Exploring Neuromorphic Computing Based on Spiking Neural Networks: Algorithms to Hardware

ACM Computing Surveys, Год журнала: 2022, Номер 55(12), С. 1 - 49

Опубликована: Ноя. 17, 2022

Neuromorphic Computing, a concept pioneered in the late 1980s, is receiving lot of attention lately due to its promise reducing computational energy, latency, as well learning complexity artificial neural networks. Taking inspiration from neuroscience, this interdisciplinary field performs multi-stack optimization across devices, circuits, and algorithms by providing an end-to-end approach achieving brain-like efficiency machine intelligence. On one side, neuromorphic computing introduces new algorithmic paradigm, known Spiking Neural Networks (SNNs), which significant shift standard deep transmits information spikes (“1” or “0”) rather than analog values. This has opened up novel research directions formulate methods represent data spike-trains, develop neuron models that can process over time, design for event-driven dynamical systems, engineer network architectures amenable sparse, asynchronous, achieve lower power consumption. other parallel thrust focuses on development efficient platforms algorithms. Standard accelerators are workloads not particularly suitable handle processing multiple timesteps efficiently. To effect, researchers have designed hardware rely sparse computations matrix operations. While most large-scale systems been explored based CMOS technology, recently, Non-Volatile Memory (NVM) technologies show toward implementing bio-mimetic functionalities single devices. In article, we outline several strides spiking networks (SNNs) taken recent past, present our outlook challenges needs overcome make bio-plausibility route successful one.

Язык: Английский

---

Where is the error? Hierarchical predictive coding through dendritic error computation

Trends in Neurosciences, Год журнала: 2022, Номер 46(1), С. 45 - 59

Опубликована: Ноя. 18, 2022

Top-down feedback in cortex is critical for guiding sensory processing, which has prominently been formalized the theory of hierarchical predictive coding (hPC). However, experimental evidence error units, are central to theory, inconclusive and it remains unclear how hPC can be implemented with spiking neurons. To address this, we connect existing work on efficient balanced networks lateral inhibition computation at apical dendrites. Together, this points an implementation neurons, where prediction errors computed not separate but locally dendritic compartments. We then discuss correspondence model experimentally observed connectivity patterns, plasticity, dynamics cortex.

Язык: Английский

---

The locus coeruleus broadcasts prediction errors across the cortex to promote sensorimotor plasticity

eLife, Год журнала: 2023, Номер 12

Опубликована: Янв. 25, 2023

Prediction errors are differences between expected and actual sensory input thought to be key computational signals that drive learning related plasticity. One way prediction could is by activating neuromodulatory systems gate The catecholaminergic locus coeruleus (LC) a major system involved in neuronal plasticity the cortex. Using two-photon calcium imaging mice exploring virtual environment, we found activity of LC axons cortex correlated with magnitude unsigned visuomotor errors. response profiles were similar both motor visual cortical areas, indicating broadcast throughout dorsal While layer 2/3 primary cortex, optogenetic stimulation facilitated stimulus-specific suppression responses during locomotion. This - induced minutes recapitulated effect on scale normally observed development across days. We conclude activity, facilitates sensorimotor consistent role modulating rates.

Язык: Английский

---

Cortico-cortical feedback engages active dendrites in visual cortex

Nature, Год журнала: 2023, Номер 617(7962), С. 769 - 776

Опубликована: Май 3, 2023

Abstract Sensory processing in the neocortex requires both feedforward and feedback information flow between cortical areas 1 . In processing, higher-level representations provide contextual to lower levels, facilitate perceptual functions such as contour integration figure–ground segmentation 2,3 However, we have limited understanding of circuit cellular mechanisms that mediate influence. Here use long-range all-optical connectivity mapping mice show influence from lateromedial higher visual area (LM) primary cortex (V1) is spatially organized. When source target represent same space, relatively suppressive. By contrast, when offset facilitating. Two-photon calcium imaging data this facilitating nonlinearly integrated apical tuft dendrites V1 pyramidal neurons: retinotopically (surround) stimuli drive local dendritic signals indicative regenerative events, two-photon optogenetic activation LM neurons projecting identified feedback-recipient spines can similar branch-specific signals. Our results how neocortical nonlinear together form a substrate support predictive cooperative interactions.

Язык: Английский

---

An integrative, multiscale view on neural theories of consciousness

Neuron, Год журнала: 2024, Номер 112(10), С. 1531 - 1552

Опубликована: Март 5, 2024

How is conscious experience related to material brain processes? A variety of theories aiming answer this age-old question have emerged from the recent surge in consciousness research, and some are now hotly debated. Although most researchers so far focused on development validation their preferred theory relative isolation, article, written by a group scientists representing different theories, takes an alternative approach. Noting that various often try explain aspects or mechanistic levels consciousness, we argue do not necessarily contradict each other. Instead, several them may converge fundamental neuronal mechanisms be partly compatible complementary, multiple can simultaneously contribute our understanding. Here, consider unifying, integration-oriented approaches been largely neglected, seeking combine valuable elements theories.

Язык: Английский

---