About

Thomas (Tommy) Sprague is an Assistant Professor in the Psychological & Brain Sciences department at the University of California, Santa Barbara. He received his BA in Cognitive Sciences from Rice University in 2010 and his PhD in Neurosciences with a Specialization in Computational Neurosciences from the University of California, San Diego in 2016. His graduate work with John Serences focused on developing and applying novel multivariate analysis methods to human neuroimaging techniques to understand how neural systems represent information in support of dynamic behavioral goals. Prior to joining UCSB, Dr. Sprague completed a postdoctoral fellowship with Clayton Curtis and Wei Ji Ma, studying how neural systems represent both the contents of visual working memory and their uncertainty by building new multivariate analysis methods.

Research topics

• Psychology
• Cognitive psychology
• Neuroscience
• Computer science
• Artificial intelligence

Selected publications

• Computational methods for extracting neural correlates of working memory and mental imagery from intracortical recordings in human visual cortex

  Journal of Vision · 2025-07-15

  articleOpen access

  Neural activity in early visual cortex (EVC) is known to contribute to visual working memory (WM) and mental imagery (MI), but the role of spiking activity in humans remains unclear. This study investigates computational techniques for extracting spiking activity and their ability to reveal correlates of WM and MI. Intracortical recordings were collected from two awake blind humans implanted with a 96-channel visual prosthesis in EVC during a delayed-match-to-sample (DMTS) WM task and a MI visualization task. In 465 trials of the WM task, participants encoded visual perceptions (phosphenes) elicited by stimulation of one of three electrodes, maintained them over a 5-second delay, and recalled whether a subsequent phosphene was the same or different. The MI task followed a similar structure, with recall replaced by vivid mental visualization. Neural activity during stimulation, delay, recall, and spontaneous periods was analyzed using methods to extract multi-unit activity (MUA), entire spiking activity (ESA), and local field potential (LFP) signals. Significant differences were observed in MUA, ESA, and LFP (theta, alpha, and beta bands) across trial periods (t-tests, p < 0.05). ESA and MUA exhibited electrode-specific neural signatures during delay and recall periods, with over 90% classification accuracy in leave-one-trial-out cross-validation (LOOCV). Stimulus-specific ESA changes remained decodable throughout delay and recall (random forest classifier sliding window, LOOCV, 70% of windows above chance), indicating sustained stimulus-selective activity for both tasks despite day-to-day variability. These findings reveal sustained stimulus-selective spiking activity in human EVC during WM and MI tasks, underscoring its critical role in retaining and recalling information and providing new insights into the neural mechanisms underlying perception and cognition.

  PublisherDOI

• Retinotopic Spatial Working Memory Representations Are Not Affected by Task-irrelevant Visual Stimuli

  Journal of Cognitive Neuroscience · 2025-09-30

  articleOpen accessSenior author

  Visual working memory (WM) enables the maintenance and manipulation of information no longer accessible in the world. Previous research has identified spatial WM representations in sustained activation patterns in visual, parietal, and frontal cortex, while MEG/EEG studies have additionally supported a role for "activity-silent" mechanisms revealed by transient reactivation or amplification of an existing representation by a task-irrelevant "ping" stimulus. In natural vision, the delay period between encoding information into WM and its use to guide behavior is rarely "empty," as is the case in many laboratory experiments. Instead, eye movements, movement of the individual, and events in the environment result in visual inputs that may overwrite or impair the fidelity of WM representations, especially in early sensory cortices. Here, we evaluated the extent to which a brief, irrelevant interrupting visual stimulus presented during a spatial WM delay period impaired behavioral performance and retinotopic WM representation fidelity assayed using an inverted encoding model. On each trial, participants (both sexes) viewed two target dots and were immediately postcued to remember the precise spatial position of one dot. On 50% of trials, a brief interrupter stimulus appeared. While we observed strong transient univariate visual responses to the interrupter stimulus, we saw no change in reconstructed neural WM representations due to this interruption, nor a change in behavioral performance on a continuous recall task. This suggests that spatial WM representations can be robust to interference from incoming task-irrelevant visual information, perhaps related to their role in guiding movements.

  PublisherOA PDFDOI

• Efficient Multi Subject Visual Reconstruction from fMRI Using Aligned Representations

  ArXiv.org · 2025-05-03

  preprintOpen access

  This work introduces a novel approach to fMRI-based visual image reconstruction using a subject-agnostic common representation space. We show that the brain signals of the subjects can be aligned in this common space during training to form a semantically aligned common brain. This is leveraged to demonstrate that aligning subject-specific lightweight modules to a reference subject is significantly more efficient than traditional end-to-end training methods. Our approach excels in low-data scenarios. We evaluate our methods on different datasets, demonstrating that the common space is subject and dataset-agnostic.

  PublisherOA PDFDOI

• Spatial attention to multiple stimuli does not reduce evoked SSVEP power relative to focal attention

  Journal of Vision · 2025-07-15

  articleOpen accessSenior author

  Typically, when spatial attention is directed to multiple locations simultaneously, perceptual sensitivity to stimuli at those locations is reduced relative to attending only a single location (e.g., Ling & Carrasco, 2006; Popovkina et al., 2021). Additionally, neural responses to visual stimuli are generally attenuated under distributed attention compared to focal attention conditions (e.g., McMains & Somers, 2005; Toffanin et al., 2009). However, fMRI data from a previous study in our lab show a different pattern: focal spatial attention enhances the neural representation of a stimulus compared to when the stimulus is ignored, but the degree of enhancement does not decrease as an additional stimulus is attended (Harrison et al., VSS 2024). To resolve the discrepancies between this finding and prior literature, we conducted an EEG experiment in which participants performed a selective attention task modeled after our previous fMRI experiment. fMRI and EEG have been suggested to assay complementary attentional modulations arising from distinct physiological processes (e.g., Itthipuripat et al., 2019), motivating its use for better understanding the neural mechanisms surrounding distributed spatial attention. Participants were cued on each trial to attend to the fixation point, to one cued peripheral location, or to two cued peripheral locations where flickering stimuli (20 and 24 Hz) appeared on every trial. In our pilot dataset, we compared steady-state evoked potential (SSVEP) amplitudes for each stimulus across attentional conditions and found that while attending to a single stimulus increases evoked power, the effects of distributed attention, compared to focal attention, are variable. In line with our prior fMRI results, there is no clear evidence that distributed attention reduces attentional enhancement compared to focal attention. This is consistent with a model where attentional feedback signals enhance sensory responses to a similar degree when attention is focused as when it is distributed.

  PublisherDOI

• Neural mechanisms of resource allocation in working memory

  Science Advances · 2025-04-09 · 8 citations

  articleOpen accessCorresponding

  To mitigate capacity limits of working memory, people allocate resources according to an item's relevance. However, the neural mechanisms supporting such a critical operation remain unknown. Here, we developed computational neuroimaging methods to decode and demix neural responses associated with multiple items in working memory with different priorities. In striate and extrastriate cortex, the gain of neural responses tracked the priority of memoranda. We decoded higher-priority memoranda with smaller error and lower uncertainty. Moreover, these neural differences predicted behavioral differences in memory prioritization between and within participants. Trial-wise variability in the magnitude of delay activity in the frontal cortex predicted differences in decoded precision between low- and high-priority items in visual cortex. These results support a model in which feedback signals broadcast from frontal cortex sculpt the gain of memory representations in the visual cortex according to behavioral relevance, thus identifying a neural mechanism for resource allocation.

