P-62: Human-Centric AI Training: Leveraging Feedback Through a Novel UI
| 発表番号 | P-62 |
| 発表タイトル | Human-Centric AI Training: Leveraging Feedback Through a Novel UI |
| 筆者氏名・所属 |
Yan Jingbo （SOKENDAI） Yamada Seiji （NII） |
| アブストラクト | Human-centered research has garnered significant insights and feedback from active human participants. However, effectively handling and utilizing this feedback to train AI agents remains a challenge. To ensure optimal performance in downstream tasks such as image classification or segmentation and to efficiently distill human input for large dataset tasks, We propose a novel UI designed for the rapid collection of feedback pertaining to clustering on a large scale. This feedback is subsequently incorporated into an end-to-end training model, facilitating the transfer of human metrics to the AI system. Unlike typical accuracy-oriented models, our approach emphasizes the interpretability of decision-making processes. It can provide insights into the prediction results obtained after the reference selective prototype, offering a unique perspective on the deep model controlling. |
| 論文 | PDFファイル |
