Recently, undergraduate student Wang Jiayun from the School of Microelectronics of XJTU found that using adaptive margin strategy is more appropriate for dynamically adjusting the learning objectives of the deep convolutional neural network model to match the dynamic characteristic space for study. The entire research was under the guidance of Professor Wang Jinjun from the Institute of Artificial Intelligence and Robotic of XJTU. The weaknesses of conventional contrastive loss function and triplet loss function were also successfully overcome by using adaptive margin loss function in his further studies.
The above findings were published on the journal Pattern Recognition (five-year impact factor 4.991) with the title of “Deep Ranking Model by Large Adaptive Margin Learning for Person Re-identification”. Wang Jiayun authored the article.
Link to the article: