Perceptual Training, Learning and Plasticity

Talk Session: Sunday, May 17, 2026, 5:15 – 7:15 pm, Talk Room 1
Moderator: Aaron Seitz, Northeastern University

Schedule of Events | Search Abstracts | Symposia | Talk Sessions | Poster Sessions

Talk 1, 5:15 pm, 35.11

Increase in Neural Conduction Speed Following Treatment for Extended Visual Deprivation

Pawan Sinha¹, Chetan Ralekar², Charles Shvartsman¹, Suma Ganesh³; ¹Massachusetts Institute of Technology, ²Indian Institute of Technology, Roorkee, India, ³Dr. Shroff's Charity Eye Hospital, New Delhi, India

Understanding whether the human visual system retains the capacity for myelin-related plasticity beyond early developmental critical periods is essential for both basic neuroscience and clinical practice. Here, we report longitudinal evidence from a cohort of children who experienced profound congenital visual deprivation until late childhood and then received sight-restoring cataract surgery. Using flash visual evoked potentials (fVEPs), we examined how the speed of neural transmission in early visual pathways changes over the year following treatment. Across six participants (mean age 12 years), we measured the latency of the P1 component - a well-established marker of conduction speed in the retino-cortical pathway - at five time points from three days to one year post-surgery. Immediately after treatment, P1 latencies were significantly delayed relative to neurotypical adults, consistent with slowed transmission in long-deprived pathways. However, we observed systematic and statistically significant latency reductions over time, converging to typical adult values by one year. A mixed-effects model revealed a strong effect of log-transformed time on latency (p < 0.001). The rate of convergence was markedly slower, by nearly an order of magnitude, than that documented in infancy, but the eventual approach to normal ranges demonstrates substantial residual plasticity. Comparisons to published neonatal data suggest that some myelination may have occurred prior to treatment due to residual light stimulation through cataracts. Nevertheless, the pronounced post-surgical latency decreases likely reflect activity-dependent increases in myelination within early visual white-matter tracts. This interpretation is supported by our prior structural imaging work showing myelin-associated changes in similar late-sighted populations. These findings challenge long-standing assumptions about the rigidity of myelination timelines, revealing meaningful plasticity in the visual system well past early childhood. They further point to the possibility that experience-dependent myelin remodeling may serve as a modulator of visual learning across the lifespan.

This work was supported by grant R01EY020517 from NEI(NIH)

Talk 2, 5:30 pm, 35.12

Short-term monocular deprivation facilitates behavioral transfer of perceptual learning to adjacent retinal locations

Zahide Pamir^1,2 (), Tutku Karahan^1,2, Berk Yüce^1,2; ¹Bilkent University, Ankara, Türkiye, ²Aysel Sabuncu Brain Research Center, Ankara, Türkiye

Topographic reorganization refers to an adaptive neural change driven by sensory deprivation, in which the deprived neurons begin responding to stimuli from neighboring locations. Although this phenomenon has been extensively studied at the neural level, its functional consequences remain unclear. The present study combines behavioral and fMRI measurements to investigate whether perceptual learning that occurs in a later deprived cortical region can enhance performance and induce corresponding neural changes at a nearby, untrained retinal location following short-term, reversible monocular deprivation. Fourteen healthy participants were trained for 15–20 days on a 2-IFC orientation discrimination task at the retinal location corresponding to the blindspot in the behavioral study. Thresholds were measured at the trained location and at an adjacent near location before and after training, under both undeprived and deprived viewing conditions. After confirming significant perceptual improvement at the trained blind spot location (p < .001), 2 (Time: pre, post) × 2 (ROI: deprived-near, undeprived-near) repeated-measures ANOVA revealed a significant Time × ROI interaction (p = .015), demonstrating that learning transferred more strongly to the near location in the deprived condition than in the undeprived condition. Moreover, eight participants completed fMRI sessions before and after training while performing the orientation-discrimination task. Our fMRI analyses focused on the blind-spot ROI during the presentation of near stimuli in the deprived condition. However, contrary to previous reports, we were unable to replicate findings indicating topographic reorganization in the cortical region corresponding to the blindspot during short-term deprivation, nor did we observe neural correlates of the behavioral enhancement associated with the transfer of perceptual learning under short-term deprivation. Although this study provides the first evidence suggesting a functional benefit of topographic reorganization by showing that short-term monocular deprivation may facilitate the transfer of learned abilities to adjacent, untrained locations, the underlying neural mechanisms remain unclear.

This project is funded by the Scientific and Technological Research Council of Turkey, TUBITAK through 2232-B Program (Project No: 121C139).

Talk 3, 5:45 pm, 35.13

Low-dimensional and optimised representations of high-level information in the expert brain

Andrea Ivan Costantino^1,2, Artem Platonov¹, Felipe Fontana Vieira^1,4, Emily Van Hove^1,2, Merim Bilalić³, Hans Op de Beeck^1,2; ¹KU Leuven, ²Leuven Brain Institute, ³Northumbria University, ⁴Ghent University

Chess expertise requires rapid extraction of relational structure from complex visual arrays. Previous research shows that experts organize chessboards into familiar relational configurations (chunks), exhibit distinctive eye movement patterns, and show activations in distinct brain areas, yet the underlying neural mechanisms remain unclear. We move from classic univariate localization approaches to representational geometry, asking what information experts encode, how it is organized, and where it resides. Using multivariate fMRI with 20 chess experts and 20 novices evaluating chess positions, we show that visual features are processed similarly by both groups, but experts compress this information and shift representational load to nonvisual, domain general control networks. Representational similarity analysis reveals comparable geometry for visual properties across early, intermediate, and posterior visual cortex in both groups. However, manifold dimensionality analyses show that experts systematically compress this information, providing more structured and efficient neural codes to downstream, higher level regions. Critically, only experts implement a neural geometry that aligns with high level relational properties such as checkmate status and strategic motifs, and these compressed manifolds paradoxically support richer decoding of task relevant features. Meta analytic network correlations reveal that novices rely on visual and semantic networks, whereas experts shift representational load to domain general control systems associated with working memory and navigation, implementing low dimensional, domain tuned, structured representations. Preliminary analyses of deep neural networks trained to play chess show similar geometrical properties. Both trained and untrained networks encode visual properties, but only trained networks encode relational features. However, in these networks relational coding emerges through dimensionality expansion rather than compression, suggesting that compression might reflect biological constraints. These results ground classic cognitive theories in measurable geometry and reveal how visual compression and network reallocation enable efficient extraction of relational structure from complex visual scenes.

Talk 4, 6:00 pm, 35.14

Recurrent Inhibition Preserves Existing Representations During New Visual Learning—Evidence from Psychophysics, Recurrent Circuit Models, and Deep Neural Networks

Yu-Ang Cheng¹, Yuka Sasaki¹, Thomas Serre¹, Takeo Watanabe¹; ¹Brown University

While balancing the acquisition of new information with the stability of existing representations is a fundamental challenge in biological and artificial learning systems, the underlying mechanisms remain elusive. This plasticity-stability dilemma concerns how training on a visual feature affects untrained features, yet little work has examined whether untrained features beyond the vicinity of the trained feature are altered or remain unchanged. Here, using psychophysics, recurrent circuit models, and deep neural network models (DNNs), we investigated how training influences untrained features in visual perceptual learning (VPL). First, we measured how untrained features changed after training. After 3 days of training on an orientation detection task, participants were tested with 18 orientations ranging from -90° to +90° relative to the trained orientation. Performance enhancement was confined to ±30° around the trained orientation, whereas orientations farther away showed neither enhancement nor deterioration. To explain why distant orientations remained stable, we modeled the result using a recurrent circuit of early visual areas and performed ablation analyses to identify which plasticity mechanisms were critical. Plasticity in excitatory or readout components produced nonspecific performance increases or decreases, inconsistent with behavior. In contrast, inhibitory-component plasticity generated a normalization-like computation that mitigated these unwanted changes and matched the psychophysical pattern. Next, to test whether this mechanism generalizes to modern deep learning systems, we introduced a trainable LayerNorm (TLN) module into DNNs. Compared with full fine-tuning, which produced interference with untrained features, TLN-only training preserved untrained orientations and reproduced the psychophysical results. Finally, we examined whether new learning degraded visual function. Linear-probe evaluation on ImageNet revealed substantial accuracy loss after full fine-tuning, whereas TLN-only training maintained performance close to the original model. These converging findings suggest that normalization-like processes mediated by recurrent inhibition serve as a key mechanism for protecting existing representations from interference during new learning.

NIH R01EY019466, R01EY027841 and NSF-BSF BCS2241417

Talk 5, 6:15 pm, 35.15

Neurochemical and BOLD signal change in hMT+ and attentional networks associated with vision restoration after occipital stroke

Hanna Willis^1,2, Lucy Starling², Isobel Rout², Brendan Sargent², Aaron Kay², Rebecca Millington-Truby², I. Betina Ip², Matthew Cavanaugh³, Sara Ajina⁴, Krystel Huxlin³, Marco Tamietto^5,6, Holly Bridge²; ¹École normale supérieure, CNRS, ²University of Oxford, ³University of Rochester, ⁴UCL, ⁵Tilburg University, ⁶University of Torino

BACKGROUND: Damage to early visual cortex after occipital stroke causes visual field loss. Training within the blind field can enhance and restore perception. We used multimodal MRI before and after training to investigate whether changes in the function of visual motion area hMT+ and higher-level attentional networks supports observed improvements in vision and awareness. METHODS: Eighteen participants (6 female; aged 24–74 years; >6 months post-stroke) completed at least six months of at-home rehabilitation using a two-alternative-forced-choice motion discrimination and integration task. All participants visited the research centre for two visits (pre-training and post-training). Twelve participants also returned after a further three months without additional training (follow-up). At each research visit, participants completed behavioural tasks and an MRI. To assess transfer of learning, they completed a Gabor detection task in the trained locations (50% and 100% contrast; σ=0.8°; spatial frequency=1c/°). GABA and glutamate concentrations were measured in ipsilesional hMT+ and a control voxel in sensorimotor cortex. fMRI assessed signal change in hMT+ and the rest of the brain during passive viewing of high-contrast Gabor stimuli in the blind field. RESULTS: Participants showed significant improvement in motion discrimination that generalised to Gabor detection (t(16)=–2.19; p=0.044) and awareness (t(16)=–2.66; p=0.017). These changes in detection (paired t-test: t(10)=-1.13; p=0.28) and awareness (paired t-test: t(10)=-0.76; p=0.46) persisted at a three-month follow up visit. Reduced GABA (r=–0.60; p=0.02) and glutamate (r=–0.58; p=0.02) in ipsilesional hMT+ were linked to improved detection. Increased activity in hMT+ and dorsolateral prefrontal cortex correlated with improved detection, while awareness changes were linked to signal changes in contralesional prefrontal cortex (area 46) and inferior parietal lobule. CONCLUSIONS: Extensive training improves visual detection and awareness for both trained and untrained tasks. Gains are supported by selective changes in brain regions involved in motion perception and attention, suggesting that a broad network supports recovery.

H.E.W was supported by a DPhil scholarship from the Medical Research Council United Kingdom (MR/N013468/ 1), a Waverley Scholarship from The Queen’s College, Oxford, UK, and a European Research Council (ERC) Horizon 2020 Research and Innovation Programme Grant No 948366-HOPLA.

Talk 6, 6:30 pm, 35.16

Trial-by-trial Evolution of Visuocortical Tuning and Functional Connectivity During Aversive Conditioning: A Simultaneous EEG-fMRI Study

Laura Ahumada¹ (), Andrew Farkas¹, Faith E. Gilbert¹, Hanna M. Engle¹, Judith Escobar¹, Jourdan J. Pouliot¹, Katherine McCain¹, Arash Mirifar¹, Mingzhou Ding¹, Andreas Keil¹; ¹University of Florida

Responses in visuocortical neurons reflect learning and experience about the visual world. The time course and large-scale mechanisms underlying learning in the human visual cortex, however, remain elusive due to a range of methodological limitations. In the present study, simultaneous EEG and fMRI signals were recorded to address some of these limitations. A sample of 25 healthy observers participated in an aversive conditioning paradigm, where one Gabor patch (conditioned stimulus, CS+) was paired with an aversive electrical pulse (unconditioned stimulus). Three additional Gabor patches with different orientations served as generalization stimuli (GS1, GS2, and GS3) and were never paired with the electric pulse. The patches flickered at 15 Hz to generate steady-state visual potentials (ssVEPs). The experimental session consisted of three phases: habituation, acquisition, and extinction. Tuning in the visual area was measured by modeling, at the single-trial level, the ssVEP amplitude across the four Gabor patches, using a Different-of-Gaussian function (Ricker model). Results showed that visuocortical tuning resembled a generalization trend from the CS+ to the most similar patches in the initial part of acquisition, while from middle acquisition through extinction, the tuning narrowed, representing sharpened tuning of the ssVEP amplitude relative to the CS+. This electrophysiological sharpening predicted the increment in BOLD activity in early visual areas (V1/V2). On the other hand, inter-site phase locking of the ssVEP was strong between occipital and parietal sensors, and between occipital and frontal sensors, particularly at the beginning of acquisition. EEG-BOLD integration showed this early strong connectivity between V1/V2 and areas like the superior frontal cortex, orbitofrontal cortex, and the ventral visual pathway. Overall, these findings are consistent with the notion that changes in tuning to a conditioned visual stimulus are supported by extra-striate inputs, while the late narrowing of the tuning bandwidth in V1/V2 involves local suppressive activity.

Talk 7, 6:45 pm, 35.17

The complex role of mind wandering in implicit statistical learning

Teodóra Vékony¹ (), Bianka Brezóczki^2,3,4, Bence C. Farkas^5,6,7, Gábor Csifcsák⁸, Péter Simor^9,10, Dezső Németh^1,11,12; ¹Gran Canaria Cognitive Research Center, Atlántico Medio University, ²Doctoral School of Psychology, Eötvös Loránd University, ³Institute of Psychology, Eötvös Loránd University, ⁴Institute of Cognitive Neuroscience and Psychology, HUN-REN Research Centre for Natural Sciences, ⁵UVSQ, Inserm, Centre de Recherche en Epidémiologie et Santé des Populations, Université Paris-Saclay, ⁶Pôle Pilotage activités et Projets, Direction Générale Adjointe Enfance Famille Santé (DGAEFS), ⁷LNC2, Département d’études Cognitives, École Normale Supérieure, INSERM, PSL Research University, ⁸Department of Psychology, UiT The Arctic University of Norway, ⁹Institute of Behavioral Sciences, Semmelweis University, ¹⁰IMéRA Institute for Advanced Studies of Aix-Marseille University, ¹¹Centre de Recherche en Neurosciences de Lyon CRNL U1028 UMR5292, INSERM, CRNS, Université Claude Bernard Lyon 1, ¹²BML-NAP Research Group, Institute of Psychology, Eötvös Loránd University & Institute of Cognitive Neuroscience and Psychology, HUN-REN Research Centre for Natural Sciences

Mind wandering (MW) is often regarded as a state of mind associated with impaired cognitive performance, yet recent studies suggest it plays a surprising, adaptive role in statistical learning (the implicit acquisition of environmental regularities). Our prior work established that periods of MW are associated with enhanced statistical learning outcomes and identified a plausible neural mechanism, demonstrating that both efficient statistical learning and increased MW are linked to local slow waves in the waking brain. Building upon this foundation, our latest study directly addresses the complex interplay between MW, inhibitory control, and statistical learning. Healthy young participants completed an online experiment that combined simultaneous probabilistic learning and response inhibition (a go/no-go task). Throughout the task, participants provided continuous reports on their attentional state, indicating whether their focus was on-task or MW. We report a clear functional dissociation: increased MW simultaneously impaired inhibitory control (evidenced by reduced response suppression) but remarkably facilitated statistical learning (improved pattern detection in the visuomotor sequence). Crucially, we show that inhibitory control performance significantly modulates the relationship between MW and statistical learning efficiency. These findings provide a compelling framework linking off-task thought to the dynamics of predictive cognition and waking neural dynamics, thereby highlighting MW's essential and multifaceted role in optimizing cognitive resource allocation.

This research was funded by the Spanish Ministry of Science, Innovation and Universities (MICIU) – State Research Agency (AEI), grant number PID2024-160183NA-I00.

Talk 8, 7:00 pm, 35.18

Gaze-contingent scotoma awareness training in patients with central vision loss

Marcello Maniglia¹ (), Jason Vice², Elliot Maxwell², Kristina Visscher², Aaron Seitz³; ¹Rochester Institute of Technology, ²University of Alabama at Birmingham, ³Northeastern University

Macular degeneration (MD) is the leading cause of central vision loss in the Western world. With the aging population worldwide, its prevalence is expected to increase, taking a toll on individuals’ quality of life, their caregivers’ burden, and the healthcare system. Patients often develop spontaneous compensatory strategies, most commonly adopting a preferred retinal locus (PRL), a peripheral region used as a new fixation point and as a new oculomotor reference. However, PRL development is often inefficient, limited by patients’ poor awareness of their scotoma’s size, shape and location, as well as our lack of mechanistic understanding of PRL development. Studies using simulated scotomas, a framework for lab-based investigations of central vision loss, show that scotoma visibility and contour characteristics can accelerate oculomotor adaptation, suggesting that awareness of visual field loss may promote PRL development. To test the translational potential of this hypothesis, we designed a visual training paradigm to enhance patients’ awareness of their blind region using eye-tracker–guided gaze-contingent displays. In Experiment 1, a group of patients with MD trained on a visual discrimination task while a gaze-contingent display drew the perimeter of their individual scotoma on a computer screen in real time. In Experiment 2, a new cohort of patients with MD trained on a modified version incorporating a trained retinal locus (TRL) component, intended to guide PRL development to a specific peripheral location chosen based on their individual structural and functional fundus image exams. A control group of patients was trained on the same visual task without the gaze-contingent component. Results reveal substantial heterogeneity in response to these interventions. We present this variability in terms of changes in oculomotor behavior and visual performance across training conditions and discuss its implications for laboratory-inspired rehabilitation protocols

NEI R21EY033623-01