Brain-inspired learning in artificial neural networks: A review

APL Machine Learning, Journal Year: 2024, Volume and Issue: 2(2)

Published: May 9, 2024

Artificial neural networks (ANNs) have emerged as an essential tool in machine learning, achieving remarkable success across diverse domains, including image and speech generation, game playing, robotics. However, there exist fundamental differences between ANNs’ operating mechanisms those of the biological brain, particularly concerning learning processes. This paper presents a comprehensive review current brain-inspired representations artificial networks. We investigate integration more biologically plausible mechanisms, such synaptic plasticity, to improve these networks’ capabilities. Moreover, we delve into potential advantages challenges accompanying this approach. In review, pinpoint promising avenues for future research rapidly advancing field, which could bring us closer understanding essence intelligence.

Language: Английский

---

Bottom-Up and Top-Down Approaches for the Design of Neuromorphic Processing Systems: Tradeoffs and Synergies Between Natural and Artificial Intelligence

Proceedings of the IEEE, Journal Year: 2023, Volume and Issue: 111(6), P. 623 - 652

Published: June 1, 2023

While Moore's law has driven exponential computing power expectations, its nearing end calls for new avenues improving the overall system performance. One of these is exploration alternative brain-inspired architectures that aim at achieving flexibility and computational efficiency biological neural processing systems. Within this context, neuromorphic engineering represents a paradigm shift in based on implementation spiking network which memory are tightly co-located. In paper, we provide comprehensive overview field, highlighting different levels granularity realized comparing design approaches focus replicating natural intelligence (bottom-up) versus those solving practical artificial applications (top-down). First, present analog, mixed-signal digital circuit styles, identifying boundary between through time multiplexing, in-memory computation, novel devices. Then, highlight key tradeoffs each bottom-up top-down approaches, survey their silicon implementations, carry out detailed comparative analyses to extract guidelines. Finally, identify necessary synergies missing elements required achieve competitive advantage systems over conventional machine-learning accelerators edge applications, outline ingredients framework toward intelligence.

Language: Английский

---

Brain-inspired methods for achieving robust computation in heterogeneous mixed-signal neuromorphic processing systems

Neuromorphic Computing and Engineering, Journal Year: 2023, Volume and Issue: 3(3), P. 034002 - 034002

Published: July 11, 2023

Abstract Neuromorphic processing systems implementing spiking neural networks with mixed signal analog/digital electronic circuits and/or memristive devices represent a promising technology for edge computing applications that require low power, latency, and cannot connect to the cloud off-line processing, either due lack of connectivity or privacy concerns. However, these are typically noisy imprecise, because they affected by device-to-device variability, operate extremely small currents. So achieving reliable computation high accuracy following this approach is still an open challenge has hampered progress on one hand limited widespread adoption other. By construction, hardware have many constraints biologically plausible, such as heterogeneity non-negativity parameters. More more evidence showing applying artificial networks, including those used in intelligence, promotes robustness learning improves their reliability. Here we delve even into neuroscience present network-level brain-inspired strategies further improve reliability neuromorphic systems: quantify, chip measurements, what extent population averaging effective reducing variability responses, demonstrate experimentally how coding cortical models allow silicon neurons produce representations, show robustly implement essential computational primitives, selective amplification, restoration, working memory, relational exploiting strategies. We argue can be instrumental guiding design robust ultra-low power implemented using imprecise substrates subthreshold emerging memory technologies.

Language: Английский

---

Neuromorphic bioelectronic medicine for nervous system interfaces: from neural computational primitives to medical applications

Progress in Biomedical Engineering, Journal Year: 2023, Volume and Issue: 5(1), P. 013002 - 013002

Published: Jan. 1, 2023

Abstract Bioelectronic medicine treats chronic diseases by sensing, processing, and modulating the electronic signals produced in nervous system of human body, labeled ‘neural signals’. While circuits have been used for several years this domain, progress microelectronic technology is now allowing increasingly accurate targeted solutions therapeutic benefits. For example, it becoming possible to modulate specific nerve fibers, hence targeting diseases. However, fully exploit approach crucial understand what aspects are important, effect stimulation, circuit designs can best achieve desired result. Neuromorphic represent a promising design style achieving goal: their ultra-low power characteristics biologically plausible time constants make them ideal candidate building optimal interfaces real neural processing systems, enabling real-time closed-loop interactions with biological tissue. In paper, we highlight main features neuromorphic that ideally suited interfacing show how they be build hybrid artificial systems. We present examples computational primitives implemented carrying out computation on sensed these systems discuss way use outputs stimulation. describe applications follow approach, open challenges need addressed, propose actions required overcome current limitations.

