July 26, 2024
Recently, Professor Xia Tian’s team has publishedthreepapers in top journals and conferences:Nature Machine Intelligence, ECCV 2024, and ICPP 2024.Thethreepapers cover cutting-edge research in medical AI privacy protection, machine vision, and high-performance genomic data.
The paper titled Shielding Sensitive Medical Imaging Datawas published on July 11thin Nature Machine Intelligence, a top-ranked journal in AI. With the widespread application of AI in medical image analysis, it has played a revolutionary role in disease diagnosis and patient prognosis prediction. However,protectingpatient privacyhas become increasingly prominent,andthis issuereceived widespread attention. Differential privacy technology, as an effective means of protecting data privacy, has traditionally been thought to sacrifice the performance of AI models in exchange for data privacy security.Invited bythe editor-in-chief ofNature Machine Intelligence, Prof.Xia Tian's team published an articletodiscuss the fundamental concepts of differential privacy technology andreviewlatest breakthroughs in applying differential privacy to large datasets: effectively protecting user data privacy without significantly impacting AI model performance.Breakingthe traditional notion that privacy protection and model performance cannot coexist,the proposedtechniqueprovidesnew perspectives and possibilities for the application of differential privacy in the field of AI.
View article:
https://www.nature.com/articles/s42256-024-00865-z
The paper titledTowards Dual Transparent Liquid Level Estimation in Biomedical Lab: Dataset, Methods and Practiceswas accepted by the top-levelconferencein computer vision,ECCV 2024.Accurately estimatesthe levels of such a liquid from arbitrary viewpoints is fundamental and crucial, especially in AI-guided autonomous biomedical laboratories for tasks like liquid dispensing, aspiration, and mixing. Existing methods focus only on single instances from fixed viewpoints, which significantly diverges from real-world applications.Hence, Prof.Xia Tian's team proposed a new benchmark for dual transparent liquid level height estimation, including a dataset, methods, andcode implementation. The proposed Dual Transparent Liquid Level Estimation Dataset (DTLD) contains 27,458 images covering four types of transparent biomedical laboratory vessels from multiple perspectives. Based on DTLD, the team proposed an end-to-end learning method to detect the liquid level contact line and estimate the liquid level.Aliquidlevelcorrection module was also introducedto enhance detection robustness. Experiments demonstrated that this method reduced the mean absolute percentage error of liquid level estimation by 43.4%.
The paper titledHardware Acceleration of Minimap2 Genomic Sequence Alignment Algorithmwas accepted by ICPP 2024. The paper introducesthe optimization of Minimap2, a widely used alignment algorithm for mapping variable-length reads to extensive reference sequences. Utilizing FPGA technology to expedite the time-consuming extension step,theteam employstheCycle Variable Logic Length(CVLL) method to optimize the pulse array, simplifying backtracking direction recording with two-bit variables to reduce memory usage. Theyalsocombinethepipelineand multi-channel technology to enhance performance. Experimental results on the FX410QL FPGA platform demon strate the notable speedups of our design, reaching a peak improve ment of 2.84× over CPU implementation, and achieving improve ments compared to GPU implementation at various input lengths. Beyond algorithm acceleration, our design providesnewinsights into enhancing genome data processing overall.
Prof.Xia Tian currently serves as the director of the Research Centerof Medical and Artificial IntelligenceatSSE,andthe lab focuses on AI applications in medicine, pharmaceuticals, and robotics, with multiple researchachievementspublished in top-rankedinternational journals such asScience,Nature Machine Intelligence,PNAS, andNature Communications.
