J. Brendan Ritchie

Reaction time for object categorization is predicted by representational distance

How does the brain translate an internal representation of an object into a decision about the objects category? Recent studies have uncovered the structure of object representations in inferior temporal cortex (IT) using multivariate pattern analysis methods. These studies have shown that representations of individual object exemplars in IT occupy distinct locations in a high-dimensional activation space, with object exemplar representations clustering into distinguishable regions based on category (e.g., animate vs. inanimate objects). In this study, we hypothesized that a representational boundary between category representations in this activation space also constitutes a decision boundary for categorization. We show that behavioral RTs for categorizing objects are well described by our activation space hypothesis. Interpreted in terms of classical and contemporary models of decision-making, our results suggest that the process of settling on an internal representation of a stimulus is itself partially constitutive of decision-making for object categorization.

Emerging Object Representations in the Visual System Predict Reaction Times for Categorization

Recognizing an object takes just a fraction of a second, less than the blink of an eye. Applying multivariate pattern analysis, or “brain decoding”, methods to magnetoencephalography (MEG) data has allowed researchers to characterize, in high temporal resolution, the emerging representation of object categories that underlie our capacity for rapid recognition. Shortly after stimulus onset, object exemplars cluster by category in a high-dimensional activation space in the brain. In this emerging activation space, the decodability of exemplar category varies over time, reflecting the brain’s transformation of visual inputs into coherent category representations. How do these emerging representations relate to categorization behavior? Recently it has been proposed that the distance of an exemplar representation from a categorical boundary in an activation space is critical for perceptual decision-making, and that reaction times should therefore correlate with distance from the boundary. The predictions of this distance hypothesis have been born out in human inferior temporal cortex (IT), an area of the brain crucial for the representation of object categories. When viewed in the context of a time varying neural signal, the optimal time to “read out” category information is when category representations in the brain are most decodable. Here, we show that the distance from a decision boundary through activation space, as measured using MEG decoding methods, correlates with reaction times for visual categorization during the period of peak decodability. Our results suggest that the brain begins to read out information about exemplar category at the optimal time for use in choice behaviour, and support the hypothesis that the structure of the representation for objects in the visual system is partially constitutive of the decision process in recognition.

Neural Decoding and “Inner” Psychophysics: A Distance-to-Bound Approach for Linking Mind, Brain, and Behavior

A fundamental challenge for cognitive neuroscience is characterizing how the primitives of psychological theory are neurally implemented. Attempts to meet this challenge are a manifestation of what Fechner called “inner” psychophysics: the theory of the precise mapping between mental quantities and the brain. In his own time, inner psychophysics remained an unrealized ambition for Fechner. We suggest that, today, multivariate pattern analysis (MVPA), or neural “decoding,” methods provide a promising starting point for developing an inner psychophysics. A cornerstone of these methods are simple linear classifiers applied to neural activity in high-dimensional activation spaces. We describe an approach to inner psychophysics based on the shared architecture of linear classifiers and observers under decision boundary models such as signal detection theory. Under this approach, distance from a decision boundary through activation space, as estimated by linear classifiers, can be used to predict reaction time in accordance with signal detection theory, and distance-to-bound models of reaction time. Our “neural distance-to-bound” approach is potentially quite general, and simple to implement. Furthermore, our recent work on visual object recognition suggests it is empirically viable. We believe the approach constitutes an important step along the path to an inner psychophysics that links mind, brain, and behavior.

Decoding the Brain: Neural Representation and the Limits of Multivariate Pattern Analysis in Cognitive Neuroscience

