Modular representations emerge in neural networks trained to perform context-dependent tasks DOI Creative Commons
W. Jeffrey Johnston, Stefano Fusi

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: Oct. 1, 2024

Abstract The brain has large-scale modular structure in the form of regions, which are thought to arise from constraints on connectivity and physical geometry cortical sheet. In contrast, experimental theoretical work argued both for against existence specialized sub-populations neurons (modules) within single regions. By studying artificial neural networks, we show that this local modularity emerges support context-dependent behavior, but only when input is low-dimensional. No anatomical required. We also specialization at population level (different modules correspond orthogonal subspaces). Modularity yields abstract representations, allows rapid learning generalization novel tasks, facilitates related contexts. Non-modular representations facilitate unrelated Our findings reconcile conflicting results make predictions future experiments.

Language: Английский

Tuned geometries of hippocampal representations meet the computational demands of social memory DOI Creative Commons
Lara M. Boyle, Lorenzo Posani, Sarah Irfan

et al.

Neuron, Journal Year: 2024, Volume and Issue: 112(8), P. 1358 - 1371.e9

Published: Feb. 20, 2024

Language: Английский

Citations

38
Mixed selectivity: Cellular computations for complexity DOI Creative Commons
Kay M. Tye, Earl K. Miller, Felix Taschbach

et al.

Neuron, Journal Year: 2024, Volume and Issue: 112(14), P. 2289 - 2303

Published: May 9, 2024

The property of mixed selectivity has been discussed at a computational level and offers strategy to maximize power by adding versatility the functional role each neuron. Here, we offer biologically grounded implementational-level mechanistic explanation for in neural circuits. We define pure, linear, nonlinear discuss how these response properties can be obtained simple Neurons that respond multiple, statistically independent variables display selectivity. If their activity expressed as weighted sum, then they exhibit linear selectivity; otherwise, Neural representations based on diverse are high dimensional; hence, confer enormous flexibility downstream readout circuit. However, circuit cannot possibly encode all possible mixtures simultaneously, this would require combinatorially large number neurons. Gating mechanisms like oscillations neuromodulation solve problem dynamically selecting which transmitted readout.

Language: Английский

Citations

36
A neural circuit targeting technique for investigating functional input-output organization in the nervous system DOI Creative Commons
Yusuke Kasuga,

Xiao-Wei Gu,

Tomoya Ohnuki

et al.

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2025, Volume and Issue: unknown

Published: April 19, 2025

Abstract Neurons communicate information across circuits and the function of cells in these is determined by both afferent inputs they receive efferent outputs send to other brain regions 1,2 . To study activity specific neuronal populations, transneuronal anterograde 3 retrograde 4–6 viral approaches have been employed define neural circuit elements or outputs, respectively. However, what missing a way neurons based on their outputs. Applying combination multiple recombinases anterograde/retrograde viruses, we developed technique called input-output P rojection-based IN tersectional C ircuit-tagging E nabled R ecombinases (PINCER) target cell types investigate functional organization circuits. We show logic application this with vivo calcium imaging optogenetic reveal distinct functions dynamics connectivity defined populations amygdala for emotional processing. Specifically, PINCER allowed parsing valence salience an type selectively mediating aversive memory formation. This allows neuroscientists identify novel subclasses combinatorial anatomical connectivity, providing tool fine dissection properties

Language: Английский

Citations

0
Modular representations emerge in neural networks trained to perform context-dependent tasks DOI Creative Commons
W. Jeffrey Johnston, Stefano Fusi

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: Oct. 1, 2024

Abstract The brain has large-scale modular structure in the form of regions, which are thought to arise from constraints on connectivity and physical geometry cortical sheet. In contrast, experimental theoretical work argued both for against existence specialized sub-populations neurons (modules) within single regions. By studying artificial neural networks, we show that this local modularity emerges support context-dependent behavior, but only when input is low-dimensional. No anatomical required. We also specialization at population level (different modules correspond orthogonal subspaces). Modularity yields abstract representations, allows rapid learning generalization novel tasks, facilitates related contexts. Non-modular representations facilitate unrelated Our findings reconcile conflicting results make predictions future experiments.

Language: Английский

Citations

0