  PublisherDOI

• Pinpointing the Sources of Top-Down Feedback in Visual Working Memory

  Journal of Vision · 2025-07-15

  articleOpen access

  How are working memory (WM) representations stored over brief delays? Neuroimaging studies consistently report persistent above-baseline activity in frontal cortex during WM maintenance but not visual cortex, while conversely, WM content can be decoded from visual cortex but less consistently in frontal cortex. Based on these findings, we hypothesized that persistent activity in frontal cortex reflects top-down feedback that enhances the fidelity of WM representations in visual cortex. To test this hypothesis, we developed a novel approach combining univariate analysis to assess trial-by-trial delay-period activation with decoding to quantify the quality of WM representations on a trial-wise basis. We applied this method to four WM datasets spanning three distinct studies, enabling a comprehensive test of our hypothesis and including testing how results generalize across tasks. Consistent with previous findings, we observed persistent activity in multiple frontal and parietal regions (but not visual cortex) during WM delays, and were able to decode memory context most robustly in V1-V3 during these same memory delays. Critically, across all datasets, trial-by-trial variations in delay activity amplitude in frontal and parietal cortices - specifically the superior precentral sulcus and intraparietal sulcus - predicted the quality of decoded WM representations in V1-V3. Thus, persistent activity in frontal and parietal cortices may reflect feedback signals targeting WM representations in visual cortex. We propose that these feedback signals may sculpt population activity in visual cortex, improving the quality memory representations.

  PublisherDOI

• Interactions between working memory and attention depend on remembered feature dimension

  2025-12-12

  articleOpen accessSenior author

  Prominent theories of visual search state that goal-relevant information is maintained in working memory (WM) to guide attention to relevant, searched-for items. Prioritization from WM is sufficiently strong to direct attention towards items that are irrelevant to current search goals. These ‘incidental capture’ results have typically been observed when remembering color or shape stimuli, but, in principle, should similarly occur using other strong guiding features such as motion. However, recent work suggests that neural representations of remembered motion directions are transformed into spatial coordinates. If so, then remembered stimuli which can easily be spatially recoded (e.g., remembered motion direction) should not interfere with visual search when feature-matching distractors are present any more than interference caused by a salient distractor that does not match the contents of WM. We replicated the traditional incidental capture finding for color: distractors matching the remembered color captured attention more than feature singletons. In contrast, when participants were required to remember a motion direction, capture was equal across distractor conditions: salient distractors captured attention, but no additional capture was observed when the remembered motion direction matched the distracting motion direction. Intriguingly, responses were faster when the search target was at a location that neural studies predict would be prioritized if remembered motion was spatially recoded. Our results indicate that some features in WM are more likely to impact attentional processing than others and that this difference in attentional capture between feature dimensions may emerge due to the ability to reformat memory representations.

  PublisherDOI

• A Dense Sampling Study on Visual Working Memory Across the Human Menstrual Cycle

  Journal of Vision · 2025-07-15

  articleOpen accessSenior author

  Working memory (WM) is the ability to temporarily store and manipulate information to guide later behavior (Baddeley, 2010; Ma et al., 2014), and recent studies have shown that WM information is stored by distributed activity across early visual cortex and higher areas including parietal and prefrontal cortex (PFC) (Christophel et al., 2017; Curtis & Sprague, 2021). While most studies average data across participants, there is important evidence that these areas (especially PFC) show variations in neural function across the menstrual cycle, and measures of cognitive function (e.g., n-back) reflect these hormone-related fluctuations (Jacobs & D’Esposito, 2011). However, this previous work does not address how these hormone fluctuations impact aspects of visual WM performance, including capacity, precision, and inter-trial serial dependence (Fischer & Whitney, 2014; Bliss et al., 2017). This is important, because visual WM is a tractable system for modeling relationships between neural function and behavior in humans (Li et al., 2021). Here, we employed dense sampling methods (Pritschet et al., 2021) to assay WM performance across n = 6 participants’ natural menstrual cycles. At each of ~15 sessions per participant (approximately every other day), we measured WM capacity (Change Localization Task; Zhao et al., 2023) and WM precision (memory guided saccade [MGS] task, Funhashi et al., 1989; Li & Sprague, 2023), along with salivary measures of ovarian hormones (estradiol/progesterone) and survey measures of state anxiety, sleep quality, and caffeine intake. We observed fluctuations in visual WM performance as measured by MGS response time, precision, and magnitude of serial dependence throughout the menstrual cycle, while WM capacity was remarkably stable. Importantly, these differences could not be explained by variations in measures of sleep quality or anxiety. This implies that the rapid fluctuation of ovarian hormones may be responsible for the change in visual WM performance measured with the MGS task.

