ms, SD = 27 ms). Conversely, invalid cues incurred an increased behavioral cost when the prior was high for both accuracy ( Cost ACC = ACC High,Invalid − ACC Low , Invalid ; M Cost = 4.1%; SD = 9.1%) but not for reaction time ( Cost RT = RT High , invalid − RT Low , Invalid ; M Cost = -7 ms; SD = 29 ms). These analyses indicate that participants adaptively used the cue information during their decision-making, relying on it more when their latent prior was high and sensory evidence was confirmatory. After validating that the task promoted the flexible use of priors in visual decisions across trials, we next examined individual differences in prior integration. Download figure Open in new tab Fig. 2 Priors induce confirmation biases and speed-accuracy trade-offs. a, Accuracy (ACC) of motion choice as a function of cue validity and prior strength (subject-level median split into high and low trials). Green and red arrows illustrate the group-level behavioral benefit ( Benefit ACC ) and cost ( Cost ACC ) of high prior when compared to low prior trials within the same cue validity, respectively. Black arrows indicate the behavioral modulation of high prior trials depending on cue validity (Δ ACC ). b, Same conventions as in panel ( a ) but for reaction time (RT). c, Correlation between the difference in accuracy for high prior trials when the cue was valid and invalid (Δ ACC = ACC High , Valid − ACC High , Invalid ) and the difference in reaction time for the same conditions (ΔRT, excluding incorrect trials). d, Correlation between the benefit in reaction time ( Benefit RT = RT Low , Valid − RT High , Valid ) and the cost in accuracy ( Cost ACC = ACC Low , Invalid − ACC High , Invalid ). Individual differences reveal rigid and flexible prior integration To test individual differences in how priors are integrated, we correlated the costs, benefits, and validity modulation in both accuracy and reaction time. First, when priors were strong, the modulation of validity in accuracy (Δ ACC = ACC High , Valid − ACC High , Invalid ) was negatively associated with the modulation of reaction time ( Fig. 2c ; Δ RT = RT High , Valid − RT High , Invalid ; r = -.79, p < .001). This indicates that participants who were more biased to judge the motion direction toward the cue also responded faster when the cue was valid. We will denote the variance along this shared dimension as confirmation bias . Additionally, the costs in accuracy ( Cost ACC ) were positively correlated with the reaction time benefits ( Fig. 2d ; Benefit RT ; r = .67, p < .001), reflecting individual differences in the speed-accuracy trade-off. Specifically, participants who made faster decisions when the cue was valid made more mistakes when the cue was invalid. The variance along this shared dimension will be denoted as speed-accuracy trade-off . In sum, these correlations indicate that participants employ distinct strategies to integrate priors into visual decisions. This raises the question of whether and how these strategies are associated with sensory and action representations. Priors distinctively stabilize sensory and action information To investigate how priors affect sensory and action information during visual decision-making, both uni- and multivariate analyses of high-density EEG recordings were employed. To investigate whether priors modulate the total activation of channels, a cluster-based permutation test across channels and time was used to compare the event-related potential (ERP) of the data time-locked to the RDK onset and reaction time (RT) for each prior strength (high, low) and cue validity (valid, invalid) condition (see Supplementary Figure 1 ). However, no statistically significant clusters across time and channels were observed before or after RDK onset when comparing high versus low prior in each validity condition. Therefore, a multivariate decoding approach was applied to test if priors affect channel covariance. Before assessing how priors modulate sensory and action information, we first decoded both the sensory (left or right motion) and the action information (left or right button response) from high-density EEG in a time-resolved manner ( Fig. 3a /b ). All decoding analyses relied on a Linear Discriminant Analysis 13 , a dimensionality reduction method that finds a linear combination of features (channels) to maximize the separation between classes (left/right motion or choice) while minimizing the within-class variance. To avoid training biases, we employed a 10-fold cross-validation with balanced training samples regarding the decoded feature (action or sensory), the prior strength (high, low), and the cue validity (valid, invalid). Download figure Open in new tab Fig. 3 Stronger priors enhance the generalization of pre-stimulus sensory activity. a, Average true class probability (TCP) of action (left/right button response) or sensory (left or right motion) for each time point across subjects. All decoding analyses were run using a linear discriminant analysis on the data from 257 EEG channels across trials, 10-fold cross-validation, and balancing the trials according to the decoded feature, the prior strength, and cue validity using an oversampling procedure. Data was smoothed with a Gaussian-weighted average over 100 ms window for visualization only. Shaded areas indicate standard deviation. b, Same as ( a ) but for data locked to the response. c, To estimate the common effect of prior strength irrespective of validity, the low prior condition matrix was subtracted from the high prior condition matrix for both valid (V) and invalid (IV) conditions, and the resulting differences were summed for each subject. d, Average generalization performance of action decoding across time for all trials and f, sensory decoding. e, Estimated common effect of prior strength for each training and test time for action and g, sensory decoding. For all time generalization matrices, red contours indicate statistically significant generalization times using a cluster-based permutation test. Black contours indicate the cluster with the highest