Research | C. Huang-YO

Back

Memristive Neural Computing and Medical Pathology Image Segmentation

Grant No.2019YFG0190

Memristor is memory resistor. It was proposed by Prof. Cai, teaching at the University of California, Berkeley, in the US. When studying the relationship between charge, current, voltage and magnetic flux, he deduced that in addition to resistors, capacitors and inductors, there should be another component that represents the relationship between charge and magnetic flux. The effect of this component is that its resistance changes with the amount of current passing through it, and even if the current stops, its resistance will still stay at the previous value until it receives a reverse current, and it will be pushed back. It adjusts resistance as current changes and can be used to store information. Because of their small size and low energy consumption, memristors are good at storing and processing information. The workload of a memristor is equivalent to the effectiveness produced by a dozen transistors in a CPU chip. Memristor-based random access memory has superior integration, power consumption, and read and write speeds than traditional random access memory. In addition, memristor is the best way to implement artificial neural network synapses in hardware [1].

In this research program, our team first implemented the memristor on Matlab, and then we made it into a network, which is a memristive neural network. This network can be used for machine learning, deep learning accelerators, chip design, acceleration, etc. After implementing this network, we encapsulate it into a package, and then we design or apply existing deep learning networks based on it to detect and segment pigmented skin diseases. By calling this package, fast calculations are achieved and the training and testing of neural networks are accelerated.

For corresponding research, you can refer to papers or patents:

N. Yang, Y. Yu, S. Zhong, X. Wang, K. Shi and J. Cai, "Exponential synchronization of stochastic delayed memristive neural networks via a novel hybrid control," Neural Networks, Volume 131, 2020, Pages 242-250. Link

X. Wang et al., "Dynamic Pinning Synchronization of Fuzzy-Dependent-Switched Coupled Memristive Neural Networks With Mismatched Dimensions on Time Scales," in IEEE Transactions on Fuzzy Systems, vol. 30, no. 3, pp. 779-793, March 2022. Link

C. Huang, Y. Yu and M. Qi, "Skin Lesion Segmentation Based on Deep Learning," 2020 IEEE 20th International Conference on Communication Technology (ICCT), Nanning, China, 2020, pp. 1360-1364. Link

Y. Yu, E. Favour and P. Mazumder, "Convolutional Neural Network Design for Breast Cancer Medical Image Classification," 2020 IEEE 20th International Conference on Communication Technology (ICCT), Nanning, China, 2020, pp. 1325-1332. Link

M. Ayidzoe, Y. Yu, P. Mensah, J. Cai, A. Kwabena and N. Tashi, "PressFeature amplification capsule network for complex images," Journal of Intelligent and fuzzy systems, 2021. Link

X. Wang et al., "Relaxed Exponential Stabilization for Coupled Memristive Neural Networks With Connection Fault and Multiple Delays via Optimized Elastic Event-Triggered Mechanism," in IEEE Transactions on Neural Networks and Learning Systems, vol. 34, no. 7, pp. 3501-3515, July 2023. Link

Y. Yu, X. Hu, Y. Fu, C. Huang, etc."Nanomemristors and Neuromorphic Computing (Neuromorphic Circuits for Nanoscale Devices)," 2022. (BooK)

......

memristor simulation and memristor network construction

neural network simulation and implementation

Industrial Computing and Portable AI Inspection

Grant No.2022ZHCG0033

Many AI models now have good laboratory results. As AI moves out of the laboratory and into the industrial market, more and more companies are demanding more than just pure accuracy for AI products. An industrial-grade AI that can be applied to the market should consider factors such as computing time, algorithm complexity, hardware consumption, development cost, and stability. At the same time, developers will also face some complex scenarios, such as extreme environments, low computing efficiency equipment (no high-performance computing power such as GPU), no continuous supply of energy, etc. Only by considering these factors can we design high-performance algorithms to achieve better results. Adapt to needs.

For this reason, in this research program, we used the city as the background and based on ensuring information security and communication security, we designed a system that can ensure the operation of urban power and use AI-assisted detection. At the same time, on this basis, we will also try our best to complete other functions, such as tracking, anomaly detection, system centralization, remote sensing,etc., which can make our city more intelligent.

For corresponding research, you can refer to papers or patents:

Q. Lv, Y. Rao, S. Zeng, C. Huang and Z. Cheng, "Small-Scale Robust Digital Recognition of Meters Under Unstable and Complex Conditions," in IEEE Transactions on Instrumentation and Measurement, vol. 71, pp. 1-13, 2022. Link

Y. Yi, Y. Rao, C. Huang, S. Zeng, Y. Yang, Q. He, X. Chen, "Optimization of Quantum Key Distribution Parameters Based on Random Forest," 2021 4th International Conference on Pattern Recognition and Artificial Intelligence (PRAI), Yibin, China, 2021, pp. 164-168. Link

Y. Wang, Y. Rao, C. Huang, Y. Yang, Y. Huang and Q. He, "Using the Improved Mask R-cnn and Softer-nms for Target Segmentation of Remote Sensing Image," 2021 4th International Conference on Pattern Recognition and Artificial Intelligence (PRAI), Yibin, China, 2021, pp. 91-95. Link

J. Xue, Y. Rao, J. Wu, H. Gou, Y. Liu and Y. Yang, "A Lightweight Object Detection Method for Bank Operation and Maintenance Scenarios," 2022 5th International Conference on Pattern Recognition and Artificial Intelligence (PRAI), Chengdu, China, 2022, pp. 582-588. Link

J. Pu et al., "matExplorer: Visual Exploration on Predicting Ionic Conductivity for Solid-state Electrolytes," in IEEE Transactions on Visualization and Computer Graphics, vol. 28, no. 1, pp. 65-75, Jan. 2022. Link

......

