11/ The geometry then changes after ingressing, and the mouse no longer represents valence. Instead, only the exposed, aversive condition stands out from the bulk. This is compatible with the coding of a new variable, "safety", which has the highest decoding and CCGP.

10/ Crucially, it does so without specialized populations! Selectivity is largely mixed across all variables:

9/ This is indeed what we observe in the BLA: it represents behavioral and valence variables as "abstract", with a disentangled compositional geometry, characterized by a high CCGP (generalization) and low CVI (interference):

8/ The key is that specialized neurons are not necessary to satisfy these requirements! We can have the same properties (high CCGP, low CVI) as long as the representational geometry of the encoded variables is disentangled / compositional (left panel):

7/ But we can also characterize these properties (high generalization, low interference) on a population level. For generalization, we can use cross-condition generalization performance (CCGP). For interference, we developed a new measure: cross-variable interference (CVI):

4/ What about the neural code? All variables were highly decodable, but representing the world is not enough to act on it. What matters is "how" this information is organized!

3/ Both trembling and freezing are more frequent in response to aversive odors (CS+), but far from exclusive. Mice typically tremble and then ingress when CS+ is presented, in accordance with the predator-imminence theory (freeze -> flee).

2/ The BLA is known to have specialized circuits for valence (+/-), but what about other features of emotional experiences, such as behavior? In a virtual burrow assay, we observed that mice both tremble (~freeze) and ingress (~flee) in response to aversive (shock-paired) odors.

For a neural circuit to be "specialized", does it need specialized neurons? In our new work - now out in @NatureNeuro - we address this question in the BLA, using representational geometry to propose a new population-centric way of defining circuit specialization. 🧵👇 1/

Absolute peak graphical abstract in this Cell paper about reprogramming the leaf-cutter's brains 😂 www.sciencedirect.com/science/arti...

I feel you captain

Seems it narrows down to cats more than predators. Not many people imitating bears or gators with makeup. But what you say about cats is true - some ppl even think anime aesthetics comes from cat faces. A theory is they have baby human features (big eyes tiny mouth) and we are wired to like that.

Finally, we computed the Shattering Dimensionality (SD) - a measure of coding flexibility (fraction of linearly solvable classification problems on conditions in the activity space). When considering independent conditions, SD was maximal in all areas, including sensory ones! 13/n

Using this M we were able to verify our theory, which accurately and quantitatively predicted the relationship between clustering and dimensionality. Importantly, clustering and dimensionality are inversely correlated in the data, with PR increasing along the hierarchy. 12/n

Importantly, M is the number of independent conditions: those that are discriminable from each other in the neural activity. We developed an iterative algorithm to isolate the independent conditions in the data - finding that cognitive regions encode more conditions than sensory ones. 11/n

We studied the relation between clusters and geometry in a mathematical model where participation ratio (PR), a measure of dimensionality, can be computed analytically from Gaussian clusters - PR depends on # conditions (M), # clusters (k), and cluster quality. 10/n

What are the computational implications of categorical clusters? Intuitively, clusters reduce the dimensionality of the data (correlations). This constrains the geometry in the activity space since PR(X) = PR(X^T), limiting the flexibility typical of high-dim representations. 9/n

What about categorical clustering? We developed a pipeline that (1) finds the best clusters (2) computes quality (silhouette) (3) compares to uni-modal null model. We found that a few regions are better than the null. Most notably, they are all low-hierarchy ones (eg, VISp). 8/n

We started from a large anatomical scale, studying the avg. selectivity for each region. The more regions are anatomically connected, the more similar their selectivities are. Also, we can decode the region from single neuron response profiles. Well-connected regions are harder to decode. 7/n

To study single-neuron responses, we developed a reduced-rank regression model (RRR model), which captures well time-varying neural activity in an interpretable set of parameters, giving an 8-dimensional embedding (8 variables) for every single neuron. 6/n

We analyzed the neural representations of cognitive, sensory, and movement variables in 43 mouse cortical regions (15000+ cells, IBL BrainWide data set) and compared them with anatomical information of cortical connectivity (Allen Atlas). 5/n

What does "structure" mean in these two spaces? In the conditions space, neurons could form functionally distinct clusters (categorical representations); in the neural space, conditions could form low/high-dimensional geometries with different computational properties. 4/n

To answer this, we developed a set of analysis pipelines to systematically study the structure of neural representations from two perspectives: (a) single neuron selectivity (b) representational geometry - and a mathematical theory to understand their mutual relation. 3/n

On a larger scale, the brain is clearly functionally and anatomically organized. However, many studies at single-neuron resolution show a complex and seemingly disorganized code, especially in cognitive areas. How do we reconcile these two seemingly conflicting perspectives? 2/n

Here I refer to different areas having different selectivity patterns that reflect their anatomical organization, i.e. everything is not everywhere & what/where reflects connectivity. In general, I use structured as in "not a randomly mixed Gaussian". This is also the case on a module/cortex scale:

@jbarbosa.org what's surprising to me is that even with this generous definition of modular/categorical, most non-sensory regions fail to show any structure whatsoever (beyond representing some variables more than others). Below is cluster quality, zscore from a gaussian randomly-mixed null model:

In the IBL data we also find structure at that level: connected regions have similar response profiles, and we can decode the region from the response profile of single neurons. However, no structure within regions (except VISp). It seems there is modularity on a larger scale @benhayden.bsky.social

Note that "associated with" is vague enough to include people who recently left, who are visiting, etc. Don't be shy

Mine is definitely Hofstadter's masterpiece "Godel, Escher, Bach". As a physics student, it introduced me to the beauty of complexity and seeded in me a fascination for intelligence that eventually led me to neuroscience. He even autographed my copy in Italian!