Abstract
Population responses to object images in inferior temporal cortex (IT) are dynamic. However, it is not well understood whether the dynamics of IT predominantly reflect encoding of external variables or internal population dynamics. Here we analyze single-unit recordings from monkey IT to investigate the predominant source of the time-varying population response. Single-unit activity was recorded from 989 neurons in the inferior bank of the superior temporal sulcus of two adult macaques passively viewing 100 grayscale object images from five categories (faces, body parts, fruit, objects, indoor scenes) presented for 300 ms at fixation at 5 degrees visual angle (Bell et al., 2011). We analyzed activity patterns across visually responsive neurons, from 100 ms before to 700 ms after stimulus onset, with population tensor analysis (Seely et al. 2016). Tensor analysis treats the population response as a third-order tensor indexed by neuron, condition, and time. The tensor is modeled as a linear combination of either condition-by-time components that are shared across neurons (basis neurons) or neuron-by-time components that are shared across conditions (basis conditions). If the tensor is reconstructed most accurately from the basis neurons (conditions), this indicates that population dynamics are predominantly driven by external variables (internal dynamics). For both monkeys, the IT population response was reconstructed most accurately from the basis neurons. The advantage of basis neurons over basis conditions increased with the number of leading components used for reconstruction (1–25). We matched the number of conditions and neurons before analysis by selecting random samples of 100 neurons. Results generalized to bootstrap resamplings of the condition set. These preliminary results suggest that, for passive viewing of isolated object images, responses in monkey IT predominantly reflect encoding of external variables (tuning). Internal population dynamics may play a larger role in challenging visual tasks that rely on recurrent processing.
Acknowledgement: National Institutes of Health, Medical Research Council