Random Forest and Quantum Communication Encryption

resource-efficient AI for digits recognition

remote sense imaging

Smart City Construction and Medical Assistance based on AI

Grant No.23ZDYF0755

The unified construction of smart cities involves many fields: among them, the importance of public health and crowd statistics in large-scale health events is self-evident. How to combine 5G and corresponding mobile devices to develop a health monitoring system in cities is the focus of this research. Different from studying traditional AI, this time we will combine AI with some emerging fields, such as edge computing, Internet of Things, cloud services, etc. This allows people's health data to be uploaded and analyzed in real time.

For this reason, in this research program, we still used the city as the background and based on Epidemic spread situations, such as Covid-19, we first start with the general population to detect pocket wearing conditions, and then use individual health data to analyze whether there is a high probability of contracting the disease. Afterwards, pathological segmentation is performed from lung CT images of diseased individuals. Such a series of precise collections from the whole to the individual are all collected by the IoT system and uploaded to the central server for analysis. Finally, based on the analysis results, government departments promptly formulate the latest policies to adjust epidemic control.

For corresponding research, you can refer to papers or patents:

Y. Rao, Q. Lv, S. Zeng, Y. Yi, C. Huang, Y. Gao, Z. Cheng and J. Sun, "COVID-19 CT ground-glass opacity segmentation based on attention mechanism threshold," in Biomedical Signal Processing and Control, vol. 81, 2023. Link

C. Huang, Y. Liu, J. Li, H. Tian and H. Chen, ”Application of YOLOv5 for Mask Detection on IoT,” 2023 5th International Conference on Computing and Data Science (CDS), Macau, China, 2023. (will come soon)

C. Huang, etc., "Intelligent integrated detection: AIoT based GB-YOLO," IEEE Internet of Things, 2023. (submitted)

Y. Rao, O. T. Nartey, S. Zeng, K. N. Acheampong, C. R. Haruna and J. Sun, "LeFUNet: UNet with Learnable Feature Connections for Teeth Identification and Segmentation in Dental Panoramic X-ray Images," 2022 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), Las Vegas, NV, USA, 2022, pp. 2110-2118. Link

Y. Shao, X. Zhang, H Chu, X. Zhang, D. Zhang, Y. Rao, "AIR-YOLOv3: Aerial Infrared Pedestrian Detection via an Improved YOLOv3 with Network Pruning," Applied Sciences, 2022. Link

......

Covid 19 CT images segmentation based on machine learning model

YOLOv5 and IoT systerm: real-time detection and tracking

Memristive Neural Network and Its Interaction with Crowd Intelligence

Grant No.62276055

Ant colony algorithm is a probabilistic algorithm used to find optimal paths. It is a heuristic global optimization algorithm in evolutionary algorithms. The basic idea of applying the ant colony algorithm to solve optimization problems is: using the walking path of ants to represent the feasible solution to the problem to be optimized, and all paths of the entire ant colony constitute the solution space of the problem to be optimized. Ants with shorter paths release more pheromones. As time goes by, the accumulated pheromone concentration on the shorter path gradually increases, and the number of ants choosing this path also increases. Eventually, the entire ants will converge on the best path under the action of positive feedback, which corresponds to the optimal solution to the problem to be optimized.

In the memristor neural network, we can use the ant colony algorithm as follows: use a greedy algorithm or genetic algorithm to record the changes in gate circuit resistance, calculate the relationship between energy consumption and efficiency, and use a loss function to describe the minimum error. The network selects the optimal path (optimized memristive neural network) with minimal error, thereby achieving relatively minimal power consumption and maximum efficiency. After that, we can develop a neural network model on this basis, so that AI can reach a state with minimal resource requirements but maximum efficiency.

Based on the above situation, for this research project, we will design gate circuits and memristive neural networks on hardware, and integrate them into a usable hardware system. On top of this hardware system, the basic settings are deployed for setting up its gate circuit and differential amplification circuit, and then the AI algorithm is developed so that the AI can perform high-performance computing.

For corresponding research, you can refer to papers or patents:

C. Huang, Y. Lin, T. Qian, H. Wang and Y. Yu, "Attention EnhancedNetwork with Semantic Inspector for Medical lmage Report Generation," 2023 IEEE 35th International Conference on Tools with Artificial Intelligence (ICTAI), Atlanta, USA, 2023. (coming soon)

X. Zhong, Y. Yu, C. Zhou, X. Wang, X. Feng, Z. Zhou, J. Shen, J. Wang, X. Han and C. Huang, ”A Matrix Coding Genetic Algorithm Based on Memristor for Image Edge Detection,” 2023 the 9th International Conference on Communication and Information Processing (ICCIP), Lingshui, China, 2023. (will come soon)

J. Yang, Q. Zhong, K. Shi, Y. Yu and S. Zhong, "Stability and Stabilization for T-S Fuzzy Load Frequency Control Power System with Energy Storage System," in IEEE Transactions on Fuzzy Systems. Link

X. Han, Y. Yu, X. Wang, X. Feng and S. Zhong, "Stabilization of Delayed Memristive Neural Networks Driven by Mixed Deception Attacks," 2023 International Conference on Control, Automation and Diagnosis (ICCAD), Rome, Italy, 2023, pp. 1-6. Link

M. Abdullah, Y. Yu, K. Adu, Y. Imrana, X. Wang, J. Cai, "HCL-Classifier: CNN and LSTM based hybrid malware classifier for Internet of Things (IoT)," Future Generation Computer Systems, vol. 142, pp. 41-58, 2023. Link

......