A5100623943- Advanced Memory and Neural Computing
- Parallel Computing and Optimization Techniques
- Interconnection Networks and Systems
- Embedded Systems Design Techniques
- Neural Networks and Reservoir Computing
- Advanced Data Storage Technologies
- Ferroelectric and Negative Capacitance Devices
- CCD and CMOS Imaging Sensors
- Neural dynamics and brain function
- Distributed and Parallel Computing Systems
- Neuroscience and Neural Engineering
- Advanced Neural Network Applications
- Photoreceptor and optogenetics research
- VLSI and Analog Circuit Testing
- Neural Networks and Applications
- Radiation Effects in Electronics
- Embedded Systems and FPGA Design
- Stochastic Gradient Optimization Techniques
- Distributed systems and fault tolerance
- Photonic and Optical Devices
- Software-Defined Networks and 5G
- Machine Learning and ELM
- Quantum Information and Cryptography
- Opportunistic and Delay-Tolerant Networks
- Formal Methods in Verification
National University of Defense Technology
2015-2024
Shanghai Lixin University of Accounting and Finance
2021-2022
Xianyang Normal University
2012-2021
Changsha University
2014-2021
Zhejiang University
2019
Shandong University
2018
Fudan University
2018
Institute of Software
2016
Institute of Microelectronics
2016
University of Defence
2012
Spike-timing dependent plasticity (STDP)-based spiking neural network (SNN) is a promising choice to realize unsupervised intelligent systems with limited power budget. In addition STDP, another two bio-inspired mechanisms of lateral inhibition and homeostasis are always implemented in the training procedure STDP-based SNNs. However, existing methods achieve necessitate great number connections that proportional square learning neurons, hardware solution demands complex circuits for each...
Spiking neural network (SNN), as the third generation of artificial networks, has been widely adopted in vision and audio tasks. Nowadays, many neuromorphic platforms support SNN simulation adopt Network-on-Chips (NoC) architecture for multi-cores interconnection. However, a large volume run-time communication on interconnection significant effect performance platform. In this paper, we propose toolchain called SNEAP (Spiking NEural mAPping toolchain) mapping SNNs to with multi-cores, which...
Abstract Background Protein protein interactions (PPIs) are essential to most of the biological processes. The prediction PPIs is beneficial understanding functions and thus helpful pathological analysis, disease diagnosis drug design etc. As amount data growing fast in post genomic era, high-throughput experimental methods expensive time-consuming for PPIs. Thus, computational have attracted researcher’s attention recent years. A large number been proposed based on different sequence...
Neuromorphic processors have gained momentum recently due to their high energy efficiency in artificial intelligence applications compared DNN accelerators. Most neuromorphic are executing SNNs (Spiking Neural Networks). Liquid State Machine (LSM), as the spiking version of reservoir computing, shows advantages and great potential image classification, speech recognition, language translation, etc.. Comparing with other SNN models, LSM has characteristics easy train low resource utilization,...
Visual querying is essential for interactively exploring massive trajectory data. However, the data uncertainty imposes profound challenges to fulfill advanced analytics requirements. On one hand, many underlying does not contain accurate geographic coordinates, e.g., positions of a mobile phone only refer regions (i.e., cell stations) in which it resides, instead GPS coordinates. other domain experts and general users prefer natural way, such as using language sentence, access analyze...
The increasing control complexity of Noisy Intermediate-Scale Quantum (NISQ) systems underlines the necessity integrating quantum hardware with software. While mapping heterogeneous quantum-classical computing (HQCC) algorithms to NISQ for execution, we observed a few dissatisfactions in programming languages (QPLs), including difficult hardware, limited expressiveness, and counter-intuitive code. In addition, noisy qubits require repeatedly performed experiments, which explicitly operate...
Coverage problem is a fundamental issue in wireless ad hoc and sensor networks. Previous techniques for coverage scheduling often require accurate location information or range measurements, which cannot be easily obtained resource-limited Recently, method based on algebraic topology proposed to achieve verification using only connectivity information. The topological sheds some light the of location-free coverage. Unfortunately, needs centralized computation rigorous restriction sensing...
The training of modern deep learning neural network calls for large amounts computation, which is often provided by GPUs or other specific accelerators. To scale out to achieve faster speed, two update algorithms are mainly applied in the distributed process, i.e., Synchronous SGD algorithm (SSGD) and Asynchronous (ASGD). SSGD obtains good convergence point while speed slowed down synchronous barrier. ASGD has but lower when compared SSGD. sufficiently utilize advantages ASGD, we propose a...
Abstract Background Essential proteins are indispensable to the development and survival of cells. The identification essential not only is helpful for understanding minimal requirements cell survival, but also has practical significance in disease diagnosis, drug design medical treatment. With rapidly amassing protein–protein interaction (PPI) data, computationally identifying from networks (PINs) becomes more popular. Up now, a number various approaches protein based on PINs have been...
Deep neural network (DNN) learns hierarchical representations from big data in a multilayer structure and has achieved great successes many fields such as computer vision speech analysis. Since DNN usually contains several billions of parameters, the asynchronous stochastic gradient descent (ASGD) algorithm is often used to train an effective model on cluster. However, increase computing nodes size, ASGD suffers serious slow convergence deficiency because parameters might be wrongly updated...
Neuromorphic event-based dynamic vision sensors (DVS) have much faster sampling rates and a higher range than frame-based imagers. However, they are sensitive to background activity (BA) events which unwanted. We propose HashHeat, hashing-based BA filter with O(C) complexity. It is the first spatiotemporal that doesn't scale DVS output size N store 32-bits timestamps. HashHeat consumes 100x less memory increases signal noise ratio by 15x compared previous designs.
Clock domain crossing (CDC) is an important issue in integrated circuit (IC) design and verification. In this paper, We present the of 5 types CDC schemes our developed SOC chip with multi working mode ten clock domain, deeply describe approach using assertion-based verification (ABV) to verify proper functionality for signals. Taped sample tests show all designs work right, demonstrate that method are effective.
Explaining the causes of infeasibility formulas has practical applications in various fields, such as formal verification and electronic design automation. A minimal unsatisfiable subformula provides a succinct explanation is valuable for applications. The problem deriving cores from Boolean been addressed rather frequently recent years. However little attention concentrated on extraction subformulas Satisfiability Modulo Theories(SMT). In this paper, we propose depth-firstsearch algorithm...
With the fast development of deep learning (DL), communication is increasingly a bottleneck for distributed workloads, and series optimization works have been done to scale out successfully. Nevertheless, network behavior has not investigated much yet. We intend analyze then carry some research through simulation. Under this circumstance, an accurate measurement necessary, as it effective way study basis Therefore, we propose capture (DLC) trace achieve measurement. To best our knowledge,...
Hardware/software partitioning is a crucial problem in hardware/software co-design. In this paper, we deeply investigate genetic algorithm (GA) for partitioning, our co-design targets heterogeneous multicore system on chip (SoC) which consists of several different types processing engines(PE), Communicating structure adopts NOC, We use GA four task graphs to simulate the experiments show method an effective algorithm.