Anomaly detection platforms in real-world environments require continuous interaction between automated systems and domain experts, as anomalies evolve dynamically and their definitions vary across contexts. Therefore, an effective platform must collaborate with experts and incorporate their feedback to update the system. This paper introduces AEGIS, an anomaly detection platform that aims to support interaction between domain experts and data-driven agents through three core capabilities: (1) data-driven insights through real-time monitoring, explanations, and distribution shift detection, which invoke customized tools to generate appropriate responses, (2) an expert feedback interface for labeling and direct updates via chat-based interaction, and (3) autonomous model construction that leverages expert-labeled data with LLM-driven hyperparameter optimization. Through this design, AEGIS fosters continuous interaction in which the platform provides insights while experts guide model improvement, ensuring user intent is reflected and robustness is maintained under evolving data distributions.

Vision Transformers (ViTs) have achieved remarkable success in various computer vision tasks. However, ViTs have a huge computational cost due to their inherent reliance on multi-head self-attention (MHSA), prompting efforts to accelerate ViTs for practical applications. To this end, recent works aim to reduce the number of tokens, mainly focusing on how to effectively prune or merge them. Nevertheless, since ViT tokens are generated from non-overlapping grid patches, they usually do not convey sufficient semantics, making it incompatible with efficient ViTs. To address this, we propose ImagePiece, a novel re-tokenization strategy for Vision Transformers. Following the MaxMatch strategy of NLP tokenization, ImagePiece groups semantically insufficient yet locally coherent tokens until they convey meaning. This simple retokenization is highly compatible with previous token reduction methods, being able to drastically narrow down relevant tokens, enhancing the inference speed of DeiT-S by 54% (nearly 1.5x faster) while achieving a 0.39% improvement in ImageNet classification accuracy. For hyper-speed inference scenarios (with 251% acceleration), our approach surpasses other baselines by an accuracy over 8%.