t-value if none were statistically significant. h, Average TCP of action information across all testing times for every pre-stimulus training time. Top lines indicate the time points of significant clusters when testing differences between high and low prior trials using a cluster-based permutation test. i, Same as ( h ) but for sensory decoding. Across participants, single-trial true class probabilities (TCP) of testing samples were significantly decoded above chance after stimulus presentation for sensory (p = .041, t̅ cluster = 2.77; d = .78; cluster-based permutation test) and action information (p = .001, t̅ cluster = 3.48; d = 1.25; cluster-based permutation test), as well as reaction time (p sensory = .004, t̅ cluster = 3.40; d = .95; p action = .001, t̅ cluster = 4.59; d = 1.62; cluster-based permutation test). Additionally, for action, we observed a cluster even before the RDK onset, [-.020 s .086 s], p = .047, t̅ cluster = 2.44, d = .63; cluster-based permutation test), suggesting that the cue already influences the action information before sensory processing. Having demonstrated that sensory and action information can be decoded, we tested whether prior strength modulates sensory or action information. TCP differences between weak and strong priors were compared for each validity condition. As no statistically significant differences were found (see Supplementary Figure 2 ), we determined whether prior strength modulates the temporal evolution of sensory and action information over the trial. To examine how sensory and action information evolve over time, we assessed the temporal cross-generalization of the decoding algorithm by training it at each time step and testing it across all time steps. First, the generalization of sensory and action information was compared to chance level across all trials. We observed a statistically significant generalization above chance for action information during the entire trial (p = .005, t̅ cluster = 1.36; d = .30) but not for sensory information ( Fig. 3d /f ). These results suggest that priors might differentially impact sensory and action information according to their relative strength. To investigate the effect of prior strength irrespective of their validity. We estimated the common effect of prior strength by summing up the differences in TCP between high and low for both validity conditions ( Fig. 3c ). Strong prior exhibited an enhanced generalization of pre-stimulus sensory information (p = .043, t̅ cluster = 1.60; d = .36), but not of action information ( Fig. 3e /g ). Specifically, the decoded sensory information at the end of the cue presentation (p = .03, t̅ cluster = -3.75; d = .96) and during the delay period (p = .048, t̅ cluster = -2.53; d = .59) reliably generalized across the trial ( Fig. 3i ). These results establish that stronger priors increase the stability of pre-stimulus sensory representation. Together, these results reveal that priors stabilize action and sensory information in distinct ways. Notably, the analyses did not compare how prior validity influences these effects. This raises the question of whether differences in how priors affect sensory and action representations depend on whether the prior information is confirmed. Effects of prior validity on neural generalization On the behavioral level, the effect of strong priors depended on its validity. Individual differences in confirmation bias showed faster and more accurate decisions for valid cues, as well as slower and less accurate decisions for invalid cues ( Fig. 2c ). Additionally, individual differences in speed-accuracy trade-offs showed that faster decisions for valid cues also resulted in impaired accuracy for invalid cues ( Fig. 2d ). It is unclear whether these behavior effects are reflected in the modulation of neural information by (dis)confirmatory evidence. We therefore investigated whether and how confirmation bias and speed-accuracy trade-offs are correlated with the tuning of neural information by prior validity ( Fig. 4a ). Download figure Open in new tab Fig. 4 Dissociable behavioral correlates of action dampening and sensory sharpening. a, Analysis strategy to trace the temporal evolution of the behavioral effects of action dampening and sensory sharpening. b, Grand-mean difference between the cross-temporal generalization of valid and invalid high prior trials ( TCP High , Valid − TCP High , Invalid ) for each training and test time point. c, Grand-mean TCP over all training and testing times ([-1, 1] s) for each condition. d, Correlation of average cross-temporal action dampening with participant’s behavior modulation (transformed behavior from Fig. 2 c) and e, with participant’s speed-accuracy trade-off (transformed behavior from Fig. 2 d). f–i, display the same results as the panels ( b–e ) but for sensory-like activity. Red contours indicate statistically significant generalization times (cluster-based permutation test, p < .05). Black contours highlight non-significant clusters with the highest t-value. First, we tested whether the generalization of sensory or action information differs between valid and invalid trials when priors were strong. This method isolates the effects of confirmed versus unconfirmed priors, clarifying how information is tuned over time. Specifically, it determines whether expected sensory and action information is sharpened ( TCP Valid , High > TCP Invalid , High ) or dampened ( TCP Invalid , High > TCP Valid , High ). While sensory information displayed no significant tuning effects, a significant dampening of expected action information was observed ( Fig. 4b /f; p = .033, t̅ cluster = -1.36; d = .30) that was most pronounced for pre-stimulus activity. Furthermore, we compared the mean TCP across all training and test times between valid and invalid trials for each prior strength and decoded feature ( Fig. 4c /g ). Congruent with the time-resolved analysis, there was only a statistically significant validity effect for action information when priors were strong (p = .008, t(19) = 2.98, d = .667). These results suggest that a