Language: Английский

---

ETLP: Event-based Three-factor Local Plasticity for online learning with neuromorphic hardware

Neuromorphic Computing and Engineering, Journal Year: 2024, Volume and Issue: 4(3), P. 034006 - 034006

Published: July 24, 2024

Abstract Neuromorphic perception with event-based sensors, asynchronous hardware, and spiking neurons shows promise for real-time, energy-efficient inference in embedded systems. Brain-inspired computing aims to enable adaptation changes at the edge online learning. However, parallel distributed architectures of neuromorphic hardware based on co-localized compute memory imposes locality constraints on-chip learning rules. We propose three-factor local plasticity (ETLP) rule that uses pre-synaptic spike trace, post-synaptic membrane voltage a third factor form projected labels no error calculation, also serve as update triggers. ETLP is applied visual auditory pattern recognition using feedforward recurrent neural networks. Compared back-propagation through time, eProp DECOLLE, achieves competitive accuracy lower computational complexity. show when plasticity, threshold topology are necessary learn spatio-temporal patterns rich temporal structure. Finally, we provide proof concept implementation FPGA highlight simplicity its primitives how they can be mapped into real-time interaction low energy consumption.

Language: Английский

---

The road to commercial success for neuromorphic technologies

Nature Communications, Journal Year: 2025, Volume and Issue: 16(1)

Published: April 15, 2025

Neuromorphic technologies adapt biological neural principles to synthesise high-efficiency computational devices, characterised by continuous real-time operation and sparse event-based communication. After several false starts, a confluence of advances now promises widespread commercial adoption. Gradient-based training deep spiking networks is an off-the-shelf technique for building general-purpose applications, with open-source tools underwritten theoretical results. Analog mixed-signal circuit designs are being replaced digital equivalents in newer simplifying application deployment while maintaining benefits. Designs in-memory computing also approaching maturity. Solving two key problems-how program general applications; how deploy them at scale-clears the way success processors. Ultra-low-power technology will find home battery-powered systems, local compute internet-of-things consumer wearables. Inspiration from uptake tensor processors GPUs can help field overcome remaining hurdles.

Language: Английский

---

Bio-inspired artificial mechanoreceptors with built-in synaptic functions for intelligent tactile skin

Nature Materials, Journal Year: 2025, Volume and Issue: unknown

Published: April 28, 2025

Language: Английский

---

Vector Symbolic Finite State Machines in Attractor Neural Networks

Neural Computation, Journal Year: 2024, Volume and Issue: 36(4), P. 549 - 595

Published: March 8, 2024

Hopfield attractor networks are robust distributed models of human memory, but they lack a general mechanism for effecting state-dependent transitions in response to input. We propose construction rules such that an network may implement arbitrary finite state machine (FSM), where states and stimuli represented by high-dimensional random vectors all enacted the network's dynamics. Numerical simulations show capacity model, terms maximum size implementable FSM, be linear dense bipolar approximately quadratic sparse binary vectors. model is imprecise noisy weights, so prime candidate implementation with high-density unreliable devices. By endowing ability emulate FSMs, we plausible path which FSMs could exist as computational primitive biological neural networks.

Language: Английский

---

A bio-inspired hardware implementation of an analog spike-based hippocampus memory model

Published: April 3, 2024

The need for processing at the edge increasing amount of data that is being produced by multitudes sensors has led to demand mode power efficient computational systems, exploring alternative computing paradigms and technologies. Neuromorphic engineering a promising approach can address this developing electronic systems faithfully emulate properties animal brains. In particular, hippocampus stands out as one most relevant brain region implementing auto associative memories capable learning large amounts information quickly recalling it efficiently. work, we present spike-based memory model inspired takes advantage features analog circuits: energy efficiency, compactness, real-time operation. This learn memories, recall them from partial fragment forget. It been implemented Spiking Neural Networks directly on mixed-signal neuromorphic chip. We describe details hardware implementation demonstrate its operation via series benchmark experiments, showing how research prototype paves way development future robust low-power systems.

Language: Английский

---

Feasibility of a Personal Neuromorphic Emulation

Published: July 15, 2024

The representation of intelligence is achieved by patterns connections among neurons in brains and machines. Brains grow continuously, such that their develop through activity-dependent specification with the continuing ontogenesis individual experience. theory active inference proposes developmental organization sentient systems reflects general processes informatic self-evidencing, minimization free energy, may be described information terms are not dependent on a specific physical substrate. At certain level complexity, self-evidencing living (self-organizing) becomes hierarchic reentrant, effective consciousness emerges as consequence good regulator. These principles imply an adequate reconstruction computational dynamics human brain possible sufficient neuromorphic emulation.

Language: Английский

---