机构名称：
¥ 1.0
• Developed novel shortcut learning detection framework analyzing 750K+ samples across 13 datasets, predicting out-of-distribution performance degradation with 96% accuracy (Published - Nature npj Digital Medicine) • Created an attention-based mechanism for localizing and correcting for multiple shortcuts (spatial and spectral), surpasses SOTA by 7.5% in data with multiple spurious correlations occurring simultaneously (Under Review - CVPR 2025）•设计生成的内涂层系统，用于使用扩散模型减少医学成像任务的混淆 - 确定并纠正了隐藏分层的诊断性能，提高了20％的诊断性能•构建和部署了多个用于放射学和超声的临床AI工具，用于实施验证协议，并与医院互具
balagopal（balu）unnikrishnan
主要关键词
相关文件推荐
