A5061225111- Advanced Memory and Neural Computing
- Ferroelectric and Negative Capacitance Devices
- Neural dynamics and brain function
- Neural Networks and Reservoir Computing
- Advanced Algorithms and Applications
- Advanced Image and Video Retrieval Techniques
- Robotics and Automated Systems
- Robotics and Sensor-Based Localization
- Neuroscience and Neural Engineering
- Photoreceptor and optogenetics research
- Advanced Sensor and Control Systems
- Adversarial Robustness in Machine Learning
- Spectroscopy and Chemometric Analyses
- Image Retrieval and Classification Techniques
- Infrared Target Detection Methodologies
- Air Quality and Health Impacts
- Face and Expression Recognition
- Advanced Graph Neural Networks
- Video Surveillance and Tracking Methods
- Atmospheric chemistry and aerosols
- Plant responses to water stress
- Education and Work Dynamics
- Soil, Finite Element Methods
- Advanced Computational Techniques and Applications
- Medical Research and Treatments
Hong Kong Polytechnic University
2024-2025
Graz University of Technology
2021-2024
Tsinghua University
2018-2023
Princeton University
2023
Princeton Public Schools
2023
University of Electronic Science and Technology of China
2007-2021
Beijing Advanced Sciences and Innovation Center
2021
King Center
2021
National Chiayi University
2021
Chinese Academy of Sciences
2018
Compared with artificial neural networks (ANNs), spiking (SNNs) are promising to explore the brain-like behaviors since spikes could encode more spatio-temporal information. Although pre-training from ANN or direct training based on backpropagation (BP) makes supervised of SNNs possible, these methods only exploit networks' spatial domain information which leads performance bottleneck and requires many complicated skills. Another fundamental issue is that spike activity naturally...
Spiking neural networks (SNNs) that enables energy efficient implementation on emerging neuromorphic hardware are gaining more attention. Yet now, SNNs have not shown competitive performance compared with artificial (ANNs), due to the lack of effective learning algorithms and programming frameworks. We address this issue from two aspects: (1) propose a neuron normalization technique adjust selectivity develop direct algorithm for deep SNNs. (2) Via narrowing rate coding window converting...
Spiking neural networks (SNNs) are promising in a bio-plausible coding for spatio-temporal information and event-driven signal processing, which is very suited energy-efficient implementation neuromorphic hardware. However, the unique working mode of SNNs makes them more difficult to train than traditional networks. Currently, there two main routes explore training deep with high performance. The first convert pre-trained ANN model its SNN version, usually requires long window convergence...
There are two principle approaches for learning in artificial intelligence: error-driven global and neuroscience-oriented local learning. Integrating them into one network may provide complementary capabilities versatile scenarios. At the same time, neuromorphic computing holds great promise, but still needs plenty of useful algorithms algorithm-hardware co-designs to fully exploit its advantages. Here, we present a global-local synergic model by introducing brain-inspired meta-learning...
Place recognition is an essential spatial intelligence capability for robots to understand and navigate the world. However, recognizing places in natural environments remains a challenging task because of resource limitations changing environments. In contrast, humans animals can robustly efficiently recognize hundreds thousands different conditions. Here, we report brain-inspired general place system, dubbed NeuroGPR, that enables by mimicking neural mechanism multimodal sensing, encoding,...
It is widely believed the brain-inspired spiking neural networks have capability of processing temporal information owing to their dynamic attributes. However, how understand what kind mechanisms contributing learning ability and exploit rich properties satisfactorily solve complex computing tasks in practice still remains be explored. In this article, we identify importance capturing multi-timescale components, based on which a multi-compartment model with dendritic heterogeneity, proposed....
Despite the rapid progress of neuromorphic computing, inadequate capacity and insufficient representation power spiking neural networks (SNNs) severely restrict their application scope in practice. Residual learning shortcuts have been evidenced as an important approach for training deep networks, but rarely did previous work assessed applicability to specifics SNNs. In this article, we first identify that negligence leads impeded information flow accompanying degradation problem a version...
Toward the long-standing dream of artificial intelligence, two successful solution paths have been paved: 1) neuromorphic computing and 2) deep learning. Recently, they tend to interact for simultaneously achieving biological plausibility powerful accuracy. However, models from these domains run on distinct substrates, i.e., platforms learning accelerators, respectively. This architectural incompatibility greatly compromises modeling flexibility hinders promising interdisciplinary research....
