Talk given at the EMHFC Conference

I gave this talk at the European Meeting on Human Fear Conditioning about "Temporal Dynamics of Aversive Generalization".

Abstract:
Temporal Dynamics of Aversive Learning and Generalization in Amygdala
The amygdala is thought to orchestrate coordinated bodily responses important for the survival
of the organism during threatening situations. However its contribution to the generalization of
previously learnt aversive associations is not well understood. As amygdala responses in the
context of fear conditioning are temporally phasic and adapt quickly, we designed a new
paradigm to investigate its temporal dynamics during fear generalization. We used faces that
formed a perceptual circular similarity continuum, allowing us to gather two-sided generalization
gradients. While one face predicted an aversive outcome (UCS), the most dissimilar face was
kept neutral. Importantly, participants were compelled to learn these associations throughout the
fMRI recording which they started naive, allowing us to collect temporally resolved
generalization gradients for BOLD and skin-conductance responses. Following fMRI, we
evaluated subjective likelihood for single faces to be associated with UCS, and complemented
these with behavioral measurements using eye-movement recordings to assess how the saliency
associated with faces were modified. Aversive generalization in the amygdala emerged late
during the task, and temporal dynamics were characterized by low learning rates. We observed
significant differences in amygdala responses for participants who exhibited a behavioral effect
in addition to verbal ratings of UCS likelihood. Amygdalar responses contrasted with temporal
dynamics in the insula where generalization gradients emerged earlier and gradually increased
with higher learning rates, similar to skin-conductance responses. Overall our results imply a
weak and late contribution of the amygdala to aversive generalization, in comparison to insular
responses that are stronger and contribute early during learning.

Adaptive Changes in the Viewing Behavior of Faces Following Aversive Learning....

I decided to write few paragraphs about papers I will be publishing from now on. This will be targeted for the non-technical audience and I hope will increase the accessibility to the published results. 

Here is our latest work that shows how eye-movement patterns during viewing of faces are modified when people learn to associate faces with an aversive outcome.

Eye movements can be effortlessly recorded while humans are engaged in different situations. This can provide important insights on what the nervous system tries to achieve, as eye-movements represent the final behavioral outcome of many complex neuronal processes, which are difficult to record and understand. 

We measured eye-movements while humans were viewing faces, and analyzed the resulting exploration patterns. These faces were calibrated to have a known similarity relationship. For example, faces A, B and C were physically organised in such a way that B was perceived equally similar to A and C, whereas A and C were the most dissimilar pairs. First, using novel similarity-based analyses we show that exploration patterns are dominated by physical aspects of faces. That is, the physical similarity relationship between A, B and C could be estimated to a good degree from the similarity of eye-movement patterns that were generated during viewing of these faces. 

Following in the experiment, we selected one face to be a nasty one by associating its presentation with a mild electric current on the hand of volunteers to generate an unpleasant feeling without hurting them. Participants learnt to associate this unpleasant outcome with only one face while other faces were kept the same as before. This resulted in a gradient of unpleasantness that was not present before, and led volunteers to generalize this unpleasant association to other faces to the extent they were perceived similar with the nasty face. This is a classic phenomenon known as generalization, since early times of Pavlov.

How does this new situation modify the similarity relationships between exploration patterns? Following learning, similarity relationships of eye movement patterns started to mirror this newly learnt categories of nasty vs. safe faces, even though there were no physical changes associated with faces. This is compatible with the idea that following learning along an arbitrary continuum of stimuli, a categorization process occurs internally that distinguishes safe from nasty faces. This then biases eye-movement patterns during viewing of faces in such a way to collect information specifically associated with the safe and nasty prototypes, leading faces resembling to these prototypes to be scanned similarly.  

This study provides a nice illustration on how eye movements patterns can shed light onto neuronal processes and help us understand what the brain is trying to achieve during learning.

Eye movement patterns on 8 different, but similar faces that were carefully calibrated to form a similarity continuum. These maps show the most attended locations for a single participant before learning. Similarity analysis of these heatmaps using FPSA method can detect learning induced changes in the scanning behavior.

Reference:

Aversive Learning Changes Face-Viewing Strategies, as Revealed by Model-Based Fixation-Pattern Similarity Analysis. Lea Kampermann, Niklas Wilming, Arjen Alink, Christian Buechel, Selim Onat.

All content in this post released under CC-BY 4.0.

SFN 2016 Abstract Submitted

Investigating the influence of priors in human fear generalization using a Bayesian model

Kampermann L, Büchel C & Onat S

When organisms form associations through learning, responses often generalize to stimuli that bare resemblance to the initially reinforced stimulus (CS+). During generalization, shifts in maximal responses away from the actual reinforced stimulus are commonly observed when tested with multiple stimuli along a similarity continuum, a phenomenon known as peak shift. We have developed a Bayesian framework that can account for peak shifts observed in behavioral ratings by postulating a group-level prior that is common to all participants over the range of stimuli forming the similarity continuum. The interaction of the prior distribution with a generalization component centered on the CS+ can act as a “magnet”, resulting in shifts of maximal responses depending on which stimuli was used as CS+. This formalism allows us to reverse-engineer subjects’ latent prior distributions by evaluating to what extent a specific hypothesis on priors can explain the observed shifts.

