A research team led by Senior Researcher Seongsik Park of the Semiconductor Technology Research Division at the Korea Institute of Science and Technology (KIST) has developed a learning technique designed to make neuromorphic artificial intelligence more accurate and more efficient. Called A²SG, or Adaptive and Asymmetric Surrogate Gradients, the method improves the training of spiking neural networks, a class of AI models inspired by the brain’s sparse, event-driven communication system. The research was accepted as a regular paper at the International Conference on Machine Learning, or ICML 2026, and presented on July 7 at the conference in Seoul, the first time the event has been held in South Korea.

Unlike conventional artificial neural networks, which continuously exchange numerical values, spiking neural networks communicate through brief electrical-like events known as spikes. Neurons remain largely inactive until incoming signals reach a threshold, allowing the system to process information only when necessary. This event-driven operation can dramatically reduce energy consumption, particularly in applications where data arrive intermittently, such as cameras, wearable sensors, drones and other edge devices. Yet the same biological realism that gives spiking networks their efficiency has made them difficult to train.

Modern deep learning systems generally improve by calculating gradients. These mathematical signals indicate how much each parameter in a network contributed to an error and in which direction the parameter should be adjusted. The process, known as backpropagation, allows a model to refine itself over millions or billions of training examples. Spiking neurons, however, produce abrupt, discontinuous spikes rather than smooth numerical outputs. Because a tiny change in an input may not immediately alter whether a neuron fires, the exact gradient can become zero or undefined, preventing conventional optimization methods from effectively updating the network.

To overcome this obstacle, researchers use substitute functions called surrogate gradients. These functions approximate the missing gradient around a neuron’s firing threshold, making it possible to apply gradient-based learning to spiking models. The KIST team’s A²SG method advances this strategy in two directions. Its adaptive component changes the gradient behavior according to the circumstances of training, while its asymmetric component reflects the fact that biological neurons do not respond identically on either side of their activation threshold. Together, the two mechanisms provide a more informative learning signal than a fixed, conventional surrogate gradient.

The team applied A²SG to a large-scale spiking neural network based on the transformer architecture, the influential design that underpins systems such as ChatGPT and many modern vision and language models. In large-scale ImageNet image-recognition experiments, the approach achieved what KIST described as world-leading accuracy among spiking neural networks. The result is important because ImageNet is widely used as a demanding benchmark for visual recognition, and performance at this scale can reveal whether a method is suitable for models far larger and more complex than laboratory prototypes.

The reported gains were not limited to accuracy. According to the research team, A²SG required approximately one-sixth of the computational overhead associated with Google’s leading training method while delivering better recognition performance. The method also produced consistent improvements across a broad range of network architectures and applications, from relatively small models to large transformer-based systems. That versatility could be crucial for neuromorphic AI, since a training technique that works only for one specialized architecture would have limited practical value.

The significance of the work extends beyond software benchmarks. Because A²SG can be implemented without changing the underlying hardware, it could potentially be deployed on existing systems before specialized neuromorphic processors become widely available. Its low-power advantages could support on-device AI in smartphones, smartwatches, medical wearables and autonomous drones, where sending data continuously to a distant data center is costly, slow or impractical. The same principle could benefit smart sensors that must operate continuously while powered by small batteries or harvested energy.

KIST researchers say the technology may also help South Korea strengthen its position in AI semiconductor development. The institute plans to continue training and validating large-scale models and to connect the learning algorithm with neuromorphic hardware under development. Future work will include applying the method to next-generation processors such as the probability-based Random Processing Unit, or RPU, being developed at KIST. If those efforts succeed, the combination of efficient algorithms and specialized chips could create AI systems that respond locally, consume far less electricity and reduce dependence on centralized computing infrastructure.

Park said the research addresses structural limitations that have slowed progress in neuromorphic AI and expands the prospects for practical low-power intelligence. The broader challenge is now to demonstrate that these gains can survive the transition from controlled benchmarks to real-world devices, where latency, reliability, memory capacity and hardware variability all matter. Still, by narrowing the performance gap between brain-inspired spiking networks and conventional deep learning while reducing training costs, A²SG offers a striking glimpse of how the next generation of AI could become both more capable and less energy-intensive.

Subject of Research: Low-power neuromorphic artificial intelligence, spiking neural networks and surrogate-gradient learning.

