#China #Shanghai #Fintech #Qifu Technology #IJCAI

Milestone Achievement for Qifu Technology in Fintech

On May 7, 2025, a significant milestone was reached by Qifu Technology, a leading fintech enterprise in China, through their collaboration with Beijing Jiaotong University. Their research paper, titled "Leveraging MLLM Embeddings and Attribute Smoothing for Compositional Zero-Shot Learning," has been officially accepted for presentation at the International Joint Conference on Artificial Intelligence (IJCAI) 2025. This event is widely recognized in the AI community as one of the most prestigious, boasting an acceptance rate of just 19.3%, making the achievement even more remarkable given that it was one of only 1,042 submissions selected among over 5,400.

The paper delves into a cutting-edge area of research known as Compositional Zero-Shot Learning, which is aimed at the identification of novel combinations of attributes and objects by leveraging existing knowledge bases. A crucial aspect of this research is its ability to address several challenges that have traditionally plagued the field, such as background noise interference, limitations in capturing semantic meanings through word embeddings, and issues related to overconfidence in known entity combinations.

Introduction of TRIDENT Framework

To tackle these challenges, the team introduced an innovative framework named TRIDENT, which stands at the forefront of their findings. The TRIDENT framework employs Multimodal Large Language Model (MLLM) embeddings and implements a technique known as attribute smoothing. This framework contains several key modules, including Feature Adaptive Aggregation (FAA), which effectively minimizes background disturbances, learns conditional masks for a refined feature extraction process, and utilizes hidden states from the MLLM to significantly enhance semantic representation.

The robustness of this approach has been demonstrated through its remarkable performance across multiple datasets, and it opens up new possibilities for various applications, particularly in image recognition and content understanding, two vital areas in today's digital landscape.

Applications in Fintech

In the specific realm of fintech, the TRIDENT framework shows immense potential. For instance, in intelligent risk management, it analyzes diverse multimodal data, including transaction behaviors and user profiles. This allows for the rapid detection of emerging fraud patterns, proving to be a faster and more accurate method compared to traditional models, ultimately leading to reduced losses for financial institutions.

Furthermore, TRIDENT also enhances customer service strategies. It does this by offering a more nuanced understanding of complex user inquiries, thereby equipping organizations with the ability to provide personalized and effective support to their clients.

Commitment to Innovation

This impressive achievement underscores Qifu Technology's ongoing commitment to fostering AI innovations. By escalating its research and development investments, along with bolstering partnerships with reputable academic institutions, Qifu Technology is not only contributing to advancements in AI applications but also driving growth and progress within the entire industry. This proactive approach is envisioned as a means of enhancing technological capabilities that will benefit society at large.

In conclusion, the acceptance of this paper at IJCAI 2025 is a testament to the pioneering work being done at Qifu Technology and serves as an encouraging sign of the transformational potential that AI holds for the fintech sector and beyond.