Sie sind herzlich zu einem spannenden IBIO-Forschungsvortrag eingeladen:
Wann: Donnerstag, 21. August, um 11 Uhr im IBIO (Haus 91), Raum 001
Titel: Population Orientation Coding in Macaque V1, aufgezeigt durch Zwei-Photonen-Kalzium-Imaging und Transformator-Modellierung
Wer: Prof. Cong Yu, Fakultät für Psychologie und Kognitionswissenschaften, Peking-Universität
Bitte teilen Sie Linda Schimrock mit, wenn Sie Cong Yu am 21. August treffen möchten. Sie sind auch herzlich eingeladen, an diesem Abend mit uns zu Abend zu essen – bitte geben Frau Schimrock Bescheid.
Talk summary: l will present recent studies combining two-photon calcium imaging and machine learning to investigate population orientation coding and related issues in macaque V1.
The first study demonstrates that the orthogonality of stimulus features, including orientation, spatial frequency, and ocularity, is not represented in the relationships among functional maps as originally hypothesized by Hubel and Wiesel. The intersection angles among these maps at cellular resolution are not orthogonal, but evenly distributed. However, the orthogonality is represented by neural geometries of population responses to these features, which are orthogonal to each other in a PCA space.
The second study develops a transformer model that reveals V1 orientation decoding relies on a sparse set of “hub” neurons receiving strong self-attention from other neurons.
These neurons are tuned to stimulus orientation and the tuning properties are enhanced by self-attention. Moreover, knocking out hub neurons does not impair orientation decoding, as the transformer recruits substitute hubs that are largely non-contributing prior to knockout. These modeling results point to a combination of redundant encoding and sparse decoding of orientation information.
The third study demonstrates ocular-dominance-dependent continuous flashing suppression (CFS) of V1 responses. Classification and transformer analyses further indicate that, while CFS-suppressed orientation signals can still support low-level coarse orientation discrimination, it may not suffice for object-level visual and cognitive processing. These results may help resolve the long-standing debate regarding the nature of subconscious visual processing.
Finally, time permitting, I will preview my upcoming ECVP talk. We use a transformer with a cross-attention mechanism to analyze orientation biases in population responses induced by preceding trials. The results suggest that the repulsion effect due to orientation adaptation and the attractive serial dependence effect are likely based on the readout of distinct V1 response patterns.