  PublisherDOI

• No spontaneous representations of spatial location in feature-selective retinotopic cortex during working memory for color and motion

  Journal of Vision · 2025-07-15 · 1 citations

  articleOpen accessSenior author

  Per the ‘sensory recruitment’ model of working memory (WM), frontal and parietal cortex engage sensory regions which have precise feature selectivity to maintain information in WM during delay periods (Curtis & D’Esposito, 2003; Postle, 2006). Since this model predicts that neural populations with specialized tuning should be recruited during WM maintenance, regions with selectivity for non-spatial features, such as color (hV4/VO1/VO2) and motion (TO1/TO2), should be recruited to maintain color and motion information, respectively, and the populations best suited for robust recruitment are those spatially aligned with the sample stimulus location. Some previous studies have shown that when remembering specific object features, the associated location is automatically encoded (Foster et al., 2017; Pratte & Tong, 2014), while others suggest that features are maintained in a spatially global manner (Ester et al., 2009). Here, we tested whether encoding features in WM results in recruitment and maintenance of activation of spatially tuned populations, or instead if feature representations are encoded globally. Participants viewed a colorful moving dot stimulus at a random location on each trial and were postcued to remember its color or motion. After a 12s delay, they adjusted the relevant feature of a probe stimulus, presented at the sample location, to match the sample. We used multivariate spatial inverted encoding models to quantify multivariate activation patterns. Strikingly, in retinotopic color- and motion-selective regions, reconstructed maps contained no representation of the sample stimulus location during the delay period. In sharp contrast, parietal cortex had a robust delay-period representation of the sample stimulus location, even though location was irrelevant for the task. Together, these results indicate that space is not obligatorily encoded in feature-selective cortices. Rather, it appears that parietal cortex encodes the location of the sample stimulus, potentially instantiating feedback signals to feature-selective regions to bind features to a location.

  PublisherDOI

• Prioritizing Working Memory Resources Depends on the Prefrontal Cortex

  Journal of Neuroscience · 2025-01-27 · 5 citations

  articleOpen access

  How the prefrontal cortex contributes to working memory remains controversial, as theories differ in their emphasis on its role in storing memories versus controlling their content. To adjudicate between these competing ideas, we tested how perturbations to the human (both sexes) lateral prefrontal cortex impact the storage and control aspects of working memory during a task that requires human subjects to allocate resources to memory items based on their behavioral priority. Our computational model made a strong prediction that disruption of this control process would counterintuitively improve memory for low-priority items. Remarkably, transcranial magnetic stimulation of retinotopically-defined superior precentral sulcus, but not intraparietal sulcus, unbalanced the prioritization of resources, improving memory for low-priority items as predicted by the model. Therefore, these results provide direct causal support for models in which the prefrontal cortex controls the allocation of resources that support working memory, rather than simply storing the features of memoranda.

  PublisherOA PDFDOI

Recent grants

• Effects of behavioral priority on working memory representations

  NIH · $77k · 2017–2020

Frequent coauthors

• John T. Serences

  University of California, San Diego

  58 shared

• Sirawaj Itthipuripat

  King Mongkut's University of Technology Thonburi

  33 shared

• Clayton E. Curtis

  29 shared

• Vy A. Vo

  23 shared

• Chaipat Chunharas

  King Chulalongkorn Memorial Hospital

  21 shared

• Daniel Thayer

  University of California, Santa Barbara

  15 shared

• Masih Rahmati

  Yale University

  15 shared

• Edward F. Ester

  University of Nevada, Reno

  14 shared

Education

• PhD in Neurosciences, with a specialization in Computational Neurosciences, Neurosciences Graudate Program

  UC San Diego

  2016

Awards & honors

• March 21, 2019 Tommy Sprague awarded Alfred P. Sloan Fellows…