森研究室の掛林諒平君と森による論文「Text-to-Imageモデルに対するデータ汚染攻撃の多言語環境での脆弱性要因と防御手法の提案」が、2025年度 情報通信システムセキュリティ研究賞（ICSS研究賞）を受賞しました。おめでとうございます！
掛林 諒平, 森 達哉, 「Text-to-Imageモデルに対するデータ汚染攻撃の多言語環境での脆弱性要因と防御手法の提案」, 電子情報通信学会 ICSS研究会（2026年3月研究会, 沖縄県立博物館・美術館）, 2026年3月.
We’re excited to welcome one new PhD student, five new master’s students, and seven new undergraduate students to our lab!
We are thrilled to announce that our conference paper entitled “Cross-Lingual Vulnerabilities of Text-to-Image Models: Evaluating Data Poisoning Attacks Across Ten Languages” has been accepted for publication in the 28th International Conference on Pattern Recognition (ICPR 2026). Congraturations Kakebayashi-kun!
Ryohei Kakebayashi and Tatsuya Mori, "Cross-Lingual Vulnerabilities of Text-to-Image Models: Evaluating Data Poisoning Attacks Across Ten Languages." In Proc. of the 28th International Conference on Pattern Recognition (ICPR 2026) Lyon, France, Sep 2026. Published in Springer LNCS.
We are thrilled to announce that our conference paper entitled “TEV-IDS: CAN Intrusion Detection System via Spatial Temporal-Entropy-Variation Fingerprinting” has been