Jipeng Li
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Jipeng Li has a total of 7 co-authors. Here are the top co-authors by number of shared publications:
- Chao Li
University of Electronic Science and Technology of China
1 shared publication - Ercan Engin Kuruoglu
Tsinghua University
1 shared publication - Fangyan Shi
School of Information and Communication Engineering, University of Electronic Science and Technology of China, Chengdu, China
1 shared publication - Guangqiang Yin
School of Information and Software Engineering, University of Electronic Science and Technology of China, Chengdu 611731, China
1 shared publication Shaoqi Hou
School of Information and Communication Engineering, University of Electronic Science and Technology of China, Chengdu, China
1 shared publication- Xueqiong Yuan
Tsinghua Berkeley Shenzhen Institute, Tsinghua Shenzhen International Graduate School, Shenzhen, Guangdong, China
1 shared publication
Latest Publications
- 5 June 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Artificial Intelligence
- Vol. 5 (11) , 5519-5529
Exploring Weight Distributions and Dependence in Neural Networks With $\alpha$-Stable Distributions
The fundamental use of neural networks is in providing a nonlinear mapping between input and output data with possibly a high number of parameters that can be learned from data directly. Consequently, studying the model’s parameters, particularly the weights, is of paramount importance. The distribution and interdependencies of these weights have a direct impact on the model’s generalizability, compressibility, initialization, and convergence speed. By fitting the weights of pretrained neural networks using the α-stable distributions and conducting statistical tests, we discover widespread heavy-tailed phenomena in neural network weights, with a few layers exhibiting asymmetry. Additionally, we employ a multivariate α-stable distribution to model the weights and explore the relationship between weights within and across layers by calculating the signed symmetric covariation coefficient. The results reveal a strong dependence among certain weights. Our findings indicate that the Gaussian assumption, symmetry assumption, and independence assumption commonly used in neural network research might be inconsistent with reality. In conclusion, our research shows three properties observed in neural network weights: heavy-tailed phenomena, asymmetry and dependence on certain weights.
- 5 June 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Artificial Intelligence
- Vol. 5 (11) , 5519-5529
Exploring Weight Distributions and Dependence in Neural Networks With $\alpha$-Stable Distributions
The fundamental use of neural networks is in providing a nonlinear mapping between input and output data with possibly a high number of parameters that can be learned from data directly. Consequently, studying the model’s parameters, particularly the weights, is of paramount importance. The distribution and interdependencies of these weights have a direct impact on the model’s generalizability, compressibility, initialization, and convergence speed. By fitting the weights of pretrained neural networks using the α-stable distributions and conducting statistical tests, we discover widespread heavy-tailed phenomena in neural network weights, with a few layers exhibiting asymmetry. Additionally, we employ a multivariate α-stable distribution to model the weights and explore the relationship between weights within and across layers by calculating the signed symmetric covariation coefficient. The results reveal a strong dependence among certain weights. Our findings indicate that the Gaussian assumption, symmetry assumption, and independence assumption commonly used in neural network research might be inconsistent with reality. In conclusion, our research shows three properties observed in neural network weights: heavy-tailed phenomena, asymmetry and dependence on certain weights.
- Open Access
- 20 July 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 8, 164570-164579
Feature Pyramid Attention Model and Multi-Label Focal Loss for Pedestrian Attribute Recognition
At present, there are many challenges in the field of pedestrian attribute recognition, such as small targets of some attributes, imbalanced samples, and low recognition accuracy of complex samples. In view of the above problems, we improved the model in two perspectives: 1) We proposed Feature Pyramid Attention Model (FPAM). In order to solve the problem that attributes are distributed in different locations in the pedestrian image, FPAM adopted the attention mechanism on the basis of ResNet-50, by which the model’s attention could be focused on key areas of the image. As for the difficulty in small targets attributes, we adopted feature pyramid integration strategy; 2) We proposed Multi Label Focal Loss (MLFL). Referring to Binary Cross Entropy Loss Function (CE) and Weight Binary Cross Entropy Loss Function (WCE), we added the weight parameters of samples which are difficult to classify to improve the recognition accuracy, and the rate of convergence was increased. Results show that our proposed method achieves 84.83% mA, 79.37% Accuracy, 87.47% Precision, 86.09% Recall, and 86.77% F1 on PETA dataset.
- 20 July 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 8, 164570-164579
Open AccessFeature Pyramid Attention Model and Multi-Label Focal Loss for Pedestrian Attribute Recognition
At present, there are many challenges in the field of pedestrian attribute recognition, such as small targets of some attributes, imbalanced samples, and low recognition accuracy of complex samples. In view of the above problems, we improved the model in two perspectives: 1) We proposed Feature Pyramid Attention Model (FPAM). In order to solve the problem that attributes are distributed in different locations in the pedestrian image, FPAM adopted the attention mechanism on the basis of ResNet-50, by which the model’s attention could be focused on key areas of the image. As for the difficulty in small targets attributes, we adopted feature pyramid integration strategy; 2) We proposed Multi Label Focal Loss (MLFL). Referring to Binary Cross Entropy Loss Function (CE) and Weight Binary Cross Entropy Loss Function (WCE), we added the weight parameters of samples which are difficult to classify to improve the recognition accuracy, and the rate of convergence was increased. Results show that our proposed method achieves 84.83% mA, 79.37% Accuracy, 87.47% Precision, 86.09% Recall, and 86.77% F1 on PETA dataset.