Abstract for Defect wafer map detection: Wafer testing is an important step in judging the quality of chips on wafers and plays a crucial role in yield assessment. After wafer testing, the judge information for each chip on one wafer will form a wafer map where 0 denotes out of wafer, 1 denotes PASS and 2 denotes FAIL. The defect patterns on wafer maps are often induced by some issues in the manufacturing process and some patterns keep recurring during the process. Defect wafer map detection aims to detect the wafers with special defect patterns for root cause analysis, which is useful for yield optimization. This work used multiple machine learning strategies. We firstly used a pre-trained model called DINOv2 to convert raw wafer images into image embeddings. Then, using those embeddings, we trained models such as Neural Networks, XGBoost, and LightGBM among which LightGBM reached the best performance. After checking the incorrect predictions, we performed label enhancement on some wrongly labeled wafer maps, leading to a 2% increase in model accuracy. Moreover, we added some symbolic features like yield rate, coordinates of failed mass center and the normalized distance between failed mass center and wafer center onto embeddings for training, which resulted in another 2% increase in model accuracy. Additionally, after looking into the Renesas data, we are working towards redefining some wafer defect patterns for specific products based on their own special defect root cause. Our defect wafer map detector may be used to help process engineers make decisions during the wafer testing process. Keywords – wafer defect map, AI for test, classification, Symbolic AI