strong prior induced action dampening, particularly during pre-stimulus activity, while sensory information was not consistently altered across participants. Neural tuning dynamics is linked to behavior Having investigated whether and how neural information changes across participants, we tested whether variability in participants’ neural tuning were associated with their behavioral performance. Since action was dampened but sensory processing was not ( Fig. 4c /g ), we examined how action dampening, ( TCP Invalid , High − TCP Valid , High ) and sensory sharpening, ( TCP Valid , High − TCP Invalid , High ) correlated with confirmation bias and speed-accuracy trade-offs across participants ( Fig. 4a ). For the confirmation bias, we identified a statistically significant positive correlation cluster between the dampening of pre-stimulus action information before the button response (p = .041, r̅ cluster = -.44). In contrast, a negative correlation cluster emerged that extended the generalization of pre-stimulus sensory sharpening to the delay and stimulus presentation (p = .014, r̅ cluster = -.48). These findings indicate that participants who dampened pre-stimulus action information before their motor response exhibited greater confirmation bias, while those with stronger sensory sharpening showed less confirmation bias. For the speed-accuracy trade-off ( Fig. 4e /i ), significant clusters were only observed in the correlation with sensory tuning. Specifically, the sharpening of the pre-stimulus sensory information that generalizes to post-stimulus activity was negatively correlated with the speed-accuracy trade-off (p = .043, r̅ cluster = -.43), indicating a focus on accuracy. In contrast, the sharpening of post-stimulus sensory information to pre-stimulus activity was positively correlated (p = .023, r̅ cluster = .49), suggesting a focus on speed. These findings reveal that speed-accuracy trade-offs are predicted by the dynamics of sensory tuning. Collectively, these results demonstrate that action dampening is linked to stronger reliance on the prior information, while dynamic sensory tuning reflects the balance of costs and benefits of prior integration. Discussion Our study reveals how prior expectations influence action and sensory information processes during visual decisions. Behaviorally, prior expectations induced both rigid confirmation bias and flexible speed-accuracy trade-offs, reflecting distinct strategies for handling uncertainty. Neurally, strong priors increased the stability of pre-stimulus sensory information, while action information remained stable regardless of prior strength. By integrating behavioral and neural data, we observed that confirmation biases were linked to increased action dampening, while flexibly integrating priors when faced with new evidence was associated with a dynamic sensory tuning process. These findings suggest that prior expectations differentially modulate action and sensory processing, while driving a dynamic sensory tuning that supports flexible decision-making. Priors distinctively stabilize sensory and action neural information The central observation of the present study is that neural mechanisms that support the emergence of stable neural representations of sensory and action dynamics findings are distinct. We observed that external prior information (cue information) introduced a stable and long-lasting choice bias, characterized by a stable neural representation as revealed by significant temporal cross-decoding, which emerged already prior to onset of the sensory stimulus. Our results further suggest that this bias was largely independent of cue validity, thus, further contributing to stabilizing the neural representation of the ensuing action. In contrast to action, sensory information was stabilized in a context-dependent manner. We observed that the neural representation of pre-stimulus sensory information was boosted when the prior was strong, but not when the prior was weak. This context-dependent stability of sensory information is well aligned with sensory templates in the early visual cortex 14 , 15 , where confirmatory evidence might further strengthen the preexisting representation. Likewise, it is conceivable that confirmed prior expectations (esp. in states of high confidence) decrease neural variability within 16 , 17 or across spatial sites 18 – 20 . In sum, even though representations of prior information are abundant in early sensory as well as motor-related brain areas 21 , priors seem to differentially impact sensory and action information processing. By linking neural and behavioral data, our results demonstrate that confirmation biases are linked to increased dampening of action representations as well as reduced sensory sharpening. In contrast, dynamic sensory tuning is associated with, and potentially contributing to, speed-accuracy trade-offs. These observations can be further conceptualized by previous theoretical accounts. For example, cancellation theories suggest a prioritization of unexpected over expected information. This prediction is consistent with the action dampening observed in the current study. Here, action dampening was associated with confirmation biases. Previous studies have also linked confirmation biases to motor preparation 22 , 23 . These considerations are also fully compatible with a biased starting point of the evidence accumulation process 24 , 25 . In addition, confirmation biases have also been explained by biased evidence accumulation 26 , 27 . Computationally, confirmation biases have been shown to benefit hierarchical inference 28 and have been linked to overweighting categorical information during continuous perceptual inference 29 . By linking confirmation biases with neural information tuning, the present study offers a possible mechanistic explanation that connects evidence accumulation studies to perceptual prediction theories. In addition to cancellation theories, the Opposing Process Theory 8 proposes that sensory information is initially sharpened by prior expectations that can be subsequently