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

The work follows the idea that neural planning does not rely on brute-force computation to reach a solution. Instead, it uses principles observed in the brain—particularly mechanisms linked to the hippocampus—to generate candidate futures and refine actions toward a goal. “The brain works in a completely different way to today’s AI systems,” explains Wolfgang Maass from TU Graz, emphasizing an algorithmic translation of biological computation.

At the core of the approach are three interacting mechanisms. First, the model forms cognitive maps, converting relationships among abstract entities into geometric structure within neural codes—effectively providing a “sense of direction.” Second, it uses stochastic neural computations that continuously propose hypothetical scenarios, enabling the system to explore without calculating every pathway. Third, it applies compositional coding, breaking down information and action plans into reusable components.

Together, these mechanisms allow the model to “imagine” possible routes and test them in imagination before committing. Rather than searching exhaustively, the system samples intermediate steps: when a randomly selected move appears to align with the goal, guided by the cognitive map, that direction is pursued. At each new position, the model again evaluates options, gradually steering toward the target while maintaining flexibility.

A key claim is adaptability. Because planning is driven by sampling from cognitive maps and compositional representations, the system can respond to changed or newly introduced situations without requiring retraining. This capability contrasts with many conventional pipelines that must be retrained to handle new environmental structures or objectives.

To demonstrate the idea, the team evaluated the model on three challenges: navigating a two-dimensional space, orienting within an abstract multi-dimensional space, and assembling or disassembling a silhouette built from modular blocks. Across tasks, the system showed goal-directed planning behavior consistent with a sampling-and-map framework.

The researchers stress that the approach is not meant to replace today’s large language models. Instead, it proposes an alternative route for applications where efficient local decision-making matters. With further development, brain-inspired planning systems could broaden AI beyond cloud-scale compute.

Such energy-aware methods could be especially valuable for robots, autonomous vehicles, and edge devices—settings where hardware constraints demand low power operation. The study was conducted with collaborations including Tsinghua University and Italy’s National Research Council.

Subject of Research: Brain-inspired neural planning and cognitive-map-based sampling
Article Title: Neural sampling from cognitive maps enables goal-directed imagination and planning
News Publication Date: 21-Jul-2026
Web References: http://dx.doi.org/10.1038/s42256-026-01254-4
References: Nature Machine Intelligence (DOI: 10.1038/s42256-026-01254-4)
Image Credits: Lunghammer – TU Graz

Keywords

Brain-inspired AI; cognitive maps; neural sampling; stochastic planning; compositional coding; energy-efficient machine intelligence

At Lehigh University, associate professor Yevgeny Berdichevsky from the departments of bioengineering and electrical and computer engineering leads an innovative effort to bridge that divide. Recently awarded a $2 million grant from the National Science Foundation (NSF), his interdisciplinary team is pioneering research to unravel the complex information processing within the brain. Their goal: to harness the brain’s natural computational mechanisms within bioengineered neural organoids and inspire new, energy-efficient AI algorithms.

This ambitious project leverages cutting-edge techniques in tissue engineering. The core of the work revolves around brain organoids—miniature, three-dimensional structures cultivated from adult stem cells that imitate the developmental features of the human cortex. Unlike traditional two-dimensional cultures, these organoids provide a more realistic microenvironment to study neuronal behaviors and circuit dynamics. Yet, neurons in organoids often grow without spatial organization, limiting their computational mimicry of brain tissue.

To overcome this, Lesley W. Chow, an associate professor specializing in bioengineering and materials science, employs 3D-printed biomaterial scaffolds. These finely tuned structures serve as physical frameworks to guide neuron placement within organoids, orchestrating the formation of layered neural networks that mimic the ordered architecture of the human cortex. By inserting neural spheroids—small clusters of diverse neuron types—into pre-designed scaffold cavities and stacking these layers methodically, the team essentially engineers the organoid’s connectivity from the ground up.

But engineering the physical layout is only the first hurdle. Functional validation requires demonstrating that these organized neurons can perform dynamic computations akin to those our brains effortlessly execute. One such complex task is visual motion detection, currently approximated in machines through optical flow algorithms embedded in drone navigation and autonomous vehicle computer vision. These algorithms, despite recent advancements, remain suboptimal in energy efficiency and accuracy.

