Anomaly detection in high-dimensional data is particularly challenging. What specific problem arises?