A5056167336- Advanced Neural Network Applications
- Advanced Memory and Neural Computing
- Human Pose and Action Recognition
- Robotics and Sensor-Based Localization
- Advanced Image and Video Retrieval Techniques
- Context-Aware Activity Recognition Systems
- Genomics and Phylogenetic Studies
- Machine Learning and ELM
- Neural Networks and Applications
- Video Surveillance and Tracking Methods
- Image Enhancement Techniques
- Gene expression and cancer classification
- Radiation Effects in Electronics
- Non-Invasive Vital Sign Monitoring
- Ferroelectric and Negative Capacitance Devices
- Advanced Vision and Imaging
Southern University of Science and Technology
2021-2025
Georgia Institute of Technology
2022-2023
Multi-bit-width convolutional neural network (CNN) maintains the balance between accuracy and hardware efficiency, thus enlightening a promising method for accurate yet energy-efficient edge computing. In this work, we develop state-of-the-art multi-bit-width accelerator NAS Optimized deep learning networks. To efficiently process inferencing, multi-level optimizations have been proposed. Firstly, differential Neural Architecture Search (NAS) is adopted high generation. Secondly, hybrid...
Deep neural networks (DNNs) in safety-critical applications demand high reliability even when running on edge-computing devices. Recent works System-on-Chip (SoC) design with state-of-the-art (SOTA) hardware artificial intelligence (AI) accelerators and corresponding multi-bit-width (MBW) convolutional network (CNN) generation strategies show that MBW CNNs can effectively explore the trade-off between accuracy efficiency. However, has not been considered such analysis, though highly...
Computing-In-memory (CIM) accelerators have the characteristics of storage and computing integration, which has potential to break through limit Moore's law bottleneck Von-Neumann architecture for convolutional neural networks (CNN) implementation improvement. However, performance CIM is still limited by conventional CNN architectures inefficient readouts. To increase energy-efficient performance, an optimized model required a low-power column parallel readout necessary edge-computing...
Various industrial and domestic applications call for optimized lightweight video LSTM network models on edge. The recent tensor-train method can transform space-time features into tensors, which be further decomposed low-rank analysis rank selection of tensor is however manually performed with no optimization. This paper formulates a search algorithm to automatically decide ranks consideration the trade-off between accuracy complexity. A fast method, called RankSearch, developed find...
Tripping or falling is among the top threats in elderly healthcare, and development of automatic fall detection systems are considerable importance. With fast Internet Things (IoT), camera vision-based solutions have drawn much attention recent years. The traditional video analysis on cloud has significant communication overhead. This work introduces a lightweight network based spatio-temporal joint-point model to overcome these hurdles. Instead detecting motion by Convolutional Neural...
Visual simultaneous localization and mapping (VSLAM) is one of the core technologies in autonomous driving, intelligent robots, metaverse other fields. Besides, loop closure detection (LCD) an essential component VSLAM which can correct drift accumulated errors caused by visual odometry (VO) front-end, assist robot to build a globally consistent map. Over years, several deep-learning methods have been proposed address task. However, prior neural network-based LCD models are heavy model size,...
Falling is ranked highly among the threats in elderly healthcare, which promotes development of automatic fall detection systems with extensive concern. With fast Internet Things (IoT) and Artificial Intelligence (AI), camera vision-based solutions have drawn much attention for single-frame prediction video understanding on by using Convolutional Neural Network (CNN) 3D-CNN, respectively. However, these methods hardly supervise intermediate features good accurate efficient performance edge...
This paper proposes a new framework called ATFVO which can be deployed on the edge device to resolve monocular visual odometry problem. The vast majority of algorithms using deep learning are equivalent or beyond traditional in performance, however they do not consider computing capability equipment. In this paper, convolution neural network (CNN) and attentive tensor-compressed compression LSTM (A-T-LSTM) used, with optical flow feature as input 6-DoF absolute-scale pose output. is fused...
Neural architecture search (NAS) optimized multi-bit-width convolutional neural network (CNN) maintains the balance between performance and efficiency, thus enlightening a promising method for accurate yet energy-efficient edge computing. In this work, we propose high throughput three-dimensional (3D) systolic accelerator NAS CNNs, in which input feature matrix, weight matrix output are delivering vertically, horizontally perpendicularly through array respectively. With 3D data flow,...
DNA sequencing is a popular tool to demystify the code of living organisms and reforming medical, pharmaceutical biotech industries. The Next-Generation Sequencing (NGS) plays vital role in high-throughput with massively parallel data generation. Nevertheless, massive amount imposes great challenges for analysis. It arduous reach low error rate handling noisy and/or biased signals owing imperfect biochemical reactions imaging systems. Furthermore, homogeneous computing system lacks power...