A5002992237- Neural dynamics and brain function
- Advanced Memory and Neural Computing
- Advanced Image and Video Retrieval Techniques
- Robotics and Sensor-Based Localization
- Visual perception and processing mechanisms
- Advanced Vision and Imaging
- Neural Networks and Applications
- Ferroelectric and Negative Capacitance Devices
- Domain Adaptation and Few-Shot Learning
- Retinal Development and Disorders
- Image Retrieval and Classification Techniques
- Speech and Audio Processing
- 3D Surveying and Cultural Heritage
- Hearing Loss and Rehabilitation
- Music and Audio Processing
- Neuroscience and Neural Engineering
- Remote-Sensing Image Classification
- Neurobiology and Insect Physiology Research
- Advanced Fluorescence Microscopy Techniques
- Image and Object Detection Techniques
- Memory and Neural Mechanisms
- Multimodal Machine Learning Applications
- EEG and Brain-Computer Interfaces
- Generative Adversarial Networks and Image Synthesis
- Color Science and Applications
University of Chinese Academy of Sciences
2018-2024
Institute of Automation
2014-2024
Chinese Academy of Sciences
2014-2024
Shandong Institute of Automation
2012-2022
Center for Excellence in Brain Science and Intelligence Technology
2022
Beijing Academy of Artificial Intelligence
2019
Spiking neural networks (SNNs) serve as a promising computational framework for integrating insights from the brain into artificial intelligence (AI). Existing software infrastructures based on SNNs exclusively support simulation or brain-inspired AI, but not both simultaneously. To decode nature of biological and create we present cognitive engine (BrainCog). This SNN-based platform provides essential infrastructure developing AI simulation. BrainCog integrates different neurons, encoding...
Although feature-based methods have been successfully developed in the past decades for registration of optical images, and synthetic aperture radar (SAR) images is still a challenging problem remote sensing. In this letter, an improved version scale-invariant feature transform first proposed to obtain initial matching features from SAR images. Then, are refined by exploring their spatial relationship. The matches finally used estimating parameters. Experimental results shown effectiveness method.
This paper performs a comprehensive and comparative evaluation of the state-of-the-art local features for task image-based 3D reconstruction. The evaluated cover recently developed ones by using powerful machine learning techniques elaborately designed handcrafted features. To obtain evaluation, we choose to include both float type binary ones. Meanwhile, two kinds datasets have been used in this evaluation. One is dataset many different scene types with groundtruth points, containing images...
Feature description for local image patch is widely used in computer vision. While the conventional way to design descriptor based on expert experience and knowledge, learning-based methods designing become more popular because of their good performance data-driven property. This paper proposes a novel method binary feature descriptor, which we call receptive fields (RFD). Technically, RFD constructed by thresholding responses set fields, are selected from large number candidates according...
Extracting robust and discriminative local features from images plays a vital role for long term visual localization, whose challenges are mainly caused by the severe appearance differences between matching due to day-night illuminations, seasonal changes, human activities. Existing solutions resort jointly learning both keypoints their descriptors in an end-to-end manner, leveraged on large number of annotations point correspondence which harvested structure motion depth estimation...
Spiking Neural Networks (SNNs) serve as an appropriate level of abstraction to bring inspirations from brain and cognition Artificial Intelligence (AI). Existing software frameworks separately develop SNNs-based simulation brain-inspired intelligence infrastructures. However, the community requires open-source platform that can simultaneously support building integrating computational models for AI at multiple scales. To enable scientific quest decode nature biological create AI, we present...
Spiking neural network (SNN) has been attached to great importance due the properties of high biological plausibility and low energy consumption on neuromorphic hardware. As an efficient method obtain deep SNN, conversion exhibited performance various large-scale datasets. However, it typically suffers from severe degradation time delays. In particular, most previous work focuses simple classification tasks while ignoring precise approximation ANN output. this paper, we first theoretically...
Local image descriptors are one of the key components in many computer vision applications. Recently, binary have received increasing interest community for its efficiency and low memory cost. The similarity is measured by Hamming distance which has equal emphasis on all elements descriptors. This paper improves performance learning a weighted with larger weights assigned to more discriminative elements. What more, can be computed as fast basis pre-computed look-up-table. Therefore, proposed...
Spiking neural networks (SNNs) are rich in spatio-temporal dynamics and suitable for processing event-based neuromorphic data. However, datasets usually less annotated than static datasets. This small data scale makes SNNs prone to overfitting limits their performance. In order improve the generalization ability of on datasets, we use images assist SNN training event this paper, first discuss domain mismatch problem encountered when directly transferring trained We argue that inconsistency...
Compared to computer vision systems, the human visual system is more fast and accurate. It well accepted that V1 neurons can encode contour information. There are plenty of computational models about detection based on mechanism neurons. Multiple-cue inhibition operator one well-known model, which neurons' non-classical receptive fields. However, this model time-consuming noisy. To solve these two problems, we propose an improved integrates some additional other mechanisms primary system....
Deep convolutional neural networks (CNN) has achieved state-of-the-art result on traffic sign classification, which plays a key role in intelligent transportation system. However, it usually requires large number of labeled training data, is not always available, to guarantee good performance. In this paper, we propose synthesize images by generative adversarial (GANs). It takes standard template and background image as input the network GANs, where defines class include controls visual...
Dodging emergent dangers is an innate cognitive ability for animals, which helps them to survive in the natural environment. The retina-superior colliculus (SC)-pulvinar- amygdala-periaqueductal gray pathway responsible visual fear responses, and it able quickly detect looming obstacles dodging. Inspired by mechanism of responses pathway, we propose a brain-inspired obstacle dodging method model functions related brain regions. This first detects moving direction speed salient point objects...
Binary features have been incrementally popular in the past few years due to their low memory footprints and efficient computation of Hamming distance between binary descriptors. They shown with promising results on some real time applications, e.g., SLAM, where matching operations are relative few. However, computer vision, there many applications such as 3D reconstruction requiring lots local features. Therefore, a natural question is that feature still solution this kind applications? To...
The audio-visual event localization task requires identifying concurrent visual and auditory events from unconstrained videos within a network model, locating them, classifying their category. efficient extraction integration of audio modal information have always been challenging in this field. In paper, we introduce CACE-Net, which differs most existing methods that solely use signals to guide information. We propose an co-guidance attention mechanism allows for adaptive bi-directional...
Biological brains have the capability to adaptively coordinate relevant neuronal populations based on task context learn continuously changing tasks in real-world environments. However, existing spiking neural network-based continual learning algorithms treat each equally, ignoring guiding role of different similarity associations for network learning, which limits knowledge utilization efficiency. Inspired by context-dependent plasticity mechanism brain, we propose a Similarity-based...
Spike-based neuromorphic hardware has demonstrated substantial potential in low energy consumption and efficient inference. However, the direct training of deep spiking neural networks is challenging, conversion-based methods still require time delay owing to unresolved conversion errors. We determine that primary source errors stems from inconsistency between mapping relationship traditional activation functions input-output dynamics spike neurons. To counter this, we introduce Consistent...
Spiking Neural Networks (SNNs) can do inference with low power consumption due to their spike sparsity. ANN-SNN conversion is an efficient way achieve deep SNNs by converting well-trained Artificial (ANNs). However, the existing methods commonly use constant threshold for conversion, which prevents neurons from rapidly delivering spikes deeper layers and causes high time delay. In addition, same response different inputs may result in information loss during transmission. Inspired biological...
The human brain can self-organize rich and diverse sparse neural pathways to incrementally master hundreds of cognitive tasks. However, most existing continual learning algorithms for deep artificial spiking networks are unable adequately auto-regulate the limited resources in network, which leads performance drop along with energy consumption rise as increase In this paper, we propose a brain-inspired algorithm adaptive reorganization pathways, employs Self-Organizing Regulation reorganize...
Spiking neural networks (SNNs) have attracted extensive attentions in Brain-inspired Artificial Intelligence and computational neuroscience. They can be used to simulate biological information processing the brain at multiple scales. More importantly, SNNs serve as an appropriate level of abstraction bring inspirations from cognition Intelligence. In this paper, we present Cognitive Engine (BrainCog) for creating brain-inspired AI simulation models. BrainCog incorporates different types...