AutoML 2024 awards Best Paper Award to our paper FLIQS
Authored by researchers from Cornell University and Google, the paper introduces a novel one-shot mixed-precision quantization method that simultaneously optimizes both integer and floating-point quantization without requiring retraining. Their approach, FLIQS, achieves state-of-the-art results across multiple deep learning models, offering significant improvements in accuracy and efficiency.
Notable achievements include boosting ResNet-18 ImageNet accuracy by 1.31%, and improving MobileNetV2 accuracy by 2.69% — all while maintaining equivalent model cost. The innovative method marks a major step forward in efficient AI model deployment.
Congratulations to the entire team on this outstanding recognition!
Yixiao Du
awards publication