dampened if bottom-up processing signals conflict with prior expectations. In line with this theory, we observed that sharpening of pre-stimulus sensory information predicted faster decisions when the prior was confirmed but at the cost of less accurate decisions when the prior was disconfirmed. In contrast, dampening sensory information during the stimulus presentation benefitted task accuracy by overriding the prior when it was disconfirmed but resulted in slower reaction times. Hence, our present results also provide supporting evidence for the Opposing Process Theory in balancing speed and accuracy during visual decisions. In sum, the present results reconcile and dissociate several previous theoretical and empirical findings and indicate that neural implementations of priors differentially impact sensory and motor processes. These considerations raise the question of how different perceptual prediction theories can be reconciled or co-exist. A central question—and a key conflict in existing literature—is whether sensory processes sharpen or dampen expected information 6 , 8 . Our results are incompatible with one central prior coding scheme and rather suggest that independent coding schemes exist for the sensory and action domains. Critically, our behavioral results confirm that each scheme has a differential impact on behavioral performance. These findings raise an important question: Are the differences between action and sensory domain inherent, or are they shaped by the task demands? Inherit or functional differences between sensory and action domains Many accounts of perceptual decision-making characterize sensory and action coding as parallel processes, where sensory evidence accumulation and motor preparation occur simultaneously 30 . Under this view, sensory evidence continuously flows from sensory to motor areas. While several lines of inquiry provided convincing evidence supporting concurrent action integration in motor areas of humans 31 as well as monkeys 32 , 33 , recent findings challenged this view. For example, intracranial recordings in patients with epilepsy revealed effector-independent evidence accumulation signals across diverse perceptual decision tasks 34 . In addition, perceptual decisions seem to be formed before action preparation when evidence must be fully integrated before forming a motor plan 35 . Complementary results from large-scale neural recordings in mice further demonstrate that sensory integration in premotor neural populations operates in a parallel fashion but transitions sharply to an orthogonal action execution subspace once evidence accumulation is completed 36 . These findings across studies and species indicate that sensory evidence accumulation and action execution are inherently distinct processes during perceptual decision-making. While sensory and motor processing are widely supported to occur in relative isolation, the distinctive impact of prior information in these domains might be functional. Under a Bayesian Decision Theory perspective, the posterior belief (i.e., the final choice) depends on the prior information and its likelihood (i.e., the current sensory evidence). Under this view, both sensory evidence and prior information influence the perceptual choice, and thereby, the ensuing action execution. However, as in most experimental designs, only the sensory input can provide immediate additional evidence to update the posterior belief. It is possible that a dynamic neural tuning of predictive information only occurs when it is decision-relevant. Congruent with this idea, prior-induced sensory modulations were only observed when the feature was goal-directed 37 , 38 . Similarly, dampening of predictive information may occur when the information is not needed to update the posterior, such as with action information in most perceptual tasks. These considerations explain why cancellation theories are prevalent in the action and motor literature. Limitations Our study raises considerations that can be explored in future research. First, to experimentally distinguish the formation of a choice variable from the action execution, future studies could exploit experimental designs that prevent the formation of action plans with variable sensory-to-response mappings, such as in Sandhaeger et al. 39 . Second, participants had to choose between two cues, introducing an additional level of uncertainty, particularly due to the inherent exploration-exploitation trade-off when learning under stochastic regimes 40 . The influence of priors during exploratory behavior may differ from their effects during exploitation, enhancing variability in how priors affect sensory tuning. Finally, a sensory sharpening was not consistently observed across participants, raising questions about the timescales of priors and their impact on perceptual processes 5 . Shorter timescales may introduce greater individual variability, while longer timescales yield more homogeneous results. In support of this hypothesis, sensory sharpening has been reported only after extended task exposure 41 , with expectation suppression similarly observed following prolonged training 42 . Additionally, the consolidation of prior information during sleep may account for persistent predictions in early visual areas 43 . Longer-term priors, such as the statistics of natural environment, may optimize sensory processing through mechanisms consistent with efficient coding theories 6 , 44 . In contrast, volatile environments or tasks with shorter timescales require learning a hierarchical structure, which may increase individual variability. Conclusion This study demonstrates that priors can be integrated rigidly or flexibly. Confirmation biases (a rigid tendency to rely on cues) was linked to increased dampening of action information and reduced sensory sharpening. In contrast, speed-accuracy trade-offs (i.e., more flexible prior influence) were attributed to a dynamic sensory tuning mechanism, adjusted based on whether sensory evidence was confirmatory. While sustained sharpening of sensory information predicted faster responses when the prior was confirmed, dampening of expected sensory information upon disconfirmatory evidence overrode the prior; thus, leading to correct decisions. These findings support recent theories of predictive processing and suggest that dynamically tuning prior sensory information according to current evidence is essential for flexible decision-making. Method Participants 20 healthy volunteers (10 females, 20 to 31 years) participated in this study and were recruited from the student community. All participants had normal vision and no photosensitive epilepsy, claustrophobia, color blindness, or other neurological disorders. 