Poster Presentation

Poster D in Poster Session D: Wednesday, August 5, 2:00 – 3:45 pm, Kimmel Center, Shorin & Rosenthal Rooms

Dynamics of object coding in the inferotemporal cortex

Conor McGrory¹, Janis Hesse¹, Tatiana A Engel¹, Doris Tsao²; ¹Princeton University, ²University of California, Berkeley

Presenter: Conor McGrory

A prevailing framework in visual neuroscience holds that objects are encoded in stable neural activity patterns within the inferotemporal (IT) cortex, where each image evokes a distinct population response representing its identity. This view is supported by the high decoding accuracy of object identity from IT population activity averaged over the viewing duration and by the success of hierarchical feedforward models of the ventral stream in predicting time-averaged IT responses to images. However, object recognition occurs within just a hundred milliseconds, too brief for multi-trial averaging. Yet, the IT object code has not been studied on such short timescales, and it remains unknown whether it conforms to the standard framework within single trials. Using NHP Neuropixels probes, we recorded spiking activity from face-selective regions in the macaque IT cortex, while the animals viewed a variety of visual stimuli. Within single trials, population spiking activity varied dynamically, exhibiting synchronous fluctuations between phases of vigorous (On) and faint (Off) firing, with rates changing by up to twofold between On and Off phases. These fluctuations occurred on a timescale of ∼100 ms, similar to On-Off dynamics previously described in V1 and V4. Their statistics remained invariant across different visual experiences, indicating that On-Off fluctuations were not caused by changing visual inputs. The same linear decoder accurately discriminated visual stimuli during both On and Off phases, showing that rapid fluctuations in population activity did not alter the object code. The geometry of population responses was consistent with a model in which On–Off activity multiplicatively scales object-coding vectors relative to a shared baseline firing rate, which explains the temporal stability of object representations within single trials. Our results extend the standard framework of IT coding to shorter timescales, demonstrating how stable object representations persist despite dynamically changing firing rates within single trials.

Topic Area: Methods, Tools, Theory & Neural Coding