Since its introduction, multivariate pattern analysis (MVPA), or ‘neural decoding’, has transformed the field of cognitive neuroscience. Underlying its influence is a crucial inference, which we call the decoder’s dictum: if information can be decoded from patterns of neural activity, then this provides strong evidence about what information those patterns represent. Although the dictum is a widely held and well-motivated principle in decoding research, it has received scant philosophical attention. We critically evaluate the dictum, arguing that it is false: decodability is a poor guide for revealing the content of neural representations. However, we also suggest how the dictum can be improved on, in order to better justify inferences about neural representation using MVPA. 1. Introduction2. A Brief Primer on Neural Decoding: Methods, Application, and Interpretation 2.1. What is multivariate pattern analysis?2.2. The informational benefits of multivariate pattern analysis3. Why the Decoder’s Dictum Is False 3.1. We don’t know what information is decoded3.2. The theoretical basis for the dictum3.3. Undermining the theoretical basis4. Objections and Replies 4.1. Does anyone really believe the dictum?4.2. Good decoding is not enough4.3. Predicting behaviour is not enough5. Moving beyond the Dictum6. Conclusion Introduction A Brief Primer on Neural Decoding: Methods, Application, and Interpretation 2.1. What is multivariate pattern analysis?2.2. The informational benefits of multivariate pattern analysis What is multivariate pattern analysis? The informational benefits of multivariate pattern analysis Why the Decoder’s Dictum Is False 3.1. We don’t know what information is decoded3.2. The theoretical basis for the dictum3.3. Undermining the theoretical basis We don’t know what information is decoded The theoretical basis for the dictum Undermining the theoretical basis Objections and Replies 4.1. Does anyone really believe the dictum?4.2. Good decoding is not enough4.3. Predicting behaviour is not enough Does anyone really believe the dictum? Good decoding is not enough Predicting behaviour is not enough Moving beyond the Dictum Conclusion

Edge-Related Activity Is Not Necessary to Explain Orientation Decoding in Human Visual Cortex.

Multivariate pattern analysis is a powerful technique; however, a significant theoretical limitation in neuroscience is the ambiguity in interpreting the source of decodable information used by classifiers. This is exemplified by the continued controversy over the source of orientation decoding from fMRI responses in human V1. Recently Carlson (2014) identified a potential source of decodable information by modeling voxel responses based on the Hubel and Wiesel (1972) ice-cube model of visual cortex. The model revealed that activity associated with the edges of gratings covaries with orientation and could potentially be used to discriminate orientation. Here we empirically evaluate whether “edge-related activity” underlies orientation decoding from patterns of BOLD response in human V1. First, we systematically mapped classifier performance as a function of stimulus location using population receptive field modeling to isolate each voxels overlap with a large annular grating stimulus. Orientation was decodable across the stimulus; however, peak decoding performance occurred for voxels with receptive fields closer to the fovea and overlapping with the inner edge. Critically, we did not observe the expected second peak in decoding performance at the outer stimulus edge as predicted by the edge account. Second, we evaluated whether voxels that contribute most to classifier performance have receptive fields that cluster in cortical regions corresponding to the retinotopic location of the stimulus edge. Instead, we find the distribution of highly weighted voxels to be approximately random, with a modest bias toward more foveal voxels. Our results demonstrate that edge-related activity is likely not necessary for orientation decoding. SIGNIFICANCE STATEMENT A significant theoretical limitation of multivariate pattern analysis in neuroscience is the ambiguity in interpreting the source of decodable information used by classifiers. For example, orientation can be decoded from BOLD activation patterns in human V1, even though orientation columns are at a finer spatial scale than 3T fMRI. Consequently, the source of decodable information remains controversial. Here we test the proposal that information related to the stimulus edges underlies orientation decoding. We map voxel population receptive fields in V1 and evaluate orientation decoding performance as a function of stimulus location in retinotopic cortex. We find orientation is decodable from voxels whose receptive fields do not overlap with the stimulus edges, suggesting edge-related activity does not substantially drive orientation decoding.