19 participants were right-handed. Participants provided informed written consent to participate in the study and were paid for their participation according to their performance. This study was approved by the Ethics Committee of the Medical Faculty Tübingen (protocol number 049/2020B02) and conducted in accordance with the Declaration of Helsinki. Experimental Design Before the main task, the motion coherence threshold was estimated through QuestPlus, an adaptive psychometric method 45 , 46 to yield 75% accuracy. The motion stimuli consisted of a sequence of moving white dots (size: .06 visual angle; density: 16.7 dots/degree; velocity: .03 visual angle/second; lifetime: 12 frames) inside a circular area, centered on a fixation cross, with a diameter of 2.5 visual angle degrees, on a black background. Most dots moved in random directions, while a small subset (up to 20%) coherently moved either to the left or right. Participants were seated 120 cm from a 28” LED screen (60 Hz, 2560×1440 px) in a dark, magnetically and acoustically shielded room. Participants were instructed to respond by pressing the buttons on two response boxes (one for each hand). All stimuli were presented using PsychToolbox for MATLAB 47 , 48 . Behavioral Task The behavioral task combines reversal learning and a random dot motion discrimination task. Each trial contained three main decisions: cue choice, confidence, and motion direction, respectively. The main trial events were interleaved with fixation crosses, as shown in Fig. 1a . The start of the trial was self-paced. The cue identity was established by distinct colors (blue or orange). Immediately after the cue choice, participants rated their confidence that it would be valid. The cue direction was displayed for 300 ms, and its validity was probabilistically determined according to an underlying pseudo-randomized structure (see true contingency in Fig. 1c ). Participants were familiarized with the main task with 15 training trials, in which no validity reversals occurred. Cues within informative blocks had validities of 80% and 30%. The best cue was pseudo-randomized across informative blocks, such that the same cue was never the best on three consecutive informative blocks. Non-informative blocks had both cue validities set at 30%. Cue validity for single trials was pseudo-randomized to have an exponential moving average of the past 5 trials around the target probability (±15%). An informative block lasted between 15 and 30 trials. Switching from an informative block to a non-informative block occurred before 30 trials if a successful learning criterion was matched (exponential moving average of the accuracy for the past 10 cue choices above 80%) after the 15 th trial in the block. Each non-informative block lasted 15 trials. Participants completed a maximum of 320 trials. Reward Structure At the end of each trial, participants received two rewards: one for the motion judgment and another for the cue choice. For correct motion responses, 10 points were assigned. For incorrect motion responses, 10 points were deducted. The reward for the cue choice depended on the confidence reported earlier in the trial. If participants did not respond within the time limit, a negative reward was assigned (-20 points), and the trial was terminated. The reward of the cue choice was dependent on the actual validity of the cue and the confidence rated by the participant on a particular trial. Reward values were estimated based on a proper scoring rule 49 , in which participants received 10 points if their confidence was high or 8 points if their confidence was low, while they lost 9 points if their confidence was high or 4 points if their confidence was low. Behavioral Modeling The latent prior strength of the optimal cue across trials was modeled using a variant of the Kalman Filter 50 , the Volatile Kalman Filter 12 (VKF). In brief, its generative model assumes that the two hidden Markov chains, in which the latent prior indicates the optimal cue for a given trial, x t , diffuses over time according to the hidden (inverse) volatility of the environment, z t , such that x t = N ( x t | x t −1 , z −1 ). In its turn, z t is given by its previous values and evolves dynamically according to a multiplicative diffusion noise, z t = z t −1 ϵ t ( λ ). ϵ t is defined as a rescaled beta distribution such that ϵ t ∈ [0, (1 − λ ) −1 ], and 0 < λ < 1. Therefore, the free parameter λ controls the diffusion of the volatility. On the first trial, the inverse mean of z t is given by a free parameter v 0 . The choice, y t , is defined as a softmax function of the difference between the latent prior for each cue with a decision exploration (free) parameter β : The outcome of a given trial, o t ∈ {0,1}, is given by o t = Bernoulli( σ ( x t )), where σ ( x t ) = 1/1 + e − x t . In our case, we assume that the cue validity of each trial provides evidence for a mutually exclusive state of whether blue is currently the best cue such that For the subsequent analysis, we used the prior of the chosen cue being the best, prior = x t ⋅ 𝕀(chosen cue = Blue) + (1 − x t ) ⋅ 𝕀(chosen cue = Orange), where 𝕀(⋅) equals 1 if the condition inside is true and 0 otherwise. The prior was then categorized into low and high using a subject-level median split. Model Fitting and Comparison To evaluate the performance of the VKF, we consider three alternative learning models: the Rescorla-Wagner 51 (RW), the