A5101909496- Advanced Memory and Neural Computing
- Ferroelectric and Negative Capacitance Devices
- Multimodal Machine Learning Applications
- Domain Adaptation and Few-Shot Learning
- Machine Learning and ELM
- Neural dynamics and brain function
- Advanced Image and Video Retrieval Techniques
- Neuroscience and Neural Engineering
- Advanced Neural Network Applications
- Distributed and Parallel Computing Systems
- Human Pose and Action Recognition
- Topic Modeling
- Video Analysis and Summarization
- Power Systems and Technologies
- Neural Networks and Reservoir Computing
- Neural Networks and Applications
- Smart Grid Security and Resilience
Peking University
2017-2019
Institute of Microelectronics
2017-2019
Megvii (China)
2017
In this paper, a hardware-realized neuromorphic system for pattern recognition is presented. The directly captures images from the environment, and then conducts classification using single layer neural network. Metal-oxide resistive random access memory (RRAM) used as electronic synapses, threshold-controlled neurons are proposed postsynaptic to save area simplify operation. neuron, no capacitor utilized, which contributes higher integration density. total energy consumption of RRAM...
Abstract Resistive switching memory (RRAM) is considered as one of the most promising devices for parallel computing solutions that may overcome von Neumann bottleneck today’s electronic systems. However, existing RRAM-based architectures suffer from practical problems such device variations and extra circuits. In this work, we propose a novel architecture pattern recognition by implementing k -nearest neighbor classification on metal-oxide RRAM crossbar arrays. Metal-oxide with gradual...
Objects appear to scale differently in natural images. This fact requires methods dealing with object-centric tasks (e.g. object proposal) have robust performance over variances scales. In the paper, we present a novel segment proposal framework, namely FastMask, which takes advantage of hierarchical features deep convolutional neural networks multi-scale objects one shot. Innovatively, adapt network into three different functional components (body, neck and head). We further propose...
We study the problem of grounding distributional representations texts on visual domain, namely visual-semantic embeddings (VSE for short). Begin with an insightful adversarial attack VSE embeddings, we show limitation current frameworks and image-text datasets (e.g., MS-COCO) both quantitatively qualitatively. The large gap between number possible constitutions real-world semantics size parallel data, to a extent, restricts model establish link textual concepts. alleviate this by augmenting...
A novel leaky integrate-and-fire (LIF) neuron circuit based on the gradual switching in resistive random access memory (RRAM) device is put forward, which threshold modulation can be achieved. Its and spike generating functions are verified through HSPICE simulation. In unsupervised pattern recognition for handwritten digits MNIST dataset, its advantage improving accuracy (from about 70% to more than 95%) demonstrated. Benchmarking results indicate that this much faster save 66% area...
Objects appear to scale differently in natural images. This fact requires methods dealing with object-centric tasks (e.g. object proposal) have robust performance over variances scales. In the paper, we present a novel segment proposal framework, namely FastMask, which takes advantage of hierarchical features deep convolutional neural networks multi-scale objects one shot. Innovatively, adapt network into three different functional components (body, neck and head). We further propose...
Hardware binary neural networks (BNNs) based on resistive random access memory (RRAM) are designed and investigated in this work. RRAM devices that work mode used as electronic synapses. The simulation results indicate the BNNs can achieve an accuracy of 94% MNIST database, show remarkable tolerance to non-ideal properties RRAM-based
A spike-rate-dependent plasticity (SRDP) protocol was experimentally verified in HfOx-based RRAM array. grey pattern recognition system is designed to demonstrate the potential application of brain-inspired computing.
A novel RRAM-based pattern recognition system with locally inhibited post-neurons is developed. The able to learn the whole MNIST training set (60,000 patterns). By using system, same post-neuron fired by similar patterns in class, which causes reduction of hardware cost. With post-neuron, can achieve more than 90.73% accuracy.