The representational architecture of fast learning through abstraction
Part of the perceptual knowledge we acquire in everyday life does not rely on repeated exposure or training: a single significant event can induce robust changes on brain activity and behavior. Such one-shot perceptual learning emerges during development in parallel to incremental learning and plays a crucial role when evidence is scarce or ambiguous. However, while most cognitive neuroscientists agree on its relevance to our adaptation abilities, the neural and cognitive computations driving one-shot learning remain largely unknown.
Predictive processing accounts have been very influential in the study of one-shot perceptual learning. From this perspective, significant perceptual episodes create lingering traces in the brain, reflecting internal models of the external world, or priors. Priors are proposed to then inform downstream brain regions about the causes of sensory input. Although this framework offers an intuitive explanation of the procedures that might support one-shot perceptual learning, we currently lack an optimal description of the precise nature of priors and the information they contain.
The overall aim of FLARE is to provide fundamental insights into how internal models of single perceptual events are instantiated in patterns of brain activity. FLARE constitutes a novel approach combining the PI’s theoretical background and expertise in cutting-edge neuroimaging methods that will allow us to pursue the overall objective across two experimental series.
The project will tackle two major open questions: 1) What is the content of priors of single perceptual events across the brain? And 2) To what extent does one-shot perceptual learning rely on sensory-specific vs. abstract priors of the episode? For both goals, we will employ a combination of tailored behavioral tasks, computational modeling, and neuroimaging methods.
Funded by the Spanish Ministry of Science — PID2023-149428NB-I00
2025
-
Determinants of Visual Ambiguity Resolution
Juan Linde-Domingo*, Javier Ortiz-Tudela*, Johannah Voeller, Martin N. Hebart, and Carlos Gonzalez-Garcia
bioRxiv, 2025
Visual inputs during natural perception are highly ambiguous: objects are frequently occluded, lighting conditions vary, and object identification depends significantly on prior experiences. However, why do certain images remain unidentifiable while others can be recognized immediately, and what visual features drive subjective clarification? To address these critical questions, we developed a unique dataset of 1,854 ambiguous images and collected more than 100,000 participant ratings evaluating their identifiability before and after seeing undistorted versions of the images. Relating the representations of a brain-inspired neural network model in response to our images with human ratings, we show that subjective identification depends largely on the extent to which higher-level visual features from the original images are preserved in their ambiguous counterparts. Notably, the predominance of higher-level features over lower-level ones softens after participants disambiguate the images. In line with these results, an image-level regression analysis showed that the subjective identification of ambiguous images was best explained by high-level visual dimensions. Moreover, we found that the process of ambiguity resolution was accompanied by a notable decrease in semantic distance and a greater consistency in object naming among participants. However, the relationship between information gained after disambiguation and subjective identification was non-linear, indicating that acquiring more information does not necessarily enhance subjective clarity. Instead, we observed a U-shaped relationship, suggesting that subjective identification improves when the acquired information either strongly matches or mismatches prior predictions. Collectively, these findings highlight fundamental principles underlying the mapping between human visual perception and memory, advancing our understanding on how we resolve ambiguity and extract meaning from incomplete visual information.