Kalman Filter 50 (KF), and the Hierarchical Gaussian Filter 52 , 53 (HGF). While the KF and the HGF models account for volatilities in the environment, the RW model does not. A softmax function with a free temperature parameter, β, was used as a response model for all learning models as described for the VKF. The models were fit and compared ( Supplementary Figure 3 ) with a hierarchical Bayesian inference approach based on variational Bayes 54 , in which fitted parameters are assumed to be drawn from a normal distribution based on the group-level statistics and the model identity is considered a random effect. To estimate the subject-level posterior, a Laplace approximation was used. EEG Data Acquisition and Preprocessing Electrophysiological (EEG) data were continuously recorded with 257 scalp electrodes with a 1000 Hz sampling rate and an online reference at Cz (Electrical Geodesics, Inc.). The data were preprocessed using the FieldTrip 55 toolbox for Matlab by applying a bandpass Butterworth filter (0.3 to 40 Hz), demeaning, and re-referencing to the common average. Trials with muscle artifacts were visually identified and excluded. Eye and muscle artifacts were manually removed using Independent Component Analysis. The data were down-sampled to 500 Hz, epoched from -1000 to 1000 ms relative to the RDK onset, and z-scored using the trial-wise mean and standard deviation from the -400 to 0 ms interval relative to the cue presentation. Univariate Analysis Time-locked responses were calculated per participant and electrode by averaging across different experimental conditions relative to the onset of the motion stimulus or relative to response onset. Differences between event-related potentials (ERPs) were assessed using cluster-based permutation statistics as implemented in FieldTrip 55 . Multivariate Analysis Decoding of action (left or right button press) or sensory-like activity (left or right motion) was performed on 257 channels using a Linear Discriminant Analysis 13 , 56 using the MVPA-Light 57 toolbox for Matlab. In all decoding analyses, we performed a 10-fold cross-validation procedure balancing the selected feature (action or sensory), the prior strength (low or high), and the cue validity (valid or invalid) across the training samples, but not the testing samples, in each fold by upsampling. The true class probability (TCP) of testing samples was estimated based on the multivariate distribution for the training fold. To compare the evolution of decoded signals, we performed a time generalization analysis across all time points and trials, resulting in a N trials × N samples × N samples TCP matrix per participant. Then, this matrix was averaged across conditions, resulting in a TCP prior , validity matrix with N samples × N samples per participant. For each decoded feature, the common effect of having a high prior ( Fig. 3 c ) is defined by isolating the effect of having a high prior regardless of the validity of the cue. The tuning (sharpening if positive, dampening if negative) of the decoded signals was defined as the difference between the valid and invalid conditions within high prior trials ( Fig. 4 a ), The sensory sharpening and action dampening for each training and test time were then correlated with the behavioral modulation and speed-accuracy trade-offs. To capture the individual differences ( Fig. 1 c/d ), the participant’s values of the two correlated features were projected into their regression line, capturing their shared variance along a single dimension. This procedure was repeated for response-locked TCP prior , validity data. Statistical Analysis Unless stated otherwise, we employed paired t-tests. All electrophysiological recordings were assessed using cluster-based permutation tests as implemented in FieldTrip 55 that were based on paired t-tests. Clusters were determined across space and time (univariate analyses) or across time (multivariate analyses) by summing t-values within the thresholded cluster. The observed cluster was compared to a surrogate distribution (10.000 shuffles) where condition labels were randomly permuted between groups. Effect sizes were quantified using Cohen’s d. Conflict of interest The authors declare no competing interests. Author contributions Conceptualization: GI, RFH; Methodology: GI, RFH; Formal Analysis: GI; Investigation: GI; Resources: RFH; Writing – Original Draft, GI; Writing – Review & Editing: RFH; Supervision: RFH; Funding Acquisition: GI, RFH. Data and code availability All data and custom code used for analyses are available from the corresponding author upon request. Preprocessing of EEG data and cluster-based permutation tests were performed using Fieldtrip 55 toolbox for Matlab and decoding analysis where performed using the MVPA-Light 57 toolbox for Matlab. Acknowledgments This work was funded by the doctoral scholarship from the Baden-Württemberg Landesgraduiertenförderungsgesetz (GI), the International Max Planck Research School for The Mechanisms of Mental Function and Dysfunction (GI), the German Research Foundation (HE8329/2-1 to RFH), the Medical Faculty of the University of Tübingen (JRG Plus program, RFH), and the Jung Foundation for Research and Science (Ernst Jung Career Advancement Award in Medicine; RFH). The funders had no role in the study design, data collection, analysis, decision to publish, or preparation of the manuscript. References 1. ↵ Fritsche , M. et al. Temporal regularities shape perceptual decisions and striatal dopamine signals . Nat Commun 15 , 7093 ( 2024 ). OpenUrl CrossRef PubMed 2. ↵ Urai , A. E. , de Gee , J. W. , Tsetsos , K. & Donner , T. H . Choice history biases subsequent evidence accumulation . eLife 8 , e46331 ( 2019 ). OpenUrl CrossRef PubMed 3. ↵ Talluri , B. C. , Urai , A. E. , Tsetsos , K. , Usher , M. & Donner , T. H . Confirmation Bias through Selective Overweighting of Choice-Consistent Evidence . Current Biology 28 , 3128 – 3135.e8 ( 2018 ). OpenUrl CrossRef PubMed 4. ↵ Knill , D. C. & Richards , W . Perception as Bayesian Inference . ( Cambridge University Press , 1996 ). 