Criticality in the Brain: Evidence and Implications for Neuromorphic Computing

Abstract

We have discovered an unexpected correlation between the operational temperature of the brain and cognitive abilities across a wide variety of animal species. This correlation is extracted from available data in the literature of the temperature range Δ T at which an animal's brain can operate and its encephalization quotient EQ, which can be used as a proxy for cognitive ability. In particular, we found a power-law dependence between Δ T and EQ. These data support the theory that the brain behaves as a critical system where temperature is one of the critical parameters, tuning the performance of the neural network.

Challenges in materials and devices for resistive-switching-based neuromorphic computing

“…Some pioneer ideas regarding artificial intelligence (AI) go back to seminal thoughts of John von Neumann 4 who first pointed out that brains cognitive power does not emerge from accurate digital calculations but rather from a collective form of computation involving large number of slow, imprecise and unreliable analog components. 5,6 Our brains are amazingly more efficient at recognizing patterns than powerful digital computers. To distinguish the image of a cat from a tiger is a hard task for a computer, but not for a child.…”

Regimes of collective logic

“…Considered as an algorithm allowing individuals to exceed their individual limitations, self-organized natural intelligence is observed from the cellular level and collective insects (Camazine et al, 2001) to higher animals and the biosphere as a whole (Estep, 2006;Krause and Ruxton, 2002;Spier, 2011;Tadi c et al, 2017). The same principles are found in the organization of the human brain considered as a critically organized collective of neurons (Beggs, 2019;Cocchi et al, 2017;Pospelov et al, 2019;Trastoy and Schuller, 2018); as development and functioning of the latter is conditioned by social interaction (Buckholtz and Marois, 2012;Conte, 2002;Humphrey, 1976;Märtsin, 2008;Vygotsky, 1978), so that intelligence appears as a fundamentally collective phenomenon even when it functions in relatively standalone mode. This observation is expressed in century-old concepts of the noosphere and the collective unconscious (Jung, 2014;Levit, 2000) that are subject to selforganization regularities studied in recent decades.…”

Criticality and Heterogeneity as General Principles to Optimize Performance for Spiking Cameras

“…These concepts have been studied more generally in complex dynamic systems [29], including neural systems [30]. Many studies have discussed the possibility of computation optimization in neural systems with either criticality or heterogeneity [4,10,31]. Studies have also shown that heterogeneity can promote more robust criticality and functionality [29,31].…”

“…For spiking cameras, the large dynamic range ensures that there is less sensitivity to only weak or strong inputs to achieve a balance between underexposure and overexposure of photos. In addition to the dynamic range, criticality and heterogeneity can be used to optimize many more aspects of neural information processing, e.g., information transformation and storage [10,31]. Therefore, these principles could be applied to the future optimal design of neuromorphic devices with different application purposes.…”

“…Scale-free neural avalanche activities, the hallmark of criticality, have been observed on different scales of neural activities, suggesting that neural systems operate around critical points [9]. Furthermore, both theoretical and experimental studies suggest that neural information processing is maximized at the critical point [10]. Neural systems are heterogeneous in nature at different levels [11,12].…”