Representational dynamics of object vision: the first 1000 ms.

By Thomas A. Carlson, David A. Tovar, Arjen Alink, Nikolaus Kriegeskorte. 2013

420 citations recorded by the source corpus.

Human object recognition is remarkably efficient. In recent years, significant advancements have been made in our understanding of how the brain represents visual objects and organizes them into categories. Recent studies using pattern analyses methods have characterized a representational space of objects in human and primate inferior temporal cortex in which object exemplars are discriminable and cluster according to category (e.g., faces and bodies). In the present study we examined how category structure in object representations emerges in the first 1000 ms of visual processing. In the study, participants viewed 24 object exemplars with a planned categorical structure comprised of four levels ranging from highly specific (individual exemplars) to highly abstract (animate vs. inanimate), while their brain activity was recorded with magnetoencephalography (MEG). We used a sliding time window decoding approach to decode the exemplar and the exemplar's category that participants were viewing on a moment-to-moment basis. We found exemplar and category membership could be decoded from the neuromagnetic recordings shortly after stimulus onset (<100 ms) with peak decodability following thereafter. Latencies for peak decodability varied systematically with the level of category abstraction with more abstract categories emerging later, indicating that the brain hierarchically constructs category representations. In addition, we examined the stationarity of patterns of activity in the brain that encode object category information and show these patterns vary over time, suggesting the brain might use flexible time varying codes to represent visual object categories.

Subfields: Cognitive Neuroscience, Computer Vision and Pattern Recognition, Representation & Self-Supervised Learning

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Cognitive Neuroscience

In Computer Vision and Pattern Recognition

In Representation & Self-Supervised Learning

Citations in the corpus

Towards Conceptual Compression (2016, background): “The questions then are: can such a framework indeed discover such representations both in principle and in practice, are such networks powerful enough for modeling real data, and what techniques one needs to make it work well.”
The dynamics of invariant object recognition in the human visual system. (2014, background): “A recent study by the same group shows that more abstract categorization has a longer decoding latency (Carlson et al. 2013), supporting this explanation.”
A Toolbox for Representational Similarity Analysis (2014, background): “Such analyses can reveal the emergence and dynamics of representational geometries over the course of tens to hundreds of milliseconds after stimulus onset, reflecting recurrent neuronal computations [25,42].”
Cogni-Net: Cognitive Feature Learning Through Deep Visual Perception (2018, background): “Works such as [1, 2, 3, 4, 5, 6, 7] show that brain signals evoked by a specific stimulus contain information about image categories, with [8] proving that they can be exploited to discriminate between image classes.”
Deep neural networks: a new framework for modelling biological vision and brain information processing (2015, background): “…contribute to the automatic rapid interpretation of even static visual images (Sugase et al. 1999, Brincat & Connor 2006, Freiwald & Tsao 2010, Carlson et al. 2013, Cichy et al. 2014; Tang et al. 2014a), leading to the emergence of representations that clearly distinguish particular objects…”
Representational models: A common framework for understanding encoding, pattern-component, and representational-similarity analysis (2017, methodology): “16 using a sliding window over the time course of the trial [37-39], or by “stacking” the time series and 314”
Deep Supervised, but Not Unsupervised, Models May Explain IT Cortical Representation (2014, background): “Indeed, recent human magnetoencephalography studies using the same stimulus set [11,12] suggest that the major categorical divisions take slightly longer to emerge than a purely feedforward account would predict.”
Decoding Dynamic Brain Patterns from Evoked Responses: A Tutorial on Multivariate Pattern Analysis Applied to Time Series Neuroimaging Data (2016, methodology): “The statistical approach used in our example analysis is common in the literature (e.g., Ritchie et al., 2015; Carlson, Tovar, et al., 2013) and was chosen for its simplicity; however, there are several alternative methods that are also valid.”
Recurrence is required to capture the representational dynamics of the human visual system (2019, positive): “This stimulus set was used across multiple studies and laboratories to collect human fMRI (20), and MEG data (16, 43), human perceptual similarity judgments (44), macaque single cell data (27), and was used in previous investigations of deep neural network models (25, 45).”