5. ↵ Gold , J. I. & Stocker , A. A . Visual Decision-Making in an Uncertain and Dynamic World . Annual Review of Vision Science 3 , 227 – 250 ( 2017 ). OpenUrl CrossRef PubMed 6. ↵ de Lange , F. P. , Heilbron , M. & Kok , P . How Do Expectations Shape Perception? Trends in Cognitive Sciences 22 , 764 – 779 ( 2018 ). OpenUrl CrossRef PubMed 7. ↵ Blakemore , S. J. , Wolpert , D. M. & Frith , C. D . Central cancellation of self-produced tickle sensation . Nat Neurosci 1 , 635 – 640 ( 1998 ). OpenUrl CrossRef PubMed Web of Science 8. ↵ Press , C. , Kok , P. & Yon , D. The Perceptual Prediction Paradox . Trends in Cognitive Sciences 24 , 13 – 24 ( 2020 ). OpenUrl CrossRef PubMed 9. ↵ Kok , P. , Jehee , J. F. M. & de Lange , F. P . Less Is More: Expectation Sharpens Representations in the Primary Visual Cortex . Neuron 75 , 265 – 270 ( 2012 ). OpenUrl CrossRef PubMed Web of Science 10. ↵ Summerfield , C. & Koechlin , E . A Neural Representation of Prior Information during Perceptual Inference . Neuron 59 , 336 – 347 ( 2008 ). OpenUrl CrossRef PubMed Web of Science 11. ↵ Kilteni , K. & Ehrsson , H. H . Body ownership determines the attenuation of self-generated tactile sensations . Proceedings of the National Academy of Sciences 114 , 8426 – 8431 ( 2017 ). OpenUrl Abstract / FREE Full Text 12. ↵ Piray , P. & Daw , N. D . A simple model for learning in volatile environments . PLOS Computational Biology 16 , e1007963 ( 2020 ). OpenUrl CrossRef 13. ↵ Fisher , R. A . The Use of Multiple Measurements in Taxonomic Problems . Annals of Eugenics 7 , 179 – 188 ( 1936 ). OpenUrl CrossRef Web of Science 14. ↵ Kok , P. , Failing , M. F. & de Lange , F. P . Prior Expectations Evoke Stimulus Templates in the Primary Visual Cortex . Journal of Cognitive Neuroscience 26 , 1546 – 1554 ( 2014 ). OpenUrl CrossRef PubMed 15. ↵ Kok , P. , Mostert , P. & de Lange , F. P . Prior expectations induce prestimulus sensory templates . Proceedings of the National Academy of Sciences 114 , 10473 – 10478 ( 2017 ). OpenUrl Abstract / FREE Full Text 16. ↵ Geurts , L. S. , Cooke , J. R. H. , van Bergen , R. S. & Jehee , J. F. M . Subjective confidence reflects representation of Bayesian probability in cortex . Nat Hum Behav ( 2022 ) doi: 10.1038/s41562-021-01247-w . OpenUrl CrossRef PubMed 17. ↵ Skowron , A. et al. Neural variability compresses with increasing belief precision during Bayesian inference . 2024.01.11.575180 Preprint at doi: 10.1101/2024.01.11.575180 ( 2024 ). OpenUrl Abstract / FREE Full Text 18. ↵ van Bergen , R. S. & Jehee , J. F. M . Modeling correlated noise is necessary to decode uncertainty . NeuroImage 180 , 78 – 87 ( 2018 ). OpenUrl CrossRef PubMed 19. Lange , R. D. & Haefner , R. M . Task-induced neural covariability as a signature of approximate Bayesian learning and inference . PLOS Computational Biology 18 , e1009557 ( 2022 ). OpenUrl CrossRef PubMed 20. ↵ van den Brink, R. L., et al. Flexible sensory-motor mapping rules manifest in correlated variability of stimulus and action codes across the brain . Neuron 111 , 571 – 584.e9 ( 2023 ). OpenUrl CrossRef PubMed 21. ↵ Findling , C. et al. Brain-wide representations of prior information in mouse decision-making . 2023.07.04.547684 Preprint at doi: 10.1101/2023.07.04.547684 ( 2023 ). OpenUrl Abstract / FREE Full Text 22. ↵ de Lange , F. P. , Rahnev , D. A. , Donner , T. H. & Lau , H . Prestimulus Oscillatory Activity over Motor Cortex Reflects Perceptual Expectations . Journal of Neuroscience 33 , 1400 – 1410 ( 2013 ). OpenUrl Abstract / FREE Full Text 23. ↵ Kelly , S. P. , Corbett , E. A. & O’Connell , R. G . Neurocomputational mechanisms of prior-informed perceptual decision-making in humans . Nat Hum Behav 5 , 467 – 481 ( 2021 ). OpenUrl PubMed 24. ↵ Bogacz , R. , Brown , E. , Moehlis , J. , Holmes , P. & Cohen , J. D . The physics of optimal decision making: A formal analysis of models of performance in two-alternative forced-choice tasks . Psychological Review 113 , 700 – 765 ( 2006 ). OpenUrl CrossRef PubMed Web of Science 25. ↵ Rao , V. , DeAngelis , G. C. & Snyder , L. H . Neural Correlates of Prior Expectations of Motion in the Lateral Intraparietal and Middle Temporal Areas . J. Neurosci . 32 , 10063 – 10074 ( 2012 ). OpenUrl Abstract / FREE Full Text 26. ↵ Rollwage , M. et al. Confidence drives a neural confirmation bias . Nat Commun 11 , 2634 ( 2020 ). OpenUrl CrossRef PubMed 27. ↵ Park , H. et al. Confirmation Bias through Selective Use of Evidence in Human Cortex . 2024.06.21.600060 Preprint at doi: 10.1101/2024.06.21.600060 ( 2024 ). OpenUrl Abstract / FREE Full Text 28. ↵ Qiu , C. , Luu , L. & Stocker , A. A . Benefits of commitment in hierarchical inference . Psychological Review 127 , 622 – 639 ( 2020 ). OpenUrl CrossRef PubMed 29. ↵ Lange , R. D. , Chattoraj , A. , Beck , J. M. , Yates , J. L. & Haefner , R. M . A confirmation bias in perceptual decision-making due to hierarchical approximate inference . PLOS Computational Biology 17 , e1009517 ( 2021 ). OpenUrl CrossRef PubMed 30. ↵ Donner , T. H. , Siegel , M. , Fries , P. & Engel , A. K . Buildup of Choice-Predictive Activity in Human Motor Cortex during Perceptual Decision Making . Current Biology 19 , 1581 – 1585 ( 2009 ). OpenUrl CrossRef PubMed Web of Science 31. ↵ Wilming , N. , Murphy , P. R. , Meyniel , F. & Donner , T. H . Large-scale dynamics of perceptual decision information across human cortex . Nat Commun 11 , 5109 ( 2020 ). OpenUrl CrossRef PubMed 32. ↵ Roitman , J. D. & Shadlen , M. N . Response of Neurons in the Lateral Intraparietal Area during a Combined Visual Discrimination Reaction Time Task . J. Neurosci . 22 , 9475 – 9489 ( 2002 ). OpenUrl Abstract / FREE Full Text 33. ↵ Shushruth , S. , Mazurek , M. & Shadlen , M. N . Comparison of Decision-Related Signals in Sensory and Motor Preparatory Responses of Neurons in Area LIP . J. Neurosci . 