Stefano Fusi
@StefanoFusi2
Followers
5K
Following
219
Media
22
Statuses
114
Center for Theoretical Neuroscience, Columbia University
New York, NY
Joined December 2018
RT @VFascianelli: Excited to speak at the Davide Giri Talks at the Consulate General of Italy in New York!.We’ll be discussing complex syst….
0
3
0
RT @RNogueiraNeuro: The Grossman Center at UChicago is hiring Center Postdocs! Great scientific environment in a great city. Competitive sa….
neuroscience.uchicago.edu
0
30
0
We always see that: 1) neural responses are very diverse 2) the shattering dimensionality is as high as it can be. Now also in an extensive analysis of the IBL dataset. Wonderful collaboration with @LorenzoPosani , Shuqi Wang, Samuel Muscinelli, Liam Paninski. Many new analyses.
Long-overdue thread on our latest work using the IBL data to reveal the shared organizational principles of the neural code in the cortex. A systematic analysis of categoricality 🧱 and dimensionality 📐 of the neural code across 40+ regions. 👇 1/n
2
18
116
The geometry of adaptation! My first excursion in the V1 territory. Great collaboration with @MarioDipoppa .@MatteoCarandini and many others.
New results! Visual adaptation changes the geometry of V1 population activity: frequent stimuli elicit smaller responses but become more discriminable, consistent with our efficient coding model. You can find me on the "new neurotwitter" at mariodipoppa.
0
8
57
A beautiful work with a wonderful team! A lot of new ideas and a huge number of elegant experiments.
Excited to share our new paper out now @Nature, where we identified neural signatures of stress susceptibility and resilience in the amygdala-ventral hippocampal network to enable control of anhedonia! <gt;. Thread below:.
0
1
40
Modularity can emerge also in the absence of anatomical and metabolic constraints. What are the computational reasons? A new great theoretical study with @wjeffjohnston.
When does modular structure emerge in neural networks?.What are the consequences of this structure for learning and behavior?. New work with @StefanoFusi2 answers these questions and more: see 🧵below (1/11).
1
16
80
And finally, here are some ideas from our group about the formation of abstract representations:.
1
0
6
These geometries are also similar to those that correspond to the axis code of @doristsao see also: . Indeed, these are disentangled representations:
nature.com
Nature Communications - Little is known about the brain’s computations that enable the recognition of faces. Here, the authors use unsupervised deep learning to show that the brain...
1
0
8
Abstract disentangled representations are observed in multiple brain areas of rodents, human and non-human primates:.
1
0
10
For a similar experiment on non-human primates and more about the conceptual framework of abstract representations, see
cell.com
Social memory consists of both familiarity detection and recollection of past social episodes. Whether and how the hippocampus fulfills these roles is unclear. Boyle, Posani, et al. find that the...
3
0
3
And this is the unstructured geometry when the subjects cannot do inference
1
0
4
And this is the real data when the subjects can perform inference (neural recordings from single units in the human hippocampus). Notice the parallelism of the red lines
1
0
5
It’s 8 points in only 4 dimensions! A nice low dimensional structure. It is fun to rotate it in the original 4D space and project it onto the 2 dimensions of your screen. The two contexts are two tetrahedra, one yellow and one blue. Red=coding directions of context
1
0
5
Context and stimulus identity are disentangled when the subjects can make inferences, as in this idealized geometry. A,…,C are the 4 different stimuli. As they are unstructured, they approximately define a tetrahedron. The two tetrahedra of different contexts are aligned
1
0
5
Observing the birth of an abstract representation with disentangled variables in human HPC!!.Now published in Nature. Great collaboration with @courellis @JMinxha @amamelak @UeliRutishauser and others.
Delighted our latest finding! We discovered that abstract representations emerge in the human hippocampus when learning to perform inference. This change in neural geometry is due to disentanglement of discovered latent and observable variables. @Nature
2
35
152
And this is the unstructured geometry when the subjects cannot do inference
0
0
1
And this is the real data when the subjects can perform inference (neural recordings from single units in the human hippocampus). Notice the parallelism of the red lines
1
0
2
It’s 8 points in only 4 dimensions! A nice low dimensional structure. It is fun to rotate it in the original 4D space and project it onto the 2 dimensions of your screen. The two contexts are two tetrahedra, one yellow and one blue. Red=coding directions of context
1
0
0
Context and stimulus identity are disentangled when the subjects can make inferences, as in this idealized geometry. A,…,C are the 4 different stimuli. As they are unstructured, they approximately define a tetrahedron. The two tetrahedra of different contexts are aligned
1
0
0