Research
I am fascinated by the constructive nature of perception. I think that transforming imprecise, discrete and abstract, sensory inputs into a continuous, meaningful perception is a tour the force. Following R.L. Gregory, I believe perception is essentially an hypothesis generation realised using knowledge derived from the world's statistical structure, from basic principles of perceptual organization to higher-level expectations about what objects are and how they behave. I think that the sensory cortices play a central role in that hypothesis generation
Spatiotemporal context
An example of low level rules of perception is input grouping for object formation. In the rat somatosensory cortex, using extracellular electrophysiology and optical imaging, we showed that temporally and spatially related tactile inputs can become cortically merged in the somatotopic map, forming unified representations rather than remaining independent signals (Corbo et al. 2018). This cortical merging provides a substrate for tactile grouping, suggesting that perceptual organization can be embedded directly within early sensory circuits.
Experience-driven reshaping of representations
Where cortical merging reflects basic structural regularities of the world (inputs that co-occur likely belong to the same object), learning introduces higher-order constraints derived from task demands and experience. In mouse V1, we showed that training for orientation discrimination induces a systematic distortion of orientation representation: neurons are selectively suppressed around task-relevant stimuli in the feature encoding space (Corbo et al. 2022). This reshapes the population activity into category-like domains whose activation predicts behavioral performance on a probabilistic level (Corbo et al. 2025), and promotes generalization of the stimuli ressembling the task's targets. Ongoing work aims at elucidating the timecourse and circuit mechanism of the training-induced orientation space modulation we uncovered.
Geometry of V1 representations
The computations that make up perception can be seen as an array of representations and readouts. Building on the results obtained by investigating feature encoding, we question the potential impact that learning-induced changes in V1 have on their readout. To do so, we investigate sensory representations through the lens of high-dimensional geometry. By quantifying object manifold structure, separability and dimensionality in population activity, we aim to understand how learning alters the geometry of early cortical representations favorably for the behavioral needs. This line of inquiry is carried out in collaboration with Leyla R. Caglar, at the Windreich Institute for AI and Human Health in Mount Sinai.
From early representations to perceptual decisions
Ultimately, my goal is to bridge sensory encoding and perceptual decision-making. I seek to identify where behaviorally predictive population variables are read out and how downstream circuits transform these structured representations into categorical percepts and choices. By tracing this transformation across cortical networks, I aim to understand the neural computations that convert early representations into perception. The first step of this research program is a brain-wide functional screening of areas involved in perceptual decisions using immediate early genes expression, in collaboration with Romain Durand de Cuttolli at the Icahn School of Medicine at Mount Sinai.
Publications
- Caglar, Leyla R; Erkat, O Batuhan; Polack, Pierre-Olivier; Corbo, Julien (2026). Learning induces geometric reorganization in the primary visual cortex In prep - presented at CCN 2025 and Cosyne 2026
- Meszéna, Balázs; Murray, Keith T; Corbo, Julien; Erkat, O Batuhan; Hajnal, Márton A; Polack, Pierre-Olivier; Orbán, Gergő (2026). TAVAE: A VAE with Adaptable Priors Explains Contextual Modulation in the Visual Cortex. arXiv preprint arXiv:2602.11956. [link]
- Corbo, Julien; Erkat, O Batuhan; McClure Jr, John; Khdour, Hussein; Polack, Pierre-Olivier (2025). Discretized representations in V1 predict suboptimal orientation discrimination. Nature Communications, 16(1), 41. [link]
- Corbo, Julien; McClure, John P; Erkat, O Batuhan; Polack, Pierre-Olivier (2022). Dynamic distortion of orientation representation after learning in the mouse primary visual cortex. Journal of Neuroscience, 42(21), 4311–4325. [link]
- McClure Jr, John P; Erkat, O Batuhan; Corbo, Julien; Polack, Pierre-Olivier (2022). Estimating how sounds modulate orientation representation in the primary visual cortex using shallow neural networks. Frontiers in Systems Neuroscience, 16, 869705. [link]
- Caron-Guyon, Jeanne; Corbo, Julien; Zennou-Azogui, Yoh’i; Xerri, Christian; Kavounoudias, Anne; Catz, Nicolas (2020). Neuronal encoding of multisensory motion features in the rat associative parietal cortex. Cerebral Cortex, 30(10), 5372–5386. [link]
- Corbo, Julien; Zennou-Azogui, Yoh’I; Xerri, Christian; Catz, Nicolas (2018). Cortical merging in S1 as a substrate for tactile input grouping. eNeuro, 5(1). [link]
Links
Email | GitHub | Google Scholar | BlueSky