From November 11 to 14, 2023, in Istanbul, Turkey, the research team from QAI – FPT Software presented their scientific paper, "Towards Better Explanations for Object Detection," at the 15th Asia Conference on Machine Learning (ACML 2023). This marks the second time in 2023 that QAI has demonstrated its expertise at prestigious international AI conferences, further solidifying its position in the global AI research community.

Following its success at AAAI-23 earlier this year, the QAI R&D team has continued its research efforts, introducing a novel eXplainable AI (XAI) method called D-CLOSE. This approach enhances the interpretability of AI models, particularly in Object Detection tasks within QaiDora Vision, an AI-powered video analytics solution. By improving the quality of AI-generated explanations, D-CLOSE enables AI engineers and researchers to better understand the rationale behind object detection decisions, thereby enhancing model transparency and reliability.

The D-CLOSE method significantly optimizes the speed and accuracy of explanations, especially in scenarios where data quality is negatively impacted by external factors. This innovation not only reduces processing time but also identifies critical data features, allowing models to achieve high accuracy with fewer data inputs. Compared to traditional methods, D-CLOSE accelerates QaiDora Vision’s ability to detect, identify, and analyze object behavior, enhancing both performance and interpretability.

The D-CLOSE approach also plays a crucial role in refining near real-time AI models used in akaCam Box—a proprietary edge computing device developed and assembled by the QaiDora Vision team. By integrating this new XAI method, akaCam Box achieves higher efficiency, reduced data processing costs, and faster deployment for customers. These advancements are particularly beneficial in complex AI-driven environments such as factories, warehouses, and retail chains, where real-time video analytics is essential.

The D-CLOSE method is the focal point of the research paper "Towards Better Explanations for Object Detection," authored by the QAI research team: Trương Văn Bình, Nguyễn Trương Thành Hưng, Nguyễn Võ Thành Khang, Nguyễn Quốc Khánh, and Cao Quốc Hưng. This study was accepted by the 15th Asia Conference on Machine Learning (ACML 2023) and selected for a long presentation session, placing it alongside distinguished institutions such as NEC, Imperial College London, University of Exeter, and JP Morgan. The presentation took place on November 14, 2023, at Acibadem University, Istanbul, Turkey.

This year’s ACML conference gathered leading AI experts and research institutions worldwide, including Google DeepMind, NTT, Kyoto University, Tokyo Institute of Technology, Nanjing University, and Shanghai Jiao Tong University. A poster featuring QAI – FPT Software’s research was displayed at the conference venue, allowing direct discussions between the authors and global AI experts over the four-day event.

The research team also achieved another milestone with their paper, "Enhancing the Fairness and Performance of Edge Cameras with Explainable AI," which has been accepted for presentation at the 42nd IEEE International Conference on Consumer Electronics (ICCE 2024). This prestigious conference will take place in Las Vegas, US, in January 2024.

Beyond their research contributions, the authors of these academic papers are also the official translators of “AI 2041: Ten Visions for Our Future” by Kai-Fu Lee, recently published in Vietnamese. Their work in eXplainable AI (XAI) includes developing cutting-edge methods such as Segmentation-Class Activation Mapping (SeCAM) for image classification explainability. This approach addresses limitations of conventional XAI techniques, optimizing input data quality and reducing model analysis time. Currently, the team is finalizing another research paper focused on optimizing akaCam Box, a key differentiator of akaCam in the AI-powered video analytics market. Through D-CLOSE and ongoing XAI innovations, the QAI R&D team continues to push the boundaries of explainable AI, contributing to the future of trustworthy and efficient AI-driven solutions.