Abstract Neuromorphic systems aim to implement large‐scale artificial neural network on hardware ultimately realize human‐level intelligence. The recent development of nonsilicon nanodevices has opened the huge potential full memristive networks (FMNN), consisting neurons and synapses, for neuromorphic applications. Unlike widely reported devices less progress. Sophisticated dynamics is major obstacle behind lagging. Here a rich dynamics‐driven neuron demonstrated, which successfully...
Spiking neural networks (SNNs) that enables energy efficient implementation on emerging neuromorphic hardware are gaining more attention. Yet now, SNNs have not shown competitive performance compared with artificial (ANNs), due to the lack of effective learning algorithms and programming frameworks. We address this issue from two aspects: (1) propose a neuron normalization technique adjust selectivity develop direct algorithm for deep SNNs. (2) Via narrowing rate coding window converting...
Orchestration of diverse synaptic plasticity mechanisms across different timescales produces complex cognitive processes. To achieve comparable complexity in memristive neuromorphic systems, devices that are capable to emulate short- and long-term (STP LTP, respectively) concomitantly essential. However, this fundamental bionic trait has not been reported any existing memristors where STP LTP can only be induced selectively because the inability decoupled using loci mechanisms. In work, we...
Abstract There is a growing trend to design hybrid neural networks (HNNs) by combining spiking and artificial leverage the strengths of both. Here, we propose framework for general computation HNNs introducing units (HUs) as linkage interface. The not only integrates key features these computing paradigms but also decouples them improve flexibility efficiency. HUs are designable learnable promote transmission modulation information flows in HNNs. Through three cases, demonstrate that can...
Recent advances in artificial intelligence have enhanced the abilities of mobile robots dealing with complex and dynamic scenarios. However, to enable computationally intensive algorithms be executed locally multitask low latency high efficiency, innovations computing hardware are required. Here, we report TianjicX, a neuromorphic that can support true concurrent execution multiple cross-computing-paradigm neural network (NN) models various coordination manners for robotics. With...
As well known, the huge memory and compute costs of both artificial neural networks (ANNs) spiking (SNNs) greatly hinder their deployment on edge devices with high efficiency. Model compression has been proposed as a promising technique to improve running efficiency via parameter operation reduction, whereas this is mainly practiced in ANNs rather than SNNs. It interesting answer how much an SNN model can be compressed without compromising its functionality, where two challenges should...
Spiking neural network (SNN) is broadly deployed in neuromorphic devices to emulate brain function. In this context, SNN security becomes important while lacking in-depth investigation. To end, we target the adversarial attack against SNNs and identify several challenges distinct from artificial (ANN) attack: 1) current mainly based on gradient information that presents a spatiotemporal pattern SNNs, hard obtain with conventional backpropagation algorithms; 2) continuous of input...
Although spiking neural networks (SNNs) take benefits from the bioplausible modeling, low accuracy under common local synaptic plasticity learning rules limits their application in many practical tasks. Recently, an emerging SNN supervised algorithm inspired by backpropagation through time (BPTT) domain of artificial (ANNs) has successfully boosted SNNs, and helped improve practicability SNNs. However, current general-purpose processors suffer efficiency when performing BPTT for SNNs due to...
Cross-modal matching shows enormous potential to recognize objects across different sensory modalities, which is fundamental numerous visual-language tasks like image-text retrieval and visual captioning. Existing works generally rely on massive well-aligned data pairs for model training. Unfortunately, multimodal datasets are extremely difficult annotate collect. As an alternative, the co-occurred collected from internet have been widely exploited train a cross-modal model. However,...
In early 2020, two unique events perturbed ship emissions of pollutants around Southern China, proffering insights into the impacts on regional air quality: decline activities due to COVID-19 and global enforcement low-sulfur (<0.5%) fuel oil for ships. January February estimated NOx, SO2, primary PM2.5 over China dropped by 19, 71, 58%, respectively, relative same period in 2019. The NOx was mostly coastal waters inland waterways reduced activities. SO2 most pronounced outside Chinese...
Nosiheptide is a prototypal thiopeptide antibiotic, containing an indole side ring in addition to its thiopeptide-characteristic macrocylic scaffold. This derived from 3-methyl-2-indolic acid (MIA), product of the radical S-adenosylmethionine enzyme NosL, but how MIA incorporated into nosiheptide biosynthesis remains be investigated. Here we report functional dissection series enzymes involved biosynthesis. We show NosI activates and transfers it phosphopantetheinyl arm carrier protein NosJ....