Spatiotemporal audio feature extraction with dynamic memristor-based time-surface neurons

Science Advances, Journal Year: 2024, Volume and Issue: 10(14)

Published: April 3, 2024

Neuromorphic speech recognition systems that use spiking neural networks (SNNs) and memristors are progressing in hardware development. The conventional manual preprocessing of audio signals is shifting toward event-based with convolutional SNNs. Despite achieving high accuracy classification, the efficient extraction spatiotemporal features from events continues to be a substantial challenge. In this study, we introduce dynamic time-surface neurons (DTSNs) using volatile featuring an adjustable temporal kernel decay, enabled by series-connected transistors Au/LiCoO 2 /Au configuration. DTSNs act as feature descriptors, enhancing event data. A two-layer SNN classifier, fully connected incorporating 1T1R nonvolatile memristor array, trained recognize Our findings show classification accuracies up 95.91%, improvements computational efficiency, increased noise resilience, confirming promise our memristor-based system for practical applications.

Language: Английский

---

Context gating in spiking neural networks: Achieving lifelong learning through integration of local and global plasticity

Knowledge-Based Systems, Journal Year: 2025, Volume and Issue: unknown, P. 112999 - 112999

Published: Jan. 1, 2025

Language: Английский

---

Hybrid neural networks for continual learning inspired by corticohippocampal circuits

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

Published: Feb. 2, 2025

Current artificial systems suffer from catastrophic forgetting during continual learning, a limitation absent in biological systems. Biological mechanisms leverage the dual representation of specific and generalized memories within corticohippocampal circuits to facilitate lifelong learning. Inspired by this, we develop circuits-based hybrid neural network (CH-HNN) that emulates these representations, significantly mitigating both task-incremental class-incremental learning scenarios. Our CH-HNNs incorporate networks spiking networks, leveraging prior knowledge new concept through episode inference, offering insights into functions feedforward feedback loops circuits. Crucially, CH-HNN operates as task-agnostic system without increasing memory demands, demonstrating adaptability robustness real-world applications. Coupled with low power consumption inherent SNNs, our model represents potential for energy-efficient, dynamic environments.

Language: Английский

---

Advancing brain-inspired computing with hybrid neural networks

National Science Review, Journal Year: 2024, Volume and Issue: 11(5)

Published: Feb. 26, 2024

ABSTRACT Brain-inspired computing, drawing inspiration from the fundamental structure and information-processing mechanisms of human brain, has gained significant momentum in recent years. It emerged as a research paradigm centered on brain–computer dual-driven multi-network integration. One noteworthy instance this is hybrid neural network (HNN), which integrates computer-science-oriented artificial networks (ANNs) with neuroscience-oriented spiking (SNNs). HNNs exhibit distinct advantages various intelligent tasks, including perception, cognition learning. This paper presents comprehensive review an emphasis their origin, concepts, biological perspective, construction framework supporting systems. Furthermore, insights suggestions for potential directions are provided aiming to propel advancement HNN paradigm.

Language: Английский

---

A Survey on Continual Semantic Segmentation: Theory, Challenge, Method and Application

IEEE Transactions on Pattern Analysis and Machine Intelligence, Journal Year: 2024, Volume and Issue: 46(12), P. 10891 - 10910

Published: Aug. 21, 2024

Continual learning, also known as incremental learning or life-long stands at the forefront of deep and AI systems. It breaks through obstacle one-way training on close sets enables continuous adaptive open-set conditions. In recent decade, continual has been explored applied in multiple fields especially computer vision covering classification, detection segmentation tasks. semantic (CSS), which dense prediction peculiarity makes it a challenging, intricate burgeoning task. this paper, we present review CSS, committing to building comprehensive survey problem formulations, primary challenges, universal datasets, neoteric theories multifarious applications. Concretely, begin by elucidating definitions challenges. Based an in-depth investigation relevant approaches, sort out categorize current CSS models into two main branches including data-replay data-free sets. each branch, corresponding approaches are similarity-based clustered thoroughly analyzed, following qualitative comparison quantitative reproductions datasets. Besides, introduce four specialities with diverse application scenarios development tendencies. Furthermore, develop benchmark for encompassing representative references, evaluation results reproductions. We hope can serve reference-worthy stimulating contribution advancement field, while providing valuable perspectives related fields.

Language: Английский

---

Domino-like water film manipulation with multifunctionality

Applied Physics Letters, Journal Year: 2024, Volume and Issue: 125(5)

Published: July 29, 2024

Domino effect is widely known and intuitively understood. Although the concept frequently used, a few works combine it with liquid manipulation. Liquid manipulation essential in many fields; however, large-scale using minimal forces still challenge. Here, we show domino-like process triggered by wind on heterogeneously wettable surfaces. This was demonstrated velocities of between 2.2 3.0 m/s structured surfaces containing water film thickness range 2.5–4.5 mm. The domino dewetting were shown various patterned designs 32–224 mm length; under ideal conditions, could be infinitely transmissible. Such might apply to long-distance directional transportation floats, bed bottom dust cleaning. Other designs, such as branched tree structure, can drive larger objects, remote circuit interrupters shown. method provides an approach for movement tiny toward multifunctionality.

Language: Английский

---

A brain-inspired model for multi-step forecasting of malignant arrhythmias

Expert Systems with Applications, Journal Year: 2025, Volume and Issue: unknown, P. 126373 - 126373

Published: Jan. 1, 2025

Language: Английский

---

Large language models and brain-inspired general intelligence

National Science Review, Journal Year: 2023, Volume and Issue: 10(10)

Published: Sept. 4, 2023

---

AI meets physics: a comprehensive survey

Artificial Intelligence Review, Journal Year: 2024, Volume and Issue: 57(9)

Published: Aug. 16, 2024

Uncovering the mechanisms of physics is driving a new paradigm in artificial intelligence (AI) discovery. Today, has enabled us to understand AI wide range matter, energy, and space-time scales through data, knowledge, priors, laws. At same time, also draws on introduces knowledge laws promote its own development. Then this using physical science inspire (PhysicsScience4AI, PS4AI). Although become force for development various fields, there still "black box" phenomenon that difficult explain field deep learning. This article will briefly review connection between relevant disciplines (classical mechanics, electromagnetism, statistical physics, quantum mechanics) AI. It focus discussing how they learning paradigm, introduce some related work solves problems. PS4AI research field. end article, we summarize challenges facing physics-inspired look forward next generation technology. aims provide brief algorithms stimulate future exploration by elucidating recent advances physics.

Language: Английский

---

Brain-inspired reward broadcasting: Brain learning mechanism guides learning of spiking neural network

Neurocomputing, Journal Year: 2025, Volume and Issue: unknown, P. 129664 - 129664

Published: Feb. 1, 2025

Language: Английский

---