Avoiding illusory effects in representational similarity analysis: What (not) to do with the diagonal

Representational similarity analysis (RSA) is an important part of the methodological toolkit in neuroimaging research. The focus of the approach is the construction of representational dissimilarity matrices (RDMs), which provide a single format for making comparisons between different neural data types, computational models, and behavior. We highlight two issues for the construction and comparison of RDMs. First, the diagonal values of RDMs, which should reflect within condition reliability of neural patterns, are typically not estimated in RSA. However, without such an estimate, one lacks a measure of the reliability of an RDM as a whole. Thus, when carrying out RSA, one should calculate the diagonal values of RDMs and not take them for granted. Second, although diagonal values of a correlation matrix can be used to estimate the reliability of neural patterns, these values must nonetheless be excluded when comparing RDMs. Via a simple simulation we show that inclusion of these values can generate convincing looking, but entirely illusory, correlations between independent and entirely unrelated data sets. Both of these points are further illustrated by a critical discussion of Coggan et al. (2016), who investigated the extent to which category-selectivity in the ventral temporal cortex can be accounted for by low-level image properties of visual object stimuli. We observe that their results may depend on the improper inclusion of diagonal values in their analysis.

Ghosts in machine learning for cognitive neuroscience: Moving from data to theory

ABSTRACT The application of machine learning methods to neuroimaging data has fundamentally altered the field of cognitive neuroscience. Future progress in understanding brain function using these methods will require addressing a number of key methodological and interpretive challenges. Because these challenges often remain unseen and metaphorically “haunt” our efforts to use these methods to understand the brain, we refer to them as “ghosts”. In this paper, we describe three such ghosts, situate them within a more general framework from philosophy of science, and then describe steps to address them. The first ghost arises from difficulties in determining what information machine learning classifiers use for decoding. The second ghost arises from the interplay of experimental design and the structure of information in the brain – that is, our methods embody implicit assumptions about information processing in the brain, and it is often difficult to determine if those assumptions are satisfied. The third ghost emerges from our limited ability to distinguish information that is merely decodable from the brain from information that is represented and used by the brain. Each of the three ghosts place limits on the interpretability of decoding research in cognitive neuroscience. There are no easy solutions, but facing these issues squarely will provide a clearer path to understanding the nature of representation and computation in the human brain. HIGHLIGHTSProvides a philosophical framework for thinking about applications of machine learning to cognitive neuroscience datasets.Discussion of current challenges for neural decoding research.Gives suggestions about how to address contemporary challenges in decoding research.

On the partnership between neural representations of object categories and visual features in the ventral visual pathway

A dominant view in the cognitive neuroscience of object vision is that regions of the ventral visual pathway exhibit some degree of category selectivity. However, recent findings obtained with multivariate pattern analyses (MVPA) suggest that apparent category selectivity in these regions is dependent on more basic visual features of stimuli. In which case a rethinking of the function and organization of the ventral pathway may be in order. We suggest that addressing this issue of functional specificity requires clear coding hypotheses, about object category and visual features, which make contrasting predictions about neuroimaging results in ventral pathway regions. One way to differentiate between categorical and featural coding hypotheses is to test for residual categorical effects: effects of category selectivity that cannot be accounted for by visual features of stimuli. A strong method for testing these effects, we argue, is to make object category and target visual features orthogonal in stimulus design. Recent studies that adopt this approach support a feature-based categorical coding hypothesis according to which regions of the ventral stream do indeed code for object category, but in a format at least partially based on the visual features of stimuli.

Tool Integration and Dynamic Touch

Tool integration is important to how people perceive their bodies in relation to the objects (“tools”) that they use to manipulate the world around them. Better under-standing the nature of this integration has the potential for clinical benefits. For example, better understanding how people integrate representations into the body schema may allow for the development of more effective prosthetics for missing limbs. Several results have been interpreted as supporting the existence of tool integra-tion (see Maravita & Iriki, 2004, for an early review). In Carlson, Alvarez, Wu, and Verstraten (2010), a previous study using positive afterimages, evidence was presented from which three conclusions were drawn: first, that tool integration for simple, grasped objects was rapid; second, that this assimilation occurred without any prior training; and third, that the same assimilation did not

Massive modularity is consistent with most forms of neural reuse

Anderson claims that the hypothesis of massive neural reuse is inconsistent with massive mental modularity. But much depends upon how each thesis is understood. We suggest that the thesis of massive modularity presented in Carruthers (2006) is consistent with the forms of neural reuse that are actually supported by the data cited, while being inconsistent with a stronger version of reuse that Anderson seems to support.