Berdichevsky’s approach capitalizes on the intrinsic dynamics of cortical neurons to surpass these limitations. By stimulating neurons directly with optical pulses—bypassing the eye entirely—his team encodes visual information into patterned light sequences projected onto targeted neurons. This technique mimics the brain’s natural transformation of photons into electrical signals but allows precise experimental manipulation at the cellular level.

Through microscopy, researchers record neuronal activity by tracking a genetically expressed fluorescent protein whose brightness fluctuates depending on neuron firing. This direct visualization of active neurons, mapped spatially and temporally, provides a rich dataset to decode how the neural network interprets motion. Collaborating with assistant professor Yuntao Liu, the team is developing sophisticated decoding algorithms and computational models to analyze fluorescence patterns. These tools will elucidate not only what the organoid “perceives” but also the velocity and directionality of moving stimuli.

The computational model serves an additional purpose: shaping protocols to train these organoids, enabling learning and adaptation much like neural plasticity in vivo. In doing so, the research embodies a feedback loop—biological computation informing artificial algorithms, which in turn refine engineered neural tissues.

Ethical considerations occupy a central role in this venture. Ally Peabody Smith, an assistant professor of community and population health, investigates the social and legal implications arising from using living neural tissues. Although the organoids remain far too simplistic and minuscule to exhibit consciousness, maintaining transparent ethical boundaries is paramount as bioengineered models increasingly approach functional complexity.

This multidisciplinary endeavor, blending computational neuroscience, bioengineering, materials science, and ethical scholarship, epitomizes the synthesis necessary to translate neural principles into transformative AI technology. As Berdichevsky explains, the integrated design is the project’s greatest strength: combining hardware-inspired neural networks with biologically precise “wetware” to achieve forms of computation that are simultaneously powerful and energy efficient.

If successful, these engineered organoids could offer a groundbreaking proof of concept—showing that biological tissues can execute computations traditionally reserved for silicon processors. This prospect holds the promise of revolutionizing AI architectures, reducing power consumption, and enabling machines to perform intricate tasks with brain-like facility.

As this research moves forward, it underscores a pivotal question at the frontier of science and engineering: can we not only emulate but also evolve the brain’s computing capabilities through biofabrication? The answers emerging from Lehigh University’s labs may well define the next era of intelligent machines.

Subject of Research: Bioengineered neural organoids for biological computation and energy-efficient artificial intelligence.

Article Title: Neural Organoids in 3D Scaffolds: Pioneering Energy-Efficient Biological Computation to Inspire Next-Generation AI

News Publication Date: Information not provided.

Web References:

• Lehigh University Faculty Profile: Yevgeny Berdichevsky
• NSF Award Abstract (#2515371)
• NSF 24-508: Emerging Frontiers in Research and Innovation (EFRI-2024/25)
• Lehigh University Faculty Profile: Lesley W. Chow
• Lehigh University Faculty Profile: Yuntao Liu
• Lehigh University College of Health Faculty: Ally Peabody Smith

Image Credits: Courtesy of Yevgeny Berdichevsky / Lehigh University

Keywords: Artificial intelligence, Organoids, Organ cultures, Neurons, Neural stem cells, Systems neuroscience, Neural networks, Engineering, Bioengineering, Electrical engineering, Neuroscience, Brain tissue, Brain

A pioneering breakthrough in this search comes from a team of engineers at Texas A&M University, led by Dr. Suin Yi, who has contributed to the development of what they term “Super-Turing AI.” This innovative model takes inspiration from the workings of the human brain, challenging the traditional paradigms of AI architecture that have dominated the field for decades. Rather than the typical separation of data storage and processing, Super-Turing AI seeks to integrate these functions, mirroring the interconnected neural processes that define human cognition.

The energy crisis surrounding today’s AI technologies cannot be overstated. Data centers housing leading AI applications, including prominent large language models like OpenAI’s ChatGPT, are colossal structures consuming power in gigawatts. This energy demand poses substantial economic and environmental challenges. The sustainability of AI technology must be addressed urgently, especially as its capabilities expand and integrate further into society. Advocates of this new approach emphasize that while AI models have remarkable abilities, their operational frameworks rely heavily on computing power that comes at a tremendous environmental cost.

