A5101599355- Parallel Computing and Optimization Techniques
- Embedded Systems Design Techniques
- Low-power high-performance VLSI design
- Robotics and Sensor-Based Localization
- Radiation Effects in Electronics
- Interconnection Networks and Systems
- Infrared Target Detection Methodologies
- Video Surveillance and Tracking Methods
- Stability and Control of Uncertain Systems
- VLSI and Analog Circuit Testing
- Image Retrieval and Classification Techniques
- Neural Networks and Applications
- Advanced Vision and Imaging
- Fault Detection and Control Systems
- Robotic Path Planning Algorithms
- Face and Expression Recognition
- Medical Image Segmentation Techniques
- Advanced Optical Sensing Technologies
- Multimodal Machine Learning Applications
- Indoor and Outdoor Localization Technologies
- Advanced Neural Network Applications
- Turbomachinery Performance and Optimization
- Adaptive Control of Nonlinear Systems
- CCD and CMOS Imaging Sensors
- 3D IC and TSV technologies
PLA Academy of Military Science
2024
Soochow University
2019-2024
Tianjin University
2024
Anhui University of Technology
2023-2024
Hewlett-Packard (United States)
2024
Southeast University
2022
Ningbo University
2022
Chongqing University of Posts and Telecommunications
2021
Universidade de Sorocaba
2021
Beihang University
2007-2020
A fine-grain dynamic power gating is proposed for saving the leakage in MIPS R3000 by sleep control and applied to a processor pipeline. An execution unit divided into four small units: multiplier, divider, shifter other (CLU). The of each cut off dynamically, based on operation. We tape-outed prototype chip Geyser-0, which provides an Core with reduction technique, 16 KB caches translation lookaside buffer (TLB) using 90 nm CMOS technology. evaluation results benchmark programs embedded...
Geyser-1, a prototype MIPS R3000 CPU with fine grain runtime PG for major computational components in the execution stage is available. Function units such as CLU, shifter, multiplier and divider are power-gated controlled at that only function unit to be used powered-on minimize leakage power. The evaluation results on real chip reveals mechanism works without electric problems. It reduces power 7% 25°C 24% 80°C. using benchmark programs show consumption can reduced from 3% 30%
Virtual try-on methods based on diffusion models achieve realistic effects. They use an extra reference network or additional image encoder to process multiple conditional inputs, which results in high training costs. Besides, they require more than 25 inference steps, bringing a long time. In this work, with the development of transformer (DiT), we rethink necessity encoder, then propose MC-VTON, enabling DiT integrate minimal inputs by utilizing its intrinsic backbone. Compared existing...
A new adaptive modeling method for aircraft engine by using equilibrium manifold (EM) and its expansion (EME) model is presented, following research undertaken the authors at School of Transportation Science Engineering, Beijing University Aeronautics Astronautics, Beijing, China. The property effect mapping design to form are systematically studied. adaptivity analysis discussed, this paper also gives identification procedure approximate nonlinear model; deterioration modification...
Stable operation of aircraft engine compressions is constrained by rotating surge. In this paper, an approximate nonlinear surge margin model compression system using equilibrium manifold presented. Firstly, paper gives overview the current state modeling aerodynamic flow instabilities in compressors. Secondly, expansion form introduced, and choosing scheduling variable method discussed. Then, also identification procedure model. Finally, simulations with high pressure (HP) compressor show...
Beginning with VisualGLM and CogVLM, we are continuously exploring VLMs in pursuit of enhanced vision-language fusion, efficient higher-resolution architecture, broader modalities applications. Here propose the CogVLM2 family, a new generation visual language models for image video understanding including CogVLM2, CogVLM2-Video GLM-4V. As an model, inherits expert architecture improved training recipes both pre-training post-training stages, supporting input resolution up to $1344 \times...
In this Letter, we demonstrate a 1×4 low-crosstalk silicon photonics cascaded arrayed waveguide grating, which is fabricated on silicon-on-insulator wafer with 220-nm-thick top layer and 2 μm buried oxide layer. The measured on-chip transmission loss of grating ～4.0 dB, the fiber-to-waveguide coupling 1.8 dB/facet. channel spacing 6.4 nm. adjacent crosstalk by characterization very low, only ?33.2 dB. Compared to normal single ～?12.5 more than 20 dB dramatically improved in device.
Chromosome classification is an important way of clinical genetic diagnosis. However, it usually done manually by skilled cytologists, which requires extensive experience, domain knowledge, and tedious manual work. In this paper, we tackle the task chromosome utilizing deep learning techniques based on aggregated residual architecture. We proposed a novel method named MixNet. Firstly, design learning-based framework that consists encoding backbone adaptive header. Secondly, architecture...
The effective and reliable routing for Low Earth Orbit (LEO) satellite networks is intractable. existing approaches cannot well handle the time-varying topology, frequent link handover, imbalanced communication load. To tackle these issues, in this paper, we propose GRouting algorithm combining Graph Neural Networks (GNN) with Deep Reinforcement Learning (DRL) to dynamically find optimal paths LEO networks. First, GNN employed learn representation of non-Euclidean data structures. able...
Crowd counting and localisation are essential tasks in crowd analysis vital to ensure public safety. However, these via UAV bring new obstacles compared with video surveillance (e.g. viewpoint scale variations, background clutter, small scales). To overcome the difficulties, this research presents a novel network named PDNet. It employs multi-task learning approach combine point regression density map regression. PDNet includes backbone extract multi-scale features, Dilated Feature Fusion...
The delay-dependent variable structure control problem of a class uncertain neutral delay systems is considered. By selecting sliding surface depending on the current state and delayed state, paper gives sufficient condition in terms linear matrix inequalities (LMIs) such that closed-loop system guaranteed to be asymptotically stable. According reaching condition, method proposed. guarantees trajectory arrives at finite time interval kept here thereafter. Lastly simulation given illustrate...
In this paper, we design a highly accurate method for 3D mapping of multi-floor buildings. The basic idea is to combine the laser range sensor metric and barometric pressure detecting floor transition kinect collection environment information. Meanwhile, adopt Monte Carlo localization in 2D map improve accuracy localization. Finally, used merge By using information collected by real robot typical environment, tested compared with some other approaches, results show that efficient has better result
Cardiovascular disease is the number one killer that threatens human health. Analysis of coronary vascular geometry plays an important role in diagnosis cardiovascular disease. In order to make reliable judgments for clinical usage, accurate and robust segmentation methods are needed. We present improved approach dynamic angiography sequences, which can obtain blood vessel trees real time. start with a pre-trained convolutional neural network (CNN) on ImageNet as base initialization. then...
Geyser-2 is the second prototype MIPS CPU which provides a fine-grained run-time power gating (PG) controlled by instructions. Geyser-l, first only PG core. Although it demonstrated leakage reduction on real chip, operational frequency limited at 60MHz because of limitation I/O speed. with cache and TLB mechanism implemented to show (1) works least 200MHz commonly used clock for embedded systems, (2) also efficient environment application programs an operating system.
Stable operation of gas turbine engine compressor is constrained by rotating surge. In this paper, an approximate nonlinear surge margin model using equilibrium manifold presented. Firstly, the expansion form introduced. Secondly, property discussed, and effect mapping design to systematically studied. Then, paper also gives identification procedure modeling deteriorated performance was discussed. Finally, simulations with high pressure show that real-time has same accuracy thermodynamic...