Article Title: A²SG: Adaptive and Asymmetric Surrogate Gradients for Training Deep Spiking Neural Networks

News Publication Date: 7-Jul-2026

Web References: https://doi.org/10.48550/arXiv.2606.11236

References: A²SG research paper, arXiv:2606.11236; Proceedings of the International Conference on Machine Learning (ICML 2026).

Image Credits: Korea Institute of Science and Technology (KIST)

Keywords

Neuromorphic computing, spiking neural networks, artificial intelligence, surrogate gradients, low-power AI, AI semiconductors, transformer networks, ImageNet, KIST, A²SG

Astrocytes were once thought to be mere support cells, passive caretakers providing structural and metabolic assistance to neurons. However, a groundbreaking initiative, funded by the U.S. Army’s Multi-University Research Initiative (MURI), is challenging this narrow view. Led by a multidisciplinary team from the University of Maryland and Claremont Colleges, the project seeks to unravel the computational secrets of astrocytes and integrate these insights into next-generation AI architectures. Their goal is nothing short of transformative: to engineer “hybrid AI” systems that blend conventional computing with bio-inspired mechanisms, thereby crafting machines that learn, adapt, and think more like the human brain.

The conceptual leap embraced by this initiative stems from a critical insight about the brain’s cellular makeup. While neurons transmit information at lightning speed—milliseconds or less—astrocytes operate on a much slower timescale, processing signals over seconds. This temporal distinction hints at complementary functional roles: neurons handle rapid, moment-to-moment computations, whereas astrocytes integrate information over longer windows, acting like a “slow-burn” memory system that stabilizes and modulates neural activity. By mimicking this duality, artificial neural networks could harness new dimensions of learning and resilience.

The team’s pioneering research began with the development of a hybrid AI network that explicitly models both artificial neurons and astrocytes wired together to emulate their interaction in the brain. Published in the journal Neurocomputing, this model revealed striking performance advantages. Networks containing approximately twice as many astrocytes as neurons—the same ratio found in human brains—outperformed those composed exclusively of either cell type alone. This synergy suggests that the collaboration between neurons and astrocytes is not just complementary but essential for efficient computation.

With the initial biological framework established, the researchers delved deeper into the intrinsic dynamics of astrocytes, particularly their characteristic slow oscillating waves. Unlike the steady, static connections commonly used in traditional AI, astrocyte communication entails rhythmic, pulsatile fluctuations that modulate synaptic strengths over time. Incorporating these rhythmic variations into their neural models led to the invention of a novel algorithm dubbed “rhythmic sharing,” in which AI network connections continuously pulse and shift. This rhythmic modulation challenges the AI status quo, where link strengths remain fixed after training, potentially unlocking adaptive capabilities that mirror the brain’s plasticity.

Remarkably, this rhythmic sharing algorithm transcended theory and demonstrated tangible superiority in applied scenarios. Tested against conventional AI systems on simulated anomaly detection tasks—such as monitoring water treatment facilities under cyberattack and predicting jet engine failures—the astrocyte-inspired network detected environmental shifts more quickly and reliably. These results, published in npj Unconventional Computing, underscore the potential of astrocyte-based AI to excel in dynamic, real-world environments where traditional models can fail silently when conditions drift subtly over time.

The rhythmic sharing network’s ability to synchronize continuous internal pulse patterns makes it exquisitely sensitive to early signs of change. Unlike standard AI systems that may overlook gradual deviations until damage becomes manifest, this novel algorithm “listens” to the internal rhythms and signals disturbances before they become problematic. This anticipatory capability is akin to a sentinel always on guard, offering a new paradigm for real-time monitoring, predictive maintenance, and anomaly detection across sectors ranging from industrial infrastructure to cybersecurity.

Beyond immediate applications, this research opens a treasure trove of possibilities for understanding how the brain’s hidden half contributes to cognition. Astrocytes participate actively in modulating synaptic transmission, regulating neurotransmitter levels, and managing blood flow—all vital for learning, memory formation, and adaptive behavior. By abstracting these biological functions into computational algorithms, hybrid AI systems can bridge the gap between rigid traditional models and fluid, context-aware problem-solving, paving the way for more organic machine intelligence.

Professor Wolfgang Losert, a physicist at the University of Maryland and co-leader of the project, emphasized the significance of this paradigm shift. “We’re harnessing algorithms rooted in biological computation hidden from view because they do not rely on electrical signaling like neurons do,” he said. “Astrocytes are dynamic participants in cognitive processes, and translating their mechanisms into AI can lead to more robust and efficient learning models that outperform today’s neural-network-based approaches.”

