{"paper_id":"2a0cdc54-0f05-40c2-8ad9-e4eb139f7d6b","body_text":"Enhanced Rare Working Memory Representations \n \n \nEnhanced Working Memory Representations for Rare Events \n \nCarlos Daniel Carrasco, Aaron Matthew Simmons, John E. Kiat, & Steven J. Luck \nCenter for Mind & Brain, University of California, Davis \nAuthor Notes:  \nThe authors declare no competing financial interests.  \nContact Information:  \nCarlos Daniel Carrasco: cdcarrasco@ucdavis.edu \nAaron Matthew Simmons: amsimmons@ucdavis.edu \nJohn E. Kiat: jekiat@ucdavis.edu \nSteven, J. Luck: sjluck@ucdavis.edu \n \n \n \n \n \n \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted March 20, 2024. ; https://doi.org/10.1101/2024.03.20.585952doi: bioRxiv preprint \n\nEnhanced Rare Working Memory Representations \n \nAbstract    \nRare events (oddballs) produce a variety of enhanced physiological responses relative to frequent \nevents (standards), including the P3b component of the event-related potential (ERP) waveform. \nPrevious research has suggested that the P3b component is related to working memory, which \nimplies that working memory representations will be enhanced for rare stimuli. To test this \nhypothesis, we devised a modified oddball paradigm in which the target was a disk presented at \none of 16 different locations, which were divided into a rare set and a frequent set. Participants \nmade a binary response on each trial to report whether the target appeared in the rare set or the \nfrequent set. As expected, the P3b was much larger for stimuli appearing at a location within the \nrare set. We also included occasional probe trials in which the subject reported the exact location \nof the target. We found that these reports were more accurate for locations within the rare set \nthan for locations within the frequent set. Moreover, the mean accuracy of these reports was \ncorrelated with the mean amplitude of the P3b. We also applied multivariate pattern analysis to \nthe ERP data to “decode” the remembered location of the target. Decoding accuracy was greater \nfor locations within the rare set than for locations within the frequent set. These behavioral and \nelectrophysiological results demonstrate that although both frequent and rare events are stored in \nworking memory, the representations are enhanced for rare events.   \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted March 20, 2024. ; https://doi.org/10.1101/2024.03.20.585952doi: bioRxiv preprint \n\nEnhanced Rare Working Memory Representations \n \nSignificance Statement  \nFor many decades, researchers have observed that rare events elicit a broad range of \nphysiological responses, and there has been much speculation about the functional significance \nof these responses. One such response is the P3b component, which is a large voltage deflection \nin scalp EEG recordings. Over 40 years ago, the P3b was hypothesized to reflect “context \nupdating” (now often called “working memory updating”). However, there has been no direct \nevidence that working memory is actually enhanced for rare, P3b-eliciting events. In the present \nstudy, we found that both behavioral and electrophysiological measures of working memory \nwere enhanced for rare events. This is potentially related to the release of norepinephrine across \nthe cortex. \n  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted March 20, 2024. ; https://doi.org/10.1101/2024.03.20.585952doi: bioRxiv preprint \n\nEnhanced Rare Working Memory Representations \n \nIntroduction  \nRare task-relevant events (often called oddballs) generate a variety of physiological \nresponses, including increased firing of noradrenergic neurons in the locus coeruleus, phasic \npupil dilation, increased blood-oxygen-level dependent (BOLD) activity in a variety of cortical \nareas, and a large P3b event-related potential (ERP) component (Bledowski et al., 2004; Clark et \nal., 2000; Johnson, 1988; Krebs et al., 2018; Linden et al., 1999; Murphy et al., 2011; Polich, \n1986; Soltani & Knight, 2000). There has been considerable speculation about the functional \nsignificance of these physiological changes (Kim, 2014; Linden et al., 1999; Nieuwenhuis et al., \n2011; Paller et al., 1992). A common hypothesis is that the P3b activity elicited by oddballs is \nrelated to working memory encoding, although the details vary across theories (Donchin & \nColes, 1988; Kok, 2001; Polich, 2007, 2012). Moreover, oddballs elicit phasic increases in \nattention (Aston-Jones & Cohen, 2005; Katayama & Polich, 1998; Kim, 2014; Murphy et al., \n2011), which might also enhance working memory. However, we know of no direct evidence \nthat working memory is actually enhanced for relatively rare events compared to relatively \nfrequent events. The goal of the present study was to test this hypothesis, using both behavioral \nand electrophysiological measures. \nTypical oddball paradigms do not provide a sensitive assessment of working memory. \nFor example, a typical paradigm would involve presenting a sequence of stimuli in which 90% \nare the letter X and 10% are the letter O, and the task would be to press one button for Xs and \nanother button for Os. The responses are made immediately, so it is not necessary to store the Xs \nand Os in working memory. Moreover, the Xs and Os are so easily discriminable that memory \nperformance would likely be at ceiling if tested after a brief delay. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted March 20, 2024. ; https://doi.org/10.1101/2024.03.20.585952doi: bioRxiv preprint \n\nEnhanced Rare Working Memory Representations \n \nWe therefore developed the modified oddball paradigm shown in Figure 1. On each trial, \na target disc appeared briefly near one of eight locations around a circle, and the main task was \nto press one button if the disc appeared near one of the cardinal axes (up, down, left, or right) and \na different button if it appeared near one of the diagonal axes (upper left, upper right, lower left, \nor lower right). One of these two categories was rare (12.5%), and the other was frequent \n(87.5%). This was much like a traditional oddball task, in which participants make an immediate \nresponse to indicate whether the stimulus belonged to the rare category or the frequent category. \nWe assumed that stimuli belonging to the rare category (the oddballs) would elicit a larger P3b \ncomponent than stimuli belonging to the frequent category (the standards). To provide a \nsensitive measure of the working memory representation of the disk, we also included occasional \nprobe trials, on which participants were asked to click on the exact location of the disc from that \ntrial after a brief delay. If working memory is enhanced for rare stimuli, then the target \nlocalization response on probe trials should be more accurate following an oddball than \nfollowing a standard. Moreover, if the P3b component elicited by oddballs is associated with \nworking memory updating, then participants with greater P3b amplitudes for the oddballs should \nexhibit more accurate memory on the probe trials than individuals with smaller P3b amplitudes. \nIn addition, we applied multivariate pattern analysis (MVPA) to the ERP data to decode \nthe remembered location of the disc in the delay period following each stimulus. This provided a \nmeans of monitoring the working memory representation of the stimuli during the delay period \nfollowing each target, independently of the decision and response processes that are involved in \nthe behavioral