The brain utilizes stochastic neural processes as a source of randomness for sampling
the verdict
SUPPORTED
the evidence backs this
refutedsupported
the weight of evidence
3 sources for · 0 against
Peer-reviewed computational neuroscience literature establishes that neural models interpret brain states as Monte Carlo samples from probability distributions, utilizing stochastic neural processes and biological noise as a source of variability for probabilistic inference and sampling.
Animals perform near-optimal probabilistic inference in a wide range of psychophysical tasks. Probabilistic inference requires trial-to-trial representation of the uncertainties associated with task variables and subsequent use of this representation. Previous work has implemented such computations using neural networks with hand-crafted and task-dependent operations. We show that generic neural networks trained with a simple error-based learning rule perform near-optimal probabilistic inference in nine common psychophysical tasks. In a probabilistic categorization task, error-based learning in a generic network simultaneously explains a monkey's learning curve and the evolution of qualitative aspects of its choice behavior. In all tasks, the number of neurons required for a given level of performance grows sublinearly with the input population size, a substantial improvement on previous implementations of probabilistic inference. The trained networks develop a novel sparsity-based probabilistic population code. Our results suggest that probabilistic inference emerges naturally in generic neural networks trained with error-based learning rules.Behavioural tasks often require probability distributions to be inferred about task specific variables. Here, the authors demonstrate that generic neural networks can be trained using a simple error-based learning rule to perform such probabilistic computations efficiently without any need for task specific operations.
The Constrained Disorder Principle (CDP) offers a new framework for understanding how biological systems use and manage noise to maintain optimal functionality. This review explores the relationship between noise and biological systems at various scales, including genetic, cellular, and organ levels, and its implications for system malfunctions. According to the CDP, all systems require an optimal range of noise to function appropriately, and disease states can arise when these noise levels are disrupted. This review presents evidence supporting this principle across different biological contexts, such as genetic variability, cellular behavior, brain functions, human behavior, aging, evolution, and drug administration. For accurate clinical assessments, it is essential to distinguish between technical variability and intrinsic biological variability. When noise is adequately constrained, it serves as a fundamental mechanism for system adaptation and optimal functioning rather than simply a source of disruption. These findings have important implications for developing more effective therapeutic strategies and understanding biological systems' dynamics. CDP-based second-generation artificial intelligence systems can help regulate noise levels to address malfunctions. These systems have improved clinical outcomes in various conditions by incorporating controlled randomness. Understanding these patterns of variability has significant implications for diagnosis, treatment monitoring, and the development of more effective therapeutic strategies across various medical conditions.
ic counterpart of Hopfield networks (Hopfield, 1982 ), but where stochasticity is caused by multiplicative noise at the connections (synapses).
As in neural sampling (Fiser et al., 2010 ; Berkes et al., 2011 ), neural states (such as instantaneous firing rates or individual spikes) in our model are interpreted as Monte Carlo samples of a probability distribution. The source of the stochasticity in neural samplers is often unspecified, or is assumed to be caused by independent, background Poisson activities (Petrovici et al., 2013 ; Neftci et al., 2014 ). Background Poisson activity can be generated self-consistently in balanced excitatory-inhibitory networks (van Vreeswijk and Sompolinsky, 1996 ), or by using finite-size effects and neural mismatch (Amit and Brunel, 1997 ). Although a large enough neural sampling network could generate its own stochasticity in this fashion, the variability in the spike trains decreases strongly as the firing rate increases (Fusi and Mattia, 1999 ; Brunel, 2000 ; Moreno-Bote, 2014 ), unless there is an external source of noise or some parameters are fine-tuned. Furthermore, when a neural sampling network generates its own noise, correlations in the background activity can play an adverse role in its performance (Probst et al., 2015 ). Correlations can be mitigated by adding feed-forward connectivity between a balanced network (or other dedicated source of Poisson spike trains) and the neural sampler, but such solutions entail increased resources in neurons, connectivity, and ultimately energy. Therefore, the above techniques for generating stochasticity are not ideal for neural sampling.
On the other hand, synaptic unreliability can induce the necessary stochasticity without requiring a dedicated source of Poisson spike trains. The unreliable transmission of synaptic vesicles in biological neurons is a well studied phenomenon (Katz, 1966 ; Branco and Staras, 2009 ), and many studies suggested it as a major source of stochasticity in
Everything we examined (3)
This check searched the claim as stated. It did not run a separate search for evidence against it.