This work is the culmination of years of interdisciplinary collaboration, integrating principles from physics, chemistry, electrical engineering, and computer science. It reflects the growing recognition that innovation in AI will increasingly depend on insights gleaned from biology’s complexity and adaptability. Foundational studies of living astrocytes in the Losert lab, supported by the Air Force Office of Scientific Research’s biophysics program, provided the experimental backbone to inform and validate these groundbreaking computational frameworks.

Looking forward, the team envisions hybrid AI architectures that continuously adapt to fluctuating conditions by leveraging astrocyte-like modulation. Such systems could revolutionize not only anomaly detection but broader cognitive tasks by maintaining stable yet flexible representations of data over time, akin to human brain function. The ultimate ambition is to bridge human and machine intelligence more closely, bringing AI out of the narrow realm of static pattern recognition into an arena of dynamic understanding and real-time learning.

As these explorations continue, the implications ripple across numerous fields—from health monitoring, where early detection of physiological abnormalities could save lives, to communication technologies that adapt fluidly to changing signals, to autonomous systems functioning reliably in unpredictable environments. The astrocyte model enriches AI’s conceptual toolkit, introducing temporal layering and rhythmic dynamism as core computational ingredients.

In sum, the astrocyte-inspired hybrid AI initiative not only expands our understanding of the brain’s hidden half but also charts a promising pathway for advancing AI beyond its current neuron-centric paradigm. By embracing the rich interplay of slow and fast cellular processes, these hybrid networks could mark the dawn of a new era in which artificial intelligence learns, adapts, and senses the world in ways eerily reminiscent of the human mind.

Subject of Research: Artificial intelligence inspired by astrocyte-neuron interactions in the human brain.

Article Title: Emergent detection of concept drift within the glia-inspired ‘rhythmic sharing’ algorithm.

News Publication Date: June 3, 2026.

Web References:

• Hybrid AI project: https://hybrid-ai.umd.edu/
• University of Maryland Invention of the Year Award: https://cmns.umd.edu/news-events/news/2025-invention-year-awards
• DOI link to npj Unconventional Computing article: http://dx.doi.org/10.1038/s44335-026-00067-3

References:

• Yang et al., Neurocomputing (2026).
• Losert et al., Physical Review Research.
• Ian Whitehouse et al., npj Unconventional Computing (2026).

Image Credits: Yang et al., Neurocomputing (2026).

Keywords

Artificial intelligence, Artificial neural networks, Artificial consciousness, Quantum computing, Supercomputing, Brain, Brain structure, Gray matter, Human brain, Neural pathways

Neuromorphic computing is an evolving paradigm that mimics the brain’s intricate architecture and processing capabilities. Unlike traditional silicon-based computing platforms, which rely heavily on power-intensive operations, neuromorphic systems aspire to replicate the brain’s unparalleled efficiency and adaptability by leveraging neuronal principles at both cellular and network scales. Current understanding of brain functions remains fragmented, making it tough to translate biological computation into effective artificial systems. The centre’s researchers intend to address this knowledge gap by developing integrative models that combine biological experiments using stem-cell-derived human neurons with advanced computational frameworks and innovative photonic hardware components.

The centre will be headquartered within the Aston Institute of Photonic Technologies (AIPT) and will unite expertise from a consortium of leading UK universities including Oxford, Cambridge, Southampton, Queen Mary University of London, Loughborough, and Strathclyde. This multidisciplinary alliance brings together neuroscientists, physicists, material scientists, engineers, and computer scientists to create a holistic approach to neuromorphic system design. Their collaboration aims to develop computing architectures that not only emulate neural processes but do so in a manner that drastically reduces energy consumption while enhancing parallel processing capabilities.

A unique feature of the centre’s approach is its focus on photonic hardware—devices that use light to process and transmit information. Light-based processors have the potential to revolutionize computing speed and energy efficiency owing to their inherent advantages in data bandwidth and signal propagation velocity compared to electronic counterparts. By integrating insights from living human neurons, the researchers aspire to engineer photonic systems capable of delivering unparalleled performance on AI workloads with significantly lower power requirements. This technology could mark a transformational shift in how artificial intelligence systems are architected, addressing both scalability and environmental sustainability challenges.