We employed a fear conditioning procedure using 8 faces organized along a circular similarity continuum varying in gender and identity dimensions. Two opposite faces were randomly selected as the CS+ and CS– for each participant (n  = 141). Following aversive conditioning, we obtained fear generalization gradients by asking participants for explicit UCS expectancy ratings, thus obtaining a two-sided generalization gradient ranging from CS+ to CS–. We compared the performance of the Bayesian model to fits carried out on a single-subject basis using two versions of Gaussian functions and subsequently tested different hypotheses on priors for the Bayesian model.

As expected a flexible Gaussian model with 2 parameters (width and location parameters) fitted the data significantly better than a simple Gaussian model (only width parameter) centered on the CS+ face (r = .91 vs. r = .74) underlining the presence of peak shifts in behavioral ratings. On the other hand, the Bayesian model performed significantly better than the simple Gaussian model, despite being based on only two more free parameters (flexible Gaussian: 2n parameters; simple Gaussian: n parameters; Bayesian model: n+2 parameters). Furthermore the Bayesian model explained behavioral gradients best, when a bimodal prior distribution peaking at both gender prototypes was used (r = .84). Testing different gender categories individually, a unimodal prior centered on the male category explained as much variance as one centered on female category. Overall, the predominance of a bimodal prior indicates that peak shifts can result from “magnet” effects of categorical face representatives instead of adversity attributions to a specific gender.

Poster Presented in ABIM

Manuscript accepted for publication in Nature Neuroscience: The Neuronal Basis of Fear Generalization in Humans

Our "Neuronal Basis of Fear Generalization" manuscript has been accepted to be published in Nature Neuroscience. 

You can download the pdf here.

It has been also highlighted in Nature Reviews Neuroscience.

Effect of aversive learning on discrimination of faces

In her Msc thesis, Lea Kampermann shows that humans can perceptually discriminate faces better, when these are paired with an aversive outcome. This effect was specific to the face, which was paired with an aversive outcome and was not observed for the one which was kept neutral throughout the experiment. Furthermore the effect was strongest when these faces were presented at shorter durations (~.6 s) allowing participants to make no more than two fixations per trial.

Her thesis contains also a detailed account on the methodology for generating face-stimuli that are perceptually calibrated to form a two-dimensional similarity gradient with equal perceptual steps between faces. The methodology is an extension of work from Yue et al. (Vision Research, 2012). If you wish to use these stimuli for your experiment they are available upon request.

Perceptually calibrated set of faces according to a simple primary visual cortex forming a circular similarity gradient. Details on their production can be read in Msc Thesis of Lea (please contact any of us for a pdf) .

Categorical, yet graded--single-image activation profiles of human category-selective cortical regions.

Mur M et al. investigated the selectivity of activity levels in parahippocampal place area (PPA) and fusiform face area (FFA) evoked by single images. They focus here only on the average BOLD activity within carefully selected ROIs.

The paper is very creative in terms of new analyses methods, relies heavily on rank orders and hypothesis testing with bootstrapping.

First it establishes the fact that PPA and FFA behaves as expected, that is face stimuli for FFA and place stimuli for PPA rank highest in terms of evoked activity. Overall PPA responses are more selective than FFA responses, reaching AUC values of 1 in both hemispheres. This results from the fact that faces evokes really high activity levels in the FFA, whereas, in the case of PPA inactivation by faces contribute to the PPA selectivity.

The rest of the report focuses on characterizing the category selectivity of these areas.

If an area is category selective in an ideal sense, non-preferred stimuli should never evoke higher activity levels than any other preferred stimulus, and if so, then only by chance due to noise.

The number of inverted pairs measures exactly the number of times one could identify violation of this rule by counting the number of times a stimulus from outside the category is ranked higher than a stimulus from within category. If these inverted pairs survive across multiple sessions (as measured by PRIP metric), this would be an evidence against ideal category selectivity. However as such, PRIP is not a very sensitive metric. For example one single preferred stimulus failing by chance to evoke any activity at all would be sufficient to generate very many inverted pairs, thus the metric seems to fluctuate highly non-linearly with respect to distance between inverted pairs. Therefore the authors, used the first sessions to identify preferred-nonpreferred pairs with largest activity difference, with the idea that an inversion with the largest activity difference would be the observation with least chance level. If these pairs survive across sessions, the difference in activity would then decrease only marginally and remain positive, thus providing evidence for stable inversions (as such, however this measure is also influenced by the noise on both the preferred and non-preferred stimulus). These analyses provide supporting evidence that FFA and PPA behave like an ideal category selective area, with the exception of left FFA, in line with the fact that left FFA is the ROI where smallest AUC values were observed (only though at ROI size of 128 voxels).

Advanced Numerical Methods in Neuroscience Lecture is now online

Advanced Numerical Methods in Neuroscience is a lecture I am holding in the graduate school Neurodapt. You can reach the overview of the lecture from this link and download the course material from this link.