Watch and track your favorite playlist.
Curated by: 김성범[ 교수 / 산업경영공학부 ] (327 videos)
최근 인공지능 분야에서 safe AI는 중요한 이슈로 떠오르고 있다. 이는 단순히 모델의 우수한 예측 성능뿐만 아니라, 안전한 인공지능 활용에 대한 필요성이 커지고 있기 때문이다. 이러한 흐름 속에서 주목 받는 분야 중 하나가 바로 out-of-distribution (OOD) detection이다. 이미지 데이터에 대한 OOD detection은 학습되지 않은 클래스를 탐지하는 기술로, 기존에는 이미지 기반 Vision 모델에 의존한 연구가 주를 이루었다. 하지만 최근 vision-language model (VLM)이 등장하면서, 이미지뿐만 아니라 텍스트 정보를 함께 활용하려는 시도가 활발해지고 있다. 이는 최신 VLM이 수억 개의 대규모 데이터로 사전 학습되어, 별도의 추가 학습 없이도 높은 성능을 발휘할 수 있기 때문이다. 이때, 이러한 강력한 사전학습 모델의 지식을 학습 없이 효과적으로 활용하기 위한 방법으로, prompt engineering이 주목 받고 있다. 본 세미나에서는 VLM을 기반으로, 추가 학습 없이 prompt engineering만을 활용하여 zero-shot으로 OOD detection을 수행하는 연구들을 소개하고자 한다. 참고자료: [1] Radford, A., Kim, J. W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., ... & Sutskever, I. (2021, July). Learning transferable visual models from natural language supervision. In International conference on machine learning (pp. 8748-8763). PmLR. [2] Hendrycks, D., & Gimpel, K. (2017, February). A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networks. In International Conference on Learning Representations. [3] Miyai, A., Yu, Q., Irie, G., & Aizawa, K. (2023). Locoop: Few-shot out-of-distribution detection via prompt learning. Advances in Neural Information Processing Systems, 36, 76298-76310. [4] Ming, Y., Cai, Z., Gu, J., Sun, Y., Li, W., & Li, Y. (2022). Delving into out-of-distribution detection with vision-language representations. Advances in neural information processing systems, 35, 35087-35102. [5] Esmaeilpour, S., Liu, B., Robertson, E., & Shu, L. (2022, June). Zero-shot out-of-distribution detection based on the pre-trained model clip. In Proceedings of the AAAI conference on artificial intelligence (Vol. 36, No. 6, pp. 6568-6576). [6] Jiang, X., Liu, F., Fang, Z., Chen, H., Liu, T., Zheng, F., & Han, B. (2024) Negative Label Guided OOD Detection with Pretrained Vision-Language Models. In The Twelfth International Conference on Learning Representations. [7] Wang, H., Li, Y., Yao, H., & Li, X. (2023). Clipn for zero-shot ood detection: Teaching clip to say no. In Proceedings of the IEEE/CVF International Conference on Computer Vision (pp. 1802-1812).