The scientific team also leverages human induced pluripotent stem cell (hiPSC) technologies, allowing for the cultivation and study of living human neurons under controlled laboratory conditions. By analyzing neuronal behavior at the cellular level, the researchers can derive fundamental principles of brain computation that inform the design of new algorithms and architectural paradigms. This biohybrid approach—melding biology with photonic and computational engineering—is poised to generate novel hardware-software co-design strategies, representing a leap beyond the incremental improvements typical of current neuromorphic hardware developments.

Energy sustainability is a critical component of the centre’s mission. Contemporary AI infrastructures exhibit rapidly escalating power consumption, which threatens the scalability and long-term viability of digital technologies worldwide. Unlike the human brain, which consumes approximately 20 watts to perform extraordinarily complex cognitive tasks, traditional computing systems consume kilowatts or more for comparable AI functions. By grounding their innovation in biological principles and emerging photonic technologies, the centre aims to redefine energy benchmarks for next-generation AI computation, making dramatic reductions in carbon footprint achievable.

The establishment of this centre transcends mere technology development; it seeks to catalyse the formation of a vibrant, long-lasting UK research ecosystem dedicated to neuromorphic computing. This ecosystem will foster industry-university collaboration, knowledge exchange, and international partnerships, ensuring that innovation continues well beyond the initial funding period. Industrial partners such as Microsoft Research, Nokia Bell Labs, Hewlett Packard Labs, and others will actively participate, enriching the research context while expediting the deployment of cutting-edge neuromorphic technologies in diverse application domains.

At the helm is Professor Sergei K. Turitsyn, director of AIPT and the centre, whose vision emphasizes not only technological breakthroughs but also the creation of a nationally recognized research brand in neuromorphic computing. This recognition aims to attract the best academic and industrial minds, bolstering the UK’s global competitiveness in an area deemed critical for future digital infrastructure resilience. Professor Turitsyn highlights the centre’s potential to unify disciplines, bridging photonics, neuroscience, materials science, and computer science for a holistic, systems-oriented research approach.

Co-director and neurophysiologist Professor Rhein Parri underscores the novelty of combining living human neuronal studies with state-of-the-art computing technologies. This integration enables a deeper understanding of neural dynamics that can be translated into computational models and devices. Through this interdisciplinary lens, the centre hopes to foster entirely new AI architectures capable of mirroring brain-like flexibility and functional efficiency.

Professor Natalia Berloff from the University of Cambridge focuses on the photonic aspect, clarifying how light-based processors can exploit massive inherent parallelism and ultrafast signal propagation to surpass the capabilities of traditional electronic circuits. This combination opens pathways to AI hardware that consumes substantially less power and offers scalability essential for future-intensive workloads.

From the University of Southampton, Professor Dimitra Georgiadou stresses the importance of novel materials and device architectures that can precisely emulate neural computation and biological responses to stimuli. Overcoming the limitations of conventional electronics requires materials and techniques specifically designed to support neuromorphic functions at scale and with sustainability in mind.

Collectively, the centre’s interdisciplinary team aspires to enact a paradigm shift in computing. By drawing on biological brain principles, stem-cell technology, photonics, and advanced algorithms, it aims to break open new technological frontiers characterized by energy efficiency, scalability, and societal impact. This new era of neuromorphic computing promises not only to transform AI and digital infrastructure but also to reshape our understanding of computation itself.

In summary, the UK Multidisciplinary Centre for Neuromorphic Computing, led by Aston University and supported by a consortium of top UK institutions, represents a landmark effort to bridge biology and technology. It focuses on developing photonic neuromorphic platforms that promise leaps in energy efficiency and computational capabilities. Supported by industry giants and underpinned by fundamental research, the initiative seeks to establish the UK as a global nexus for sustainable, brain-inspired computing innovation.

Subject of Research: Neuromorphic computing, photonic hardware, brain-inspired energy-efficient computing technologies.

Article Title: Aston University Leads UK Multidisciplinary Centre to Revolutionize Brain-Inspired, Energy-Efficient Computing

News Publication Date: [Not specified in the provided content]

Web References:

• https://www.aston.ac.uk/research/eps/aipt/neuromorphic-computing-centre
• https://research.aston.ac.uk/en/persons/sergei-turitsyn
• https://research.aston.ac.uk/en/persons/rhein-parri

Image Credits: Professor Sergei K. Turitsyn

Keywords

Neuromorphic computing, photonics, brain-inspired computing, energy-efficient computing, artificial intelligence, stem cell technology, computational neuroscience, photonic hardware, sustainable digital infrastructure, interdisciplinary research, UK research centre, UKRI EPSRC.