Dr. Yi aims to change this narrative by employing principles derived from neuroscience in the design of AI systems. The fundamental issue at hand lies in the way traditional AI models operate—where training and memory are typically handled as distinct processes. Current paradigms force AI systems to execute separate training phases for learning and memory storage, followed by a cumbersome transfer of data across various hardware components. This not only elevates energy usage but also slows down processing times, creating inefficiencies that Super-Turing AI intends to remedy.

In contrast, the human brain processes learning and memory as an integrated function. The efficiency of the brain is largely due to its intricate network of neurons, which communicate via synapses that strengthen or weaken based on experience and learning—an essential feature known as synaptic plasticity. This biological mechanism allows for the optimization of cognitive processes and enables rapid adaptation and decision-making based on past experiences. By mirroring this natural efficiency in AI systems, Dr. Yi’s team hopes to reduce the computational burden that current models impose.

To illustrate the potential effectiveness of Super-Turing AI, the research team conducted a compelling experiment involving a drone tasked with navigating a challenging environment. The drone utilized a circuit built on principles derived from the new model, allowing it to learn and adapt in real-time without requiring extensive pre-training. The results were striking; the drone demonstrated heightened speed, improved efficiency, and significantly lower energy consumption compared to conventional AI approaches, showcasing the practical implications of integrating biological principles into artificial systems.

The relevance of this work extends beyond mere experimentation. The burgeoning AI industry finds itself at a critical juncture, where advancements in models such as Super-Turing AI could pave the way for sustainable development. Many companies have invested heavily in constructing expansive data centers to house increasingly powerful AI models, inadvertently driving up both economic and environmental costs. With energy constraints becoming a crucial limiting factor in scaling AI technologies, innovative hardware solutions are more important than ever.

Dr. Yi emphasizes that realizing the full potential of AI transcends software innovations alone; it necessitates equally transformative advancements in hardware technologies. The symbiotic relationship between software and hardware must be prioritized to facilitate the continuous evolution of effective and efficient AI systems. Without robust hardware to support the wherewithal of advanced algorithms, the future trajectory of artificial intelligence could be jeopardized.

The implications of this research resonate deeply within the conversation surrounding sustainable AI development. Super-Turing AI could represent a significant leap forward in designing eco-friendly AI architectures that harness the efficiency of human neural processes. Envisioning a future where AI technologies not only meet but also exceed current paradigms of performance—while being mindful of their energy footprints—aligns with growing public sentiment demanding accountability in technological advancements.

As society progresses towards greater reliance on AI, balancing performance with sustainability has never been more critical. From improved energy efficiency to reduced operational costs, Super-Turing AI holds promise as a path toward economic and environmental responsibility in a sector where the stakes are incredibly high. The urgency for such developments in AI cannot be overstated, particularly as global awareness of ecological sustainability continues to grow.

Dr. Yi’s unwavering belief is that the future of AI can and must be aligned with the best interests of both humanity and the planet. Innovations like Super-Turing AI offer a glimpse into a world where artificial intelligence operates within the sustainable parameters that nature has already established. As the team moves forward with their research, they aim to inspire both the scientific community and industry leaders to consider the long-term implications of their innovations on the environment and society.

In conclusion, Super-Turing AI stands not merely as a technical achievement, but as a cultural touchstone for the future of artificial intelligence. Its integration of biological principles could reshape the landscape of AI development, allowing the technology to thrive while also adhering to principles of sustainability. The path ahead remains bright, but it depends on how industry stakeholders approach this transformative potential.

Subject of Research: Super-Turing AI
Article Title: HfZrO-based synaptic resistor circuit for a Super-Turing intelligent system
News Publication Date: 28-Feb-2025
Web References: [To Be Confirmed]
References: [To Be Confirmed]
Image Credits: Texas A&M University College of Engineering

Keywords: Artificial Intelligence, Energy Efficiency, Sustainable AI, Super-Turing AI, Neural Processes, Circuit Development, Ecological Responsibility, Economic Costs, Hardware Innovation, Cognitive Processes, Synaptic Plasticity, Innovation in AI Technologies.