38 , 6350 – 6365 ( 2018 ). OpenUrl Abstract / FREE Full Text 34. ↵ Gherman , S. et al. Intracranial electroencephalography reveals effector-independent evidence accumulation dynamics in multiple human brain regions . Nat Hum Behav 8 , 758 – 770 ( 2024 ). OpenUrl CrossRef PubMed 35. ↵ Lisi , M. , Morgan , M. J. & Solomon , J. A . Perceptual decisions and oculomotor responses rely on temporally distinct streams of evidence . Commun Biol 5 , 189 ( 2022 ). OpenUrl CrossRef PubMed 36. ↵ Khilkevich , A. et al. Brain-wide dynamics linking sensation to action during decision-making . Nature 634 , 890 – 900 ( 2024 ). OpenUrl CrossRef PubMed 37. ↵ Park , J. , Kim , S. , Kim , H. R. & Lee , J . Prior expectation enhances sensorimotor behavior by modulating population tuning and subspace activity in sensory cortex . Science Advances 9 , eadg4156 ( 2023 ). OpenUrl CrossRef PubMed 38. ↵ Ruesseler , M. , Weber , L. A. , Marshall , T. R. , O’Reilly , J. & Hunt , L. T . Quantifying decision-making in dynamic, continuously evolving environments . eLife 12 , e82823 ( 2023 ). OpenUrl CrossRef PubMed 39. ↵ Sandhaeger , F. , Omejc , N. , Pape , A.-A. & Siegel , M . Abstract perceptual choice signals during action-linked decisions in the human brain . PLOS Biology 21 , e3002324 ( 2023 ). OpenUrl CrossRef PubMed 40. ↵ Domenech , P. , Rheims , S. & Koechlin , E . Neural mechanisms resolving exploitation-exploration dilemmas in the medial prefrontal cortex . Science 369 , eabb0184 ( 2020 ). OpenUrl Abstract / FREE Full Text 41. ↵ Walsh , K. , McGovern , D. P. , Dully , J. , Kelly , S. P. & O’Connell , R. G . Prior probability cues bias sensory encoding with increasing task exposure . eLife 12 , RP91135 ( 2024 ). OpenUrl CrossRef PubMed 42. ↵ Feuerriegel , D. , Vogels , R. & Kovács , G . Evaluating the evidence for expectation suppression in the visual system . Neuroscience & Biobehavioral Reviews 126 , 368 – 381 ( 2021 ). OpenUrl CrossRef PubMed 43. ↵ Yon , D. et al. Stubborn Predictions in Primary Visual Cortex . Journal of Cognitive Neuroscience 35 , 1133 – 1143 ( 2023 ). OpenUrl CrossRef PubMed 44. ↵ Simoncelli , E. P. & Olshausen , B. A . Natural Image Statistics and Neural Representation . Annual Review of Neuroscience 24 , 1193 – 1216 ( 2001 ). OpenUrl CrossRef PubMed Web of Science 45. ↵ Jones , P. R . QuestPlus: A MATLAB Implementation of the QUEST+ adaptive Psychometric Method . Journal of Open Research Software 6 , 27 ( 2018 ). OpenUrl CrossRef 46. ↵ Watson , A. B . QUEST+: A general multidimensional Bayesian adaptive psychometric method . Journal of Vision 17 , 10 ( 2017 ). OpenUrl CrossRef PubMed 47. ↵ Brainard , D. H. The Psychophysics Toolbox . Spatial Vis 10 , 433 – 436 ( 1997 ). OpenUrl CrossRef Web of Science 48. ↵ Pelli , D. G . The VideoToolbox software for visual psychophysics: transforming numbers into movies . Spatial Vis 10 , 437 – 442 ( 1997 ). OpenUrl CrossRef Web of Science 49. ↵ Carvalho , A. , Dimitrov , S. & Larson , K . On proper scoring rules and cumulative prospect theory . EURO J Decis Process 6 , 343 – 376 ( 2018 ). OpenUrl CrossRef 50. ↵ Kalman , R. E . A New Approach to Linear Filtering and Prediction Problems . Journal of Basic Engineering 82 , 35 – 45 ( 1960 ). OpenUrl CrossRef 51. ↵ Black , A. H. Prokosy , W. F. Rescorla , R. A. & Wagner , A. R. A theory of Pavlovian conditioning : Variations in the effectiveness of reinforcement and non-reinforcement . in Classical conditioning II: Current research and theory (eds. Black , A. H. & Prokosy , W. F. ) vol. 2 64 – 69 ( Appleton-Century-Crofts, New Yorker , 1972 ). OpenUrl 52. ↵ Mathys , C. D. et al. Uncertainty in perception and the Hierarchical Gaussian Filter . Front. Hum. Neurosci . 8 , ( 2014 ). 53. ↵ Mathys , C. , Daunizeau , J. , Friston , K. & Stephan , K . A Bayesian Foundation for Individual Learning Under Uncertainty . Frontiers in Human Neuroscience 5 , ( 2011 ). 54. ↵ Piray , P. , Dezfouli , A. , Heskes , T. , Frank , M. J. & Daw , N. D . Hierarchical Bayesian inference for concurrent model fitting and comparison for group studies . PLOS Computational Biology 15 , e1007043 ( 2019 ). OpenUrl CrossRef PubMed 55. ↵ Oostenveld , R. , Fries , P. , Maris , E. & Schoffelen , J.-M . FieldTrip: Open Source Software for Advanced Analysis of MEG, EEG, and Invasive Electrophysiological Data . Computational Intelligence and Neuroscience 2011 , e156869 ( 2010 ). OpenUrl CrossRef 56. ↵ Zhao , S. , Zhang , B. , Yang , J. , Zhou , J. & Xu , Y. Linear discriminant analysis . Nat Rev Methods Primers 4 , 1 – 16 ( 2024 ). OpenUrl CrossRef 57. ↵ Treder , M. S . MVPA-Light: A Classification and Regression Toolbox for Multi-Dimensional Data . Front. Neurosci . 14 , ( 2020 ). View the discussion thread. Back to top Previous Next Posted March 29, 2025. Download PDF Supplementary Material Email Thank you for your interest in spreading the word about bioRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following Sensory and action neural tuning explains how priors guide human visual decisions Message Subject (Your Name) has forwarded a page to you from bioRxiv Message Body (Your Name) thought you would like to see this page from the bioRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share Sensory and action neural tuning explains how priors guide human visual decisions Gabriela Iwama , Randolph F. Helfrich bioRxiv 2025.03.26.645167; doi: https://doi.org/10.1101/2025.03.26.645167 Share This Article: Copy Citation Tools Sensory and action neural tuning explains how priors guide human visual decisions Gabriela Iwama , Randolph F. Helfrich bioRxiv 2025.03.26.645167; doi: https://doi.org/10.1101/2025.03.26.645167 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Area Neuroscience Subject Areas All Articles Animal Behavior and Cognition (7635) Biochemistry (17690) Bioengineering (13892) Bioinformatics (41936) Biophysics (21451) Cancer Biology (18588) Cell Biology (25499) Clinical Trials (138) Developmental Biology (13378) Ecology (19899) Epidemiology (2067) Evolutionary Biology (24320) Genetics (15609) Genomics (22506) Immunology (17736) Microbiology (40394) Molecular Biology (17181) Neuroscience (88603) Paleontology (666) Pathology (2832) Pharmacology and Toxicology (4824) Physiology (7641) Plant Biology (15152) Scientific Communication and Education (2045) Synthetic Biology (4294) Systems Biology (9825) Zoology (2271)