responses. Specifically, we decoded which of the four locations within a category \nwas presented (e.g., up, down, left, or right for the cardinal category). We predicted that the \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted March 20, 2024. ; https://doi.org/10.1101/2024.03.20.585952doi: bioRxiv preprint \n\nEnhanced Rare Working Memory Representations \n \nwithin-category decoding accuracy would be greater for a given category when that category was \nrare than when it was frequent. \nFigure 1. (A) Overview of the oddball task. On every trial, participants made a speeded keypress \nfollowing the presentation of a target disc to indicate whether it appeared near a cardinal axis or near a \ndiagonal axis (as shown in B). For a given trial block, one axis was rare and the other was frequent. On \n12.5% of trials (probe trials), the fixation point turned red at the offset of the intertrial interval, which \nsignaled participants that they should use the mouse to click on the remembered location of the target disk \nfrom that trial. (B) Possible locations of the target disc, which could appear 1° counterclockwise or 1° \nclockwise from a cardinal or diagonal axis. Note that the distance of the targets from the axes is slightly \nexaggerated in this image. \nMethods  \nParticipants \nTwenty-two human participants from the UC Davis community completed the study (14 \nwomen, 7 men, 1 unreported gender; mean age = 20, SD = 1.69). All participants were \nneurotypical and had normal or corrected-to-normal vision with no history of neurological \nconditions. Twenty participants were right-handed, and two were left-handed. We chose to \nProbe: 12.5% occurrence\n1200-1400 ms\n200 1200-1400 1200-1400200 200\n...\nTime (ms)\nOpenEnded ResponseWindow\nA) B)\n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted March 20, 2024. ; https://doi.org/10.1101/2024.03.20.585952doi: bioRxiv preprint \n\nEnhanced Rare Working Memory Representations \n \ncollect a somewhat higher N than is typical of multivariate pattern analysis EEG studies (e.g., N \n= 16 in Bae & Luck, 2019; Bae, 2021; Bae & Luck, 2018) because of the relatively small \nnumber of oddball trials in the present study. Monetary compensation was provided at a rate of \n$15/hr. All participants provided informed consent, and the study was approved by the \nUniversity of California, Davis Institutional Review Board.  \nStimuli and Task  \nFigure 1 illustrates the stimuli and task. The experiment presentation script can be \ndownloaded at doi.org/10.17605/OSF.IO/HV7JU. \nStimuli were presented using Psychopy (Peirce et al., 2019) on an LCD monitor (Dell \nU2412M) with a gray background (27.7 cd/m2) at a viewing distance of 100 cm. A 0.05° black \nfixation dot was continuously visible in the center of the monitor, surrounded by a black circle \nwith a radius of 2.17°.  \nOn each regular trial, a black target disc (0.2°) appeared for 200 ms, followed by an \ninterstimulus interval of 1200-1400 ms (rectangular distribution). The target appeared on the \nblack circle, near one of the cardinal axes (0°, 90°, 180°, 270°) or near one of the diagonal axes \n(45°, 135°, 225°, 315°), centered either -1 or +1 degrees from one of these axes (e.g., 44°, 46°, \n89°, 91°). Thus, there were 16 possible target locations, 8 near the cardinal axes and 8 near the \ndiagonal axes. Placing the target ±1° from an axis was designed to require participants to \nremember the precise location of the target rather than relying on simple categories such as “top” \nor “lower left”. In addition, this made it possible to examine biases away from the cardinal axes \n(Bae, 2022). \n For half of the trial blocks, the target appeared near a cardinal axis on 12.5% of trials \n(oddballs) and appeared near a diagonal axis on the remaining 87.5% of trials (standards). This \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted March 20, 2024. ; https://doi.org/10.1101/2024.03.20.585952doi: bioRxiv preprint \n\nEnhanced Rare Working Memory Representations \n \nwas reversed for the other trial blocks. Each of the 8 locations within the cardinal or diagonal \ncategory occurred with equal probability. Participants were instructed to make a speeded \nresponse on a computer keyboard to indicate whether the immediately preceding target was near \na cardinal axis or near a diagonal axis. Half the participants pressed the left arrow key for \ncardinal and the right arrow key for diagonal; this was reversed for the other half.  \nAlthough most oddball studies use only one rare stimulus (e.g., a high-pitched tone) and \none frequent stimulus (e.g., a low-pitched tone), the P3b component is sensitive to the probability \nof the task-defined category rather than the probability of the physical stimulus (Luck, 2014; \nMecklinger & Ullsperger, 1993). Oddball studies have used abstract categories such as the ones \nused here for decades (e.g., Kutas et al., 1977). \nEach block contained 256 trials (32 oddballs and 224 standards). The blocks alternated \nbetween cardinal-oddball/diagonal-standard and diagonal-oddball/cardinal-standard, with the \nstarting condition counterbalanced across participants. Each participant received 256 oddballs \nand 1792 standards, with each of the 16 locations occurring equally often within each of these \ncategories (2048 trials total). Thus, there were 112 trials for a given location when that location \nwas in the standard category and 16 trials for a given location when that location was in the \noddball category. \nWorking memory for the exact target location was probed following a random 12.5% of \ntargets. When a target was probed, the fixation dot changed from black to red at the end of the \ninterstimulus interval (i.e., after the participant indicated whether the target had been near a \ncardinal or diagonal axis). Once it turned red, the participant could use the mouse to move the \nred dot. They were instructed to move the red dot to the remembered location of the target and \nthen click the mouse button (with no time pressure). The red dot then disappeared and was \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted March 20, 2024. ; https://doi.org/10.1101/2024.03.20.585952doi: bioRxiv preprint \n\nEnhanced Rare Working Memory Representations \n \nreplaced by the black fixation dot in the center of the monitor. The stream of targets then \nresumed after a delay of 1200-1400 ms. For each participant, oddballs were probed on 32 trials \nand standards were probed on 224 trials.  \nNote that participants could not predict whether a given trial would be a probe trial. Probe \ntrials were just like any other trials until the end of the intertrial interval following a given target. \nThus, the working memory representations and neural activity during the intertrial interval \nfollowing a target could not be systematically different on probe and non-probe trials. \nWe probed on only a small subset of trials for two reasons. First, we wanted the task to be \nmore like a traditional oddball task, in which exact memories are not probed. Second, probe \nresponses took considerable time, and we could obtain more targets in a session of a reasonable \nduration if we probed infrequently. A very large overall number of trials was needed to obtain a \nsufficient number of trials per location for the EEG decoding (which used the data from all trials, \nnot just probe trials). Far fewer trials were needed to obtain robust measures of behavioral \naccuracy. \nEEG Recording and Preprocessing  \nThe continuous EEG was recorded using a Brain Products actiCHamp recording system. \nWe recorded EEG signals from 27 standard 10/20 sites: FP1, Fz, F3, F7, Cz, C3, Pz, P3, P5, P7, \nP9, PO7, PO3, O1, POz, Oz, FP2, F4, F8, C4, P4, P6, P8, P10, PO4, PO8, O2. We also recorded \nsignals from horizontal electrooculogram (HEOG) electrodes lateral to the left and right external \ncanthi, from a vertical electrooculogram (VEOG) electrode located under the right eye, and from \nelectrodes over the left and right mastoids. Single-ended voltages were recorded relative to a \nground electrode located at AFz. All electrode impedances were kept < 50 KΩ. The signals were \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted March 20, 2024. ; https://doi.org/10.1101/2024.03.20.585952doi: bioRxiv preprint \n\nEnhanced Rare Working Memory Representations \n \nfiltered with a cascaded integrator-comb antialiasing filter (half-power cutoff at 130 Hz) and \ndigitized at 500 Hz.  \nAll preprocessing steps were conducted in MATLAB using the EEGLAB and ERPLAB \ntoolboxes (Delorme & Makeig, 2004; Lopez-Calderon & Luck, 2014), following the standard \npipeline described by Luck (2023).We began by shifting the event codes to account for the \nmonitor delay (56 ms, which was measured using a photodiode). The signals were then \nresampled at 250 Hz. The DC offset was removed, and the signals were high-pass filtered \n(noncausal Butterworth impulse response function, half-amplitude cutoffs at 0.1 Hz, 12 dB/oct \nroll-off). Time segments between the trial blocks were deleted, and the EEG data were \nreferenced to the P9 electrode site. A bipolar VEOG channel was created by subtracting the FP2 \nelectrode from the VEOG electrode, and a bipolar HEOG channel was created by subtracting \nHEOG-right from HEOG-left. \nIndependent component analysis (ICA) was then performed to correct for blinks and eye \nmovements (excluding the bipolar channels). The data used for the ICA decomposition were \nfiltered more aggressively (noncausal Butterworth impulse response function, half-amplitude \ncutoffs at 1 – 30 Hz, 48 dB/oct roll-off) and resampled at 100 Hz. The ICA weights were then \ntransferred back to the original data, and independent components corresponding to blinks and \nartifacts were removed from the data; typically, 1-2 components were removed per participant. \nWe used consistency between the shape, timing and spatial location of the component compared \nto the bipolar HEOG and VEOG signals to determine which components were artifacts. \nThe ICA-corrected data were then re-referenced to the average of the left and right \nmastoid electrodes. The data were then segmented from -500 to 1496 ms relative to stimulus \nonset and baseline-corrected to the mean voltage from -500 to 0 ms. Finally, epochs were \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted March 20, 2024. ; https://doi.org/10.1101/2024.03.20.585952doi: bioRxiv preprint \n\nEnhanced Rare Working Memory Representations \n \nmarked for rejection using standard ERPLAB routines if they contained large voltage deflections \nin any channel (using a simple voltage threshold) or an eyeblink that would have prevented \nperception of the target (using a moving window peak-to-peak function from 0 to 200 ms in the \nbipolar VEOG channel). This led to an average rejection of 13.7% of trials across participants \n(SE = 6.4%). We always exclude participants for whom more than 25% of trials were rejected; \nno participants exceeded this threshold in the present study. \nBehavioral Analyses \nFor the oddball categorization task, we computed the proportion correct and the mean \nresponse time (RT) for each participant, separately for the oddball and standard categories.  \nFor the probe task, the stimuli and responses were coded in terms of their angular \nposition around the circle of possible target locations. We excluded trials on which the reported \nlocation was > 40° away from the true location (0.89% ± 0.31% of trials), because such large \nerrors presumably reflect lapses of attention. On the remaining trials, we computed the response \nerror, defined as the angular distance between the true target location and the reported location. \nIn the primary analyses, we took the absolute value of the response error on each trial and \naveraged across trials for a given participant, separately for the oddball and standard categories. \nWorking memory representations tend to be biased away from the cardinal axes (Bae, \n2022) and we examined these biases in a separate analysis of the probe data. Following prior \nresearch, we expected that locations that were 1° clockwise from a cardinal axis would be \nreported as being more than 1° clockwise, and that locations that were 1° counterclockwise from \na cardinal axis would be reported as being more than 1° counterclockwise. If working memory \nrepresentations are enhanced for oddballs compared to standards, then these biases should be \nreduced for the oddballs. To analyze the biases, we took the response error on each trial and gave \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted March 20, 2024. ; https://doi.org/10.1101/2024.03.20.585952doi: bioRxiv preprint \n\nEnhanced Rare Working Memory Representations \n \nit a negative sign if it was farther away from the nearest cardinal axis than the true stimulus \nlocation (e.g., given a true location of 89°, which was near the cardinal axis at 90°, a reported \nlocation at 87° was coded as a response error of -2°). Similarly, we gave a response error a \npositive sign if it was biased in the opposite direction (e.g., given a true location of 89°, a \nreported location at 91° was coded as a response error of +2°). These values were then averaged \nacross trials for a given participant, separately for oddball and standard trials. A consistent \nrepulsion away from the nearest axis would lead to a negative average value. Little or no bias of \nthis sort would be expected for locations near the diagonal axes for either oddballs or standards, \nso trials with stimuli near the diagonal axes were excluded from this analysis. \nP3b Scoring  \nWe measured P3b amplitude at an a priori electrode size (Pz) and an a priori time \nwindow which we defined as defined as ±150 ms around the P3b peak of the grand averaged \nERP. This peak was at 512 ms so our measurement window was from 362 to 662 ms. The P3b \namplitude was scored as the mean voltage during this time window at Pz, separately for oddballs \nand standards. \nDecoding Analysis \nThe decoding analysis collapsed across the two locations that were ±1° from a given axis \n(e.g., 44° and 46°), which were too similar to be reliably differentiated by the decoding process. \nThis gave us four cardinal locations (0°, 90°, 180°, 270°) and four diagonal locations (45°, 135°, \n225°, 315°). We performed the decoding separately for these two categories, separately, when a \ncategory was the oddball and when it was the standard. For example, we decoded which of the \nfour cardinal locations was present on the cardinal trials when the cardinal category was the \noddball, and we separately decoded which of the four cardinal locations was present on the \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted March 20, 2024. ; https://doi.org/10.1101/2024.03.20.585952doi: bioRxiv preprint \n\nEnhanced Rare Working Memory Representations \n \ncardinal trials when the cardinal category was the standard. This allowed us to compare the \ndecoding accuracy for the same stimulus locations when those locations were oddballs and when \nthey were standards. Decoding was performed separately at each time point (every 4 ms from -\n500 to +1496 ms). \nWe followed the location decoding procedure developed by Bae and Luck (2018), as \nimplemented in ERPLAB Toolbox (Lopez-Calderon & Luck, 2014) using MATLAB’s fitcecoc() \nfunction. The first step was to apply a 6 Hz lowpass filter (48 dB/octave roll-off), which \nminimized contamination from alpha-band EEG oscillations. This filter reduced the temporal \nresolution of the analysis, but that was not a major problem given that we were examining long-\nduration working memory effects. The ocular channels were left out of the analysis, leaving 27 \nscalp channels. \nDecoding was performed on averaged ERP waveforms using support vector machines \n(SVM) with error-correcting output codes (Dietterich & Bakiri, 1995) and 3-fold cross-\nvalidation. We conducted four separate decoding runs for each participant: a) cardinal oddball; b) \ncardinal standard; c) diagonal oddball; and d) diagonal standard. For each run, there were four \npossible stimulus locations (the four different cardinal locations or the four different diagonal \nlocations). The decoder attempted to determine which of these four locations was presented on \nthe basis of the pattern of voltage across electrode sites at a given time point. \nBecause decoding accuracy tends to increase when more trials are available, we \nrandomly subsampled a subset of trials from the standards to equate the number of trials for \noddballs and standards. This yielded an average across participants of 19.5 trials (SD = 4.5) at \neach of the four locations for each decoding run. The available trials were randomly divided into \nthree subsets, and a separate averaged ERP waveform was created from each of these three \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted March 20, 2024. ; https://doi.org/10.1101/2024.03.20.585952doi: bioRxiv preprint \n\nEnhanced Rare Working Memory Representations \n \nsubsets. This gave us a matrix of 4 classes (4 locations) ´ 3 averages ´ 27 electrode sites ´ 500 \ntime points for each participant for each decoding run. Decoding was performed on the matrix of \nclasses ´ averages ´ electrode sites, separately at each time point for each participant for a given \nrun.  \nWe performed a 3-fold cross-validation by training the decoder on two of the three \naverages for each location and then having the decoder predict the location on the basis of the \nremaining average for each location. We then repeated the process two more times with new \ndecoders, changing which averages were used for training and which were used for testing. This \nprocess was then iterated 100 times, with different random subsets of trials used to create the \naveraged ERPs for each iteration. Completely new decoders were trained for every fold, \niteration, time point, and participant. \nFor each fold, the decoder consisted of four separate SVMs, each of which was trained to \ndistinguish between one location and the other three locations on the basis of the pattern of \nvoltage across electrode sites. To test the decoder, the vector of voltages across electrode sites \nfor a given location from the test average was passed to all four SVMs. The decoder predicted \nthe location for that test average using MATLAB’s predict() function, which minimized the \naverage binary loss over the four SVMs. Decoding accuracy was computed as the proportion of \ncorrect predictions by the decoder across the three folds and 100 iterations. Because there were \nfour locations, chance decoding accuracy was 25%.  \nWe then averaged decoding accuracy across the cardinal and diagonal decoding runs, \nseparately for oddballs and standards. This gave us a decoding accuracy for each participant at \neach time point for oddballs and for standards. To maximize statistical power, decoding accuracy \nwas then averaged across an a priori time window that began at the start of the P3b measurement \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted March 20, 2024. ; https://doi.org/10.1101/2024.03.20.585952doi: bioRxiv preprint \n\nEnhanced Rare Working Memory Representations \n \nwindow (362 ms) and extended until the end of the epoch (1496 ms). For each participant, this \ngave us one decoding accuracy value for oddballs and another for standards. \nResults  \nBehavioral results  \nOddball task. We first examined the speed and accuracy for the buttonpress task, in \nwhich participants indicated whether a given target was an oddball or a standard. As is typical in \noddball tasks, mean accuracy was lower for oddballs (80.5% ± 2.32%) than for standards (95.7% \n± 0.512%), which was significant in a paired t test (t(21) = 6.85, p < 0.0001, dz = 1.46). In \naddition, mean RTs were slower for oddballs (470.66 ± 19.48 ms) than for standards (346.04 ± \n18.75 ms), which was also a significant difference (t(21) = 14.59, p < 0.0001, dz = 3.11). Note \nthat all statistical tests reported in this paper were two-tailed and used an alpha of .05. Effect \nsizes are quantified as dz, which is the standard effect size metric corresponding to paired and \none-sample t tests (Cohen J, 1988; Lakens, 2013). When means are given, we also provide the \nstandard error of the mean (SEM). \n \n \n \nFigure 2. Probe trial results. A) Mean absolute error in the report of the exact location of the target disc. \nError bars show ±1 SEM. B) Histograms of the single-trial bias values for target discs near a cardinal \naxis. For oddballs, the distribution of errors was fairly symmetrical around zero (minimal bias). For \nstandards, the distribution is shifted toward negative values (repulsion away from the axis).  \n \nProbes:\np=0 . 0 0 3\nOddballs Standards\n0\n1\n2\n3\n4\n5Mean Absolute Error\nA) B) Oddballs Standards\n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted March 20, 2024. ; https://doi.org/10.1101/2024.03.20.585952doi: bioRxiv preprint \n\nEnhanced Rare Working Memory Representations \n \nProbe task. We next examined accuracy on the probe trials, for which participants \nclicked on the remembered location of the target. Figure 2a shows that the mean absolute \nresponse error for the probe task was smaller for oddballs (2.99 ± 0.155°) than for standards \n(3.36 ± 0.156°), which was a significant difference in a paired t test (t(21) = 3.30, p = 0.003, dz = \n0.704). This result indicates that working memory representations were more accurate for rare \nstimuli than for frequent stimuli. \nWe also asked whether working memory representations of rare stimuli are less subject to \nsystematic biases. In particular, we took advantage of the fact that stimuli presented near a \ncardinal axis are typically remembered as being shifted away from that axis (Bae, 2022; Pratte et \nal., 2017; Wei & Stocker, 2015). We asked whether this bias would be reduced for oddballs \nrelative to standards. Figure 2b shows histograms of the single-trial bias values, with negative \nvalues indicating a bias away from the nearby cardinal axis and positive values indicating a bias \ntoward the axis. On standard trials, the distribution of values was shifted toward the negative side \nof zero, indicating the typical finding of repulsion away from the axis. We computed an average \nbias score for each participant, and we found that this bias score was significantly different from \nzero for the standards in a one-sample t test (t(21) = 14.59 , p = 0.002, dz = 0.739). On oddball \ntrials, the distribution of bias values was more symmetrical around zero, and the average bias \nscore was not significantly different from zero (t(21) = 1.08 , p = 0.292, dz = 0.231). The key \nfinding was that the average bias score was significantly more negative for standards than for \noddballs in a paired t test (t(21) = 3.17 , p = 0.005, dz = 0.676). This indicates that working \nmemory representations were less biased for rare stimuli in addition to being more accurate. \nP3b amplitude \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted March 20, 2024. ; https://doi.org/10.1101/2024.03.20.585952doi: bioRxiv preprint \n\nEnhanced Rare Working Memory Representations \n \n \n \n \nFigure 3. A) Grand average ERP waveforms for oddballs and standards at representative electrode sites. \nThe P3b component peaked shortly after 500 ms and was largest at central and parietal electrode sites. B) \nMean amplitude at site Pz measured across a 300-ms time window centered around the peak (362-662 \nms). Error bars show ±1 SEM. C) Grand average scalp topography for the difference in amplitude \nbetween oddballs and standards in the P3b time window. D) Scatterplots showing the correlation between \nP3b amplitude (measured from 362-662 ms at the Pz electrode site) and behavioral accuracy on probe \ntrials (mean absolute error). Each dot shows an individual participant, and the line is the best-fit \nregression line.  \nFigure 3a shows the grand average ERP waveforms for oddballs and standards at five \nrepresentative electrode sites. As is typical, a large P3b was visible for oddballs, peaking at 512 \nms at the Pz electrode site. As shown in Figure 3b, the mean voltage during the measurement \nwindow (362-662 ms) at Pz was significantly larger for oddballs than for standards in a paired t \ntest (t(21) = 8.10, p < 0.0001, dz = 1.43). The scalp distribution of the oddball-minus-standard \ndifference wave during this window showed the typical midline centroparietal maximum (Figure \n3c). Thus, although the oddball paradigm used in the present study was somewhat unusual, it \nyielded the typical pattern of a larger P3b for oddballs than for standards. \nWe also asked whether individuals with larger P3b amplitudes had more accurate \nworking memory representations. Specifically, we examined the correlation between P3b \nA) Oddballs\nStandards\n-500 0 500 1000 1500\nTime (ms)\n-5\n0\n5\n10\n15µV\nAvg Fp1/Fp2\n-500 0 500 1000 1500\nTime (ms)\n-5\n0\n5\n10\n15\nFz\n-500 0 500 1000 1500\nTime (ms)\n-5\n0\n5\n10\n15\nCz\n-500 0 500 1000 1500\nTime (ms)\n-5\n0\n5\n10\n15\nPz\n-500 0 500 1000 1500\nTime (ms)\n-5\n0\n5\n10\n15\nOz\nOddballs Standards\n0\n2\n4\n6\n8\n10\n12µV\nPz p<0 . 0 0 0 1B) C) D)\nµV\nOddballs - Standards\n0 2 4 6\nMean Absolute Error\n0\n5\n10\n15\n20\n25Mean Amplitude (µV)\nOddballs\nr = 0.457, p = 0.0327\n0 2 4 6\nMean Absolute Error\n0\n5\n10\n15\n20\n25 Standards\nr = 0.151, p = 0.5016\n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted March 20, 2024. ; https://doi.org/10.1101/2024.03.20.585952doi: bioRxiv preprint \n\nEnhanced Rare Working Memory Representations \n \namplitude (mean voltage at Pz during the measurement window on all oddball or standard trials) \nand the absolute response error (on oddball or standard probe trials). The scatterplots are shown \nin Figure 3d. A statistically significant correlation was observed for oddballs (r = 0.457, p = \n0.033), with a smaller error for participants with a larger P3b amplitude. A correlation in the \nsame direction was observed for standards, but it was weak and not significant (r = 0.151, p = \n0.502). \nTogether, these results show that oddball stimuli produce both a larger P3b and more \naccurate working memory representations, and that these effects are correlated with each other. \nNote, however, that this does not demonstrate a causal relationship between the P3b and the \nworking memory enhancement. It is entirely plausible that both effects are results of a common \nunderlying factor (e.g., an increase in attention triggered by the oddballs, which separately \nimpacts P3b amplitude and working memory encoding). \nDecoding Results  \nOur final analyses were designed to determine whether we could see evidence of \nenhanced working memory for oddballs in the brain activity measured between the onset of the \nP3b wave and the end of the trial. Toward that end, we attempted to decode the location of the \nstimulus from the ERP activity at each moment in time across the recording epoch. We collapsed \nacross the two locations that were near a given axis (e.g., 44° and 46°), giving us four cardinal \nlocations and four diagonal locations. We then decoded which of the four cardinal locations was \npresented when the cardinals were standards, which of the four cardinal locations was presented \nwhen the cardinals were oddballs, which of the four diagonal locations was presented when the \ndiagonals were standards, and which of the four diagonal locations was presented when the \ndiagonal were oddballs. We then collapsed across the diagonal and cardinal categories to obtain \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted March 20, 2024. ; https://doi.org/10.1101/2024.03.20.585952doi: bioRxiv preprint \n\nEnhanced Rare Working Memory Representations \n \noverall decoding accuracy for oddballs and for standards. Because there were four locations \nwithin a category, chance decoding accuracy was 25%. \n \nFigure 4. A) Timepoint-by-timepoint decoding of target location on oddball and standard trials. The dark \nlines show mean decoding accuracy across participants, and the shading shows ±1 SEM. B) Mean \ndecoding accuracy between the start of the P3b measurement window (362 ms) and the end of the epoch \n(1500 ms). Error bars show ±1 SEM. \nFigure 4a shows decoding accuracy at each individual time point. Location decoding \naccuracy was slightly greater for oddballs than for standards for much of the epoch, but most \nnoticeably from approximately 800-1400 ms. To maximize statistical power, we averaged the \ndecoding accuracy across an a priori time window that began at the start of the P3b measurement \nwindow (362 ms) and extended through the end of the epoch (1496 ms). As shown in Figure 4b, \nmean decoding accuracy was greater for oddballs than for standards, which was a significant \ndifference in a paired t test (t(21) = 2.10, p = 0.048, dz = 0.425). Thus, the representation of the \ntarget was modestly but significantly enhanced for oddballs relative to standards during the \nperiod of time following stimulus offset. \nDiscussion  \n-500 0 500 1000 1500\nTime (ms)\n0.2\n0.3\n0.4\n0.5Decoding Accuracy\nOddballs\nStandards\n0.2\n0.3\n0.4\n0.5\nOddballs\nStandards\np = 0.048\n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted March 20, 2024. ; https://doi.org/10.1101/2024.03.20.585952doi: bioRxiv preprint \n\nEnhanced Rare Working Memory Representations \n \nThe current study sought to provide direct evidence that working memory representations \nare enhanced for relatively rare task-relevant events compared to relatively frequent task-relevant \nevents. Traditional oddball paradigms do not provide a sensitive assessment of working memory \nrepresentations, but we were able to assess the accuracy of target representations behaviorally by \nincluding probe trials. We also assessed the neural representation of the target by decoding the \ntarget’s location on the basis of the EEG activity following target offset.  \nAs expected, oddballs elicited a much larger P3b wave than standards, with the \nprototypical P3b scalp distribution. This indicates that the modifications we made to the oddball \nparadigm did not disrupt its fundamental nature. The behavioral data from probe trials showed \nthat participants maintained a more accurate and less biased representation of the target for \noddballs than for standards. In addition, the EEG decoding results showed that the neural \nrepresentation of the target was enhanced for oddballs relative to standards in the period \nfollowing P3b onset. These findings are, to our knowledge, the first direct evidence that working \nmemory is enhanced for rare, task-relevant stimulus categories.  \nThe present results do not indicate the nature of the neural mechanism underlying \nenhanced working memory, but such an effect could potentially be a result of increased firing by \nneurons in the locus coeruleus, which would lead to a phasic increase in norepinephrine release \nthroughout the cortex (Aston-Jones & Cohen, 2005; Krebs et al., 2018; Murphy et al., 2011; \nNieuwenhuis et al., 2005). This is consistent with prior findings that norepinephrine is critical for \nsignaling oddballs in humans and that norepinephrine depletion leads to working memory \ndeficits in nonhuman primates (Arnsten, 2006; Arnsten & Goldman-Rakic, 1985; Brozoski et al., \n1979; Cai et al., 1993; Franowicz et al., 2002; Strange & Dolan, 2007; Zhang et al., 2013). \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted March 20, 2024. ; https://doi.org/10.1101/2024.03.20.585952doi: bioRxiv preprint \n\nEnhanced Rare Working Memory Representations \n \nThe finding that working memory representations were enhanced for rare items—with \ngreater accuracy for participants with larger P3b amplitudes—might be interpreted as evidence \nfor the context updating hypothesis of the P3b wave, which was proposed over 40 years ago by \nDonchin (1981). Although Donchin did not use the term working memory, more recent \nresearchers have interpreted the context updating hypothesis as stating that the P3b is related to \nworking memory encoding (Kok, 2001; Polich, 2007). However, there is not much evidence to \nsupport the idea that the P3b specifically reflects the updating of working memory (Verleger, \n2008). Indeed, using a reference-back task that was specifically designed to distinguish between \nupdating and other processes, Rac-Lubashevsky and Kessler (2019) found no evidence that a P3b \nwas triggered when working memory updating was required. Moreover, the P3b component in \nthe present study was more than three times larger for oddballs than for standards, and yet \nbehavioral accuracy on probe trials (a fairly direct measure of working memory) was only \nslightly greater for oddballs than for standards. Similarly, although decoding accuracy greater for \noddballs than for standards, the difference in decoding accuracy was modest.  \nThus, it is unlikely that the neural mechanisms that produce the P3b component are the \nsame mechanisms that encode information into working memory. The correlation between P3b \namplitude and working memory accuracy in the present study is more likely related to a shared \nimpact of attentional allocation on P3b amplitude and working memory encoding. By analogy, \npupil dilation is also enhanced for oddballs, correlated with working memory performance, and \nlinked with locus coeruleus activity (Aminihajibashi et al., 2019; Eckstein et al., 2019; Gilzenrat \net al., 2010; Murphy et al., 2011; Robison et al., 2023; Robison & Unsworth, 2019; Zokaei et al., \n2019). However, this does mean that the mechanisms that produce pupil dilation are the same as \nthe mechanisms that encode information into working memory. Instead, the P3b may be related \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted March 20, 2024. ; https://doi.org/10.1101/2024.03.20.585952doi: bioRxiv preprint \n\nEnhanced Rare Working Memory Representations \n \nto models in which locus coeruleus activity provides a signal across the cerebral cortex that \npredictions about the environment have been strongly violated (Jordan, 2023). However, \nadditional research would be needed to provide conclusive evidence of causal links between \nlocus coeruleus activity and the P3b wave. \nBehavioral responses in working memory tasks can be influenced by decision processes \nat the time of report in addition to encoding and maintenance processes. To provide additional \ninformation about the nature of the differences in working memory between rare and frequent \nstimuli, we decoded the location of the target being held in working memory using the scalp \ntopography of the ERP signal during the interval immediately following the target, prior to the \nprobe period. The finding of greater decoding accuracy for oddballs than for standards during \nthis period suggests that rare stimuli are encoded and maintained more accurately than frequent \nstimuli (as opposed to an effect of rareness that is isolated to decision processes at the time of \nresponse). However, we cannot be certain that these ERP decoding effects reflect the same \nprocesses responsible for the improved behavioral performance for rare stimuli on probe trials. \nIndeed, the ERP decoding analyses were designed to distinguish among the four different \nlocations within a given response category, which were separated by 90°, whereas the behavioral \nresponses were typically within 5° of the true target location (see Figure 2A). The coarseness of \nthe decoding process may partly explain why the differences in decoding accuracy between \noddballs and standards were so small. We are not yet at the point where we can decode \ndifferences in stimulus location from scalp voltages with the same precision as the behavioral \nresponses. \nWe would like to emphasize that the present study used a behavioral task to create the \nrare and frequent stimulus categories in the context of the behavioral task. These were not \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted March 20, 2024. ; https://doi.org/10.1101/2024.03.20.585952doi: bioRxiv preprint \n\nEnhanced Rare Working Memory Representations \n \ncategories that vary in their frequency or novelty outside of this task. Stimuli that are intrinsically \nnovel elicit a very different pattern of neural activity, including the novelty P3a component \n(which has a much more frontal scalp distribution than the P3b component) and extensive BOLD \nactivity in temporal and inferior frontal cortex (Strobel et al., 2008). It is not known how intrinsic \nnovelty impacts working memory. Performance in visual working memory tasks is more \naccurate for familiar than for unfamiliar stimuli under some conditions (Chen et al., 2006; \nJackson & Raymond, 2008; Ngiam et al., 2019), but this likely reflects the use of longer-term \nmemory representations to aid in task performance. Click or tap here to enter text.Novelty can \nimprove working memory under other conditions, particularly encoding processes (Mayer et al., \n2011). Additional research will be necessary to fully understand how this kind of intrinsic \nnovelty is related to the encoding and maintenance of information leading to enhancements in \nworking memories.  \nProactive interference provides a potential explanation for the finding of differences in \nworking memory accuracy between the rare and frequent categories in the present study. In \nstudies using verbal materials, working memory accuracy is reduced when similar information is \npresented across many trials (Baddeley, 1986; Keppel & Underwood, 1962). In the present \nstudy, locations in the frequent category were, by definition, presented more frequently than the \nlocations in the rare category, which could have led to proactive interference. However, this kind \nof interference is not typically seen in visual working memory tasks using brief retention \nintervals like those in the current study (Lin & Luck, 2012; Oberauer et al., 2017). Future \nresearch would be needed to provide a conclusive test of this explanation. \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted March 20, 2024. ; https://doi.org/10.1101/2024.03.20.585952doi: bioRxiv preprint \n\nEnhanced Rare Working Memory Representations \n \nData and Code Availability \nAll data will be available at: doi.org/10.17605/OSF.IO/HV7JU, upon publication.  \nAuthor Contributions \nCarlos Daniel Carrasco: conceptualization, analyses, writing; Aaron Matthew Simmons: \nsoftware development, writing; John E. Kiat: analyses, writing; Steven J. Luck: \nconceptualization, writing, development of methods, funding acquisition. Editing/Review was \nperformed by all authors prior to submission of the final version of the manuscript.  \nAcknowledgements \nWe thank the Luck lab for all the help and support always.  \nFunding \nThis study was supported by grants R01MH087450, R01EY033329, and R01MH076226 from \nthe National Institute of Mental Health.   \n \n \n \n \n \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted March 20, 2024. ; https://doi.org/10.1101/2024.03.20.585952doi: bioRxiv preprint \n\nEnhanced Rare Working Memory Representations \n \n \n \nReferences \n \nAminihajibashi, S., Hagen, T., Foldal, M. D., Laeng, B., & Espeseth, T. (2019). Individual differences \nin resting-state pupil size: Evidence for association between working memory capacity and pupil \nsize variability. International Journal of Psychophysiology, 140, 1–7. \nhttps://doi.org/10.1016/j.ijpsycho.2019.03.007 \nArnsten, A. F. T. (2006). Fundamentals of Attention-Deficit/Hyperactivity Disorder: Circuits and \nPathways. In J Clin Psychiatry (Vol. 67, Issue 8). \nArnsten, A. F. T., & Goldman-Rakic, P. S. (1985). α2-Adrenergic Mechanisms in Prefrontal Cortex \nAssociated with Cognitive Decline in Aged Nonhuman Primates. In New Series (Vol. 13, Issue \n4731). \nAston-Jones, G., & Cohen, J. D. (2005). An integrative theory of locus coeruleus-norepinephrine \nfunction: Adaptive gain and optimal performance. In Annual Review of Neuroscience (Vol. 28, \npp. 403–450). https://doi.org/10.1146/annurev.neuro.28.061604.135709 \nBaddeley, A. (1986). Working memory. Oxford: Oxford University Press, Clarendon Press. \nBae, G. Y. (2022). Breaking the cardinal rule: The impact of interitem interaction and attentional \npriority on the cardinal biases in orientation working memory. Attention, Perception, and \nPsychophysics, 84(7), 2186–2194. https://doi.org/10.3758/s13414-021-02374-2 \nBae, G. Y., & Luck, S. J. (2019). Decoding motion direction using the topography of sustained ERPs \nand alpha oscillations. NeuroImage, 184, 242–255. \nhttps://doi.org/10.1016/j.neuroimage.2018.09.029 \nBae, G. Y. (2021). The Time Course of Face Representations during Perception and Working \nMemory Maintenance. Cerebral Cortex Communications, 2(1), 1–12. \nhttps://doi.org/10.1093/texcom/tgaa093 \nBae, G. Y., & Luck, S. J. (2018). Dissociable Decoding of Spatial Attention and Working Memory \nfrom EEG Oscillations and Sustained Potentials. The Journal of Neuroscience, 38(2), 2860–17. \nhttps://doi.org/10.1523/JNEUROSCI.2860-17.2017 \nBledowski, C., Prvulovic, D., Goebel, R., Zanella, F. E., & Linden, D. E. J. (2004). Attentional \nsystems in target and distractor processing: A combined ERP and fMRI study. NeuroImage, \n22(2), 530–540. https://doi.org/10.1016/j.neuroimage.2003.12.034 \nBrozoski, T. J., Brown, R. M., Rosvold, H. E., & Goldman, P. S. (1979). Cognitive Deficit Caused by \nRegional Depletion of Dopamine in Prefrontal Cortex of Rhesus Monkey. In New Series (Vol. \n205, Issue 4409). \nCai, J. X., Ma, Y.-Y., & Hu, X.-T. (1993). Reserpine impairs spatial working memory performance in \nmonkeys: reversal by the az-adrenergic agonist clonidine. In Brain Research (Vol. 614). \nChen, D., Yee Eng, H., & Jiang, Y. (2006). Visual working memory for trained and novel polygons. \nVisual Cognition, 14(1), 37–54. https://doi.org/10.1080/13506280544000282 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted March 20, 2024. ; https://doi.org/10.1101/2024.03.20.585952doi: bioRxiv preprint \n\nEnhanced Rare Working Memory Representations \n \nClark, V. P., Fannon, S., Lai, S., Benson, R., & Bauer, L. (2000). Responses to Rare Visual Target \nand Distractor Stimuli Using Event-Related fMRI. \nCohen J. (1988). Statistical Power Analysis for the Behavioral Sciences Second Edition. \nDelorme, A., & Makeig, S. (2004). EEGLAB: an open source toolbox for analysis of single-trial EEG \ndynamics including independent component analysis. In Journal of Neuroscience Methods (Vol. \n134). http://www.sccn.ucsd.edu/eeglab/ \nDietterich, T. G., & Bakiri, G. (1995). Solving Multiclass Learning Problems via Error-Correcting \nOutput Codes. In Journal of Artiicial Intelligence Research (Vol. 2). \nDonchin, E. (1981). Surprise!...Surprise? Psychophysiology, 18(5), 493–513. \nhttps://doi.org/10.1111/j.1469-8986.1981.tb01815.x \nDonchin, E., & Coles, M. G. H. (1988). Is the P300 component a manifestation of context updating? \nBehavioral and Brain Sciences, 11(3), 357–374. https://doi.org/10.1017/S0140525X00058027 \nEckstein, M. K., Starr, A., & Bunge, S. A. (2019). How the inference of hierarchical rules unfolds \nover time. Cognition, 185, 151–162. https://doi.org/10.1016/j.cognition.2019.01.009 \nFranowicz, J. S., Kessler, L. E., Dailey Borja, C. M., Kobilka, B. K., Limbird, L. E., & Arnsten, A. F. \nT. (2002). Mutation of the 2A-Adrenoceptor Impairs Working Memory Performance and Annuls \nCognitive Enhancement by Guanfacine. \nGilzenrat, M. S., Nieuwenhuis, S., Jepma, M., & Cohen, J. D. (2010). Pupil diameter tracks changes \nin control state predicted by the adaptive gain theory of locus coeruleus function. Cognitive, \nAffective and Behavioral Neuroscience, 10(2), 252–269. https://doi.org/10.3758/CABN.10.2.252 \nJackson, M. C., & Raymond, J. E. (2008). Familiarity Enhances Visual Working Memory for Faces. \nJournal of Experimental Psychology: Human Perception and Performance, 34(3), 556–568. \nhttps://doi.org/10.1037/0096-1523.34.3.556 \nJohnson, R. (1988). SCALP-RECORDED P300 ACTIVITY IN PATIENTS FOLLOWING \nUNILATERAL TEMPORAL LOBECTOMY. In Brain (Vol. 111). \nhttps://academic.oup.com/brain/article/111/6/1517/297108 \nJordan, R. (2023). The locus coeruleus as a global model failure system. In Trends in Neurosciences. \nElsevier Ltd. https://doi.org/10.1016/j.tins.2023.11.006 \nKatayama, J., & Polich, J. (1998). Stimulus context determines P3a and P3b. Psychophysiology, \n35(1), 23–33. https://doi.org/10.1111/1469-8986.3510023 \nKeppel, G., & Underwood, B. J. (1962). Proactive Inhibition in Short-Term Retention of Single Items \n1. \nKim, H. (2014). Involvement of the dorsal and ventral attention networks in oddball stimulus \nprocessing: A meta-analysis. Human Brain Mapping, 35(5), 2265–2284. \nhttps://doi.org/10.1002/hbm.22326 \nKok, A. (2001). On the utility of P3 amplitude as a measure of processing capacity. \nPsychophysiology, 38(3), 557–577. https://doi.org/10.1017/S0048577201990559 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted March 20, 2024. ; https://doi.org/10.1101/2024.03.20.585952doi: bioRxiv preprint \n\nEnhanced Rare Working Memory Representations \n \nKrebs, R. M., Park, H. R. P., Bombeke, K., & Boehler, C. N. (2018). Modulation of locus coeruleus \nactivity by novel oddball stimuli. Brain Imaging and Behavior, 12(2), 577–584. \nhttps://doi.org/10.1007/s11682-017-9700-4 \nKutas, M., Mccarthy, G., & Donchin, E. (1977). Augmenting Mental Chronometry: The P300 as a \nMeasure of Stimulus Evaluation Time. In Source: Science (Vol. 197, Issue 4305). \nLakens, D. (2013). Calculating and reporting effect sizes to facilitate cumulative science: A practical \nprimer for t-tests and ANOVAs. Frontiers in Psychology, 4(NOV). \nhttps://doi.org/10.3389/fpsyg.2013.00863 \nLin, P. H., & Luck, S. J. (2012). Proactive interference does not meaningfully distort visual working \nmemory capacity estimates in the canonical change detection task. Frontiers in Psychology, \n3(FEB). https://doi.org/10.3389/fpsyg.2012.00042 \nLinden, D. E. J., Prvulovic, D., Formisano, E., Völlinger, M., Zanella, F. E., Goebel, R., & Dierks, T. \n(1999). The functional neuroanatomy of target detection: an fMRI study of visual and auditory \noddball tasks. Cerebral Cortex, 9(8), 815–823. \nLopez-Calderon, J., & Luck, S. J. (2014). ERPLAB: An open-source toolbox for the analysis of \nevent-related potentials. Frontiers in Human Neuroscience, 8(1 APR), 1–14. \nhttps://doi.org/10.3389/fnhum.2014.00213 \nLuck, S. J. (2014). An Introduction To The Event-Related Potential Technique (2nd Editio). MIT \nPress Journals. \nLuck, S. J. (2023). ERPLAB Decoding Tutorial. https://github.com/ucdavis/erplab/wiki/ERPLAB-\nDecoding-Tutorial \nMayer, J. S., Kim, J., & Park, S. (2011). Enhancing visual working memory encoding: The role of \ntarget novelty. Visual Cognition, 19(7), 863–885. https://doi.org/10.1080/13506285.2011.594459 \nMecklinger, A., & Ullsperger, P. (1993). P3 varies with stimulus categorization rather than \nprobability. In Electroencephalography and clinical Neurophysiology (Vol. 86). \nMurphy, P. R., Robertson, I. H., Balsters, J. H., & O’connell, R. G. (2011). Pupillometry and P3 index \nthe locus coeruleus-noradrenergic arousal function in humans. Psychophysiology, 48(11), 1532–\n1543. https://doi.org/10.1111/j.1469-8986.2011.01226.x \nNgiam, W. X. Q., Khaw, K. L. C., Holcombe, A. O., & Goodbourn, P. T. (2019). Visual working \nmemory for letters varies with familiarity but not complexity. Journal of Experimental \nPsychology: Learning, Memory, and Cognition, 45(10), 1761. \nNieuwenhuis, S., Aston-Jones, G., & Cohen, J. D. (2005). Decision making, the P3, and the locus \ncoeruleus-norepinephrine system. In Psychological Bulletin (Vol. 131, Issue 4, pp. 510–532). \nhttps://doi.org/10.1037/0033-2909.131.4.510 \nNieuwenhuis, S., De Geus, E. J., & Aston-Jones, G. (2011). The anatomical and functional \nrelationship between the P3 and autonomic components of the orienting response. In \nPsychophysiology (Vol. 48, Issue 2, pp. 162–175). Blackwell Publishing Inc. \nhttps://doi.org/10.1111/j.1469-8986.2010.01057.x \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted March 20, 2024. ; https://doi.org/10.1101/2024.03.20.585952doi: bioRxiv preprint \n\nEnhanced Rare Working Memory Representations \n \nOberauer, K., Awh, E., & Sutterer, D. W. (2017). The role of long-term memory in a test of visual \nworking memory: Proactive facilitation but no proactive interference. Journal of Experimental \nPsychology: Learning Memory and Cognition, 43(1), 1–22. https://doi.org/10.1037/xlm0000302 \nPaller, K. A., Mccarthy, G., Roessler, E., Allison, T., & Wood, C. C. (1992). Potentials evoked in \nhuman and monkey medial temporal lobe during auditory and visual oddball paradigms. In \nElectroencephalography and clinical Neurophysiology (Vol. 84). \nPeirce, J., Gray, J. R., Simpson, S., MacAskill, M., Höchenberger, R., Sogo, H., Kastman, E., & \nLindeløv, J. K. (2019). PsychoPy2: Experiments in behavior made easy. Behavior Research \nMethods, 51(1), 195–203. https://doi.org/10.3758/s13428-018-01193-y \nPolich, J. (1986). A’ITENTION, PROBABILITY, AND TASK DEMANDS AS DETERMINANTS \nOF P300 LATENCY FROM AUDITORY STIMULI t. In Electroencephalograph)’ and clinical \nNeurophysiology (Vol. 63). \nPolich, J. (2007). Updating P300: An integrative theory of P3a and P3b. Clinical Neurophysiology, \n118(10), 2128–2148. https://doi.org/10.1016/j.clinph.2007.04.019 \nPolich, J. (2012). Neuropsychology of P300. The Oxford Handbook of Event-Related Potential \nComponents, 159–188. \nPratte, M. S., Park, Y. E., Rademaker, R. L., & Tong, F. (2017). Accounting for stimulus-specific \nvariation in precision reveals a discrete capacity limit in visual working memory. Journal of \nExperimental Psychology: Human Perception and Performance, 43(1), 6–17. \nhttps://doi.org/10.1037/xhp0000302 \nRac-Lubashevsky, R., & Kessler, Y. (2019). Revisiting the relationship between the P3b and working \nmemory updating. Biological Psychology, 148. https://doi.org/10.1016/j.biopsycho.2019.107769 \nRobison, M. K., Ralph, K. J., Gondoli, D. M., Torres, A., Campbell, S., Brewer, G. A., & Gibson, B. \nS. (2023). Testing locus coeruleus-norepinephrine accounts of working memory, attention \ncontrol, and fluid intelligence. Cognitive, Affective and Behavioral Neuroscience, 23(4), 1014–\n1058. https://doi.org/10.3758/s13415-023-01096-2 \nRobison, M. K., & Unsworth, N. (2019). Pupillometry tracks fluctuations in working memory \nperformance. Attention, Perception, and Psychophysics, 81(2), 407–419. \nhttps://doi.org/10.3758/s13414-018-1618-4 \nSoltani, M., & Knight, R. T. (2000). Neural Origins of the P300. In Critical Reviews in Neurobiology \n(Vol. 14). \nStrange, B. A., & Dolan, R. J. (2007). β-adrenergic modulation of oddball responses in humans. \nBehavioral and Brain Functions, 3. https://doi.org/10.1186/1744-9081-3-29 \nStrobel, A., Debener, S., Sorger, B., Peters, J. C., Kranczioch, C., Hoechstetter, K., Engel, A. K., \nBrocke, B., & Goebel, R. (2008). Novelty and target processing during an auditory novelty \noddball: A simultaneous event-related potential and functional magnetic resonance imaging \nstudy. NeuroImage, 40(2), 869–883. https://doi.org/10.1016/j.neuroimage.2007.10.065 \nVerleger, R. (2008). P3b: Towards some decision about memory. In Clinical Neurophysiology (Vol. \n119, Issue 4, pp. 968–970). https://doi.org/10.1016/j.clinph.2007.11.175 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted March 20, 2024. ; https://doi.org/10.1101/2024.03.20.585952doi: bioRxiv preprint \n\nEnhanced Rare Working Memory Representations \n \nWei, X. X., & Stocker, A. A. (2015). A Bayesian observer model constrained by efficient coding can \nexplain “anti-Bayesian” percepts. Nature Neuroscience, 18(10), 1509–1517. \nhttps://doi.org/10.1038/nn.4105 \nZhang, Z., Cordeiro Matos, S., Jego, S., Adamantidis, A., & Séguéla, P. (2013). Norepinephrine \nDrives Persistent Activity in Prefrontal Cortex via Synergistic α1 and α2 Adrenoceptors. PLoS \nONE, 8(6). https://doi.org/10.1371/journal.pone.0066122 \nZokaei, N., Board, A. G., Manohar, S. G., & Nobre, A. C. (2019). Modulation of the pupillary \nresponse by the content of visual working memory. Proceedings of the National Academy of \nSciences of the United States of America, 116(45), 22802–22810. \nhttps://doi.org/10.1073/pnas.1909959116 \n  \n \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted March 20, 2024. ; https://doi.org/10.1101/2024.03.20.585952doi: bioRxiv preprint","source_license":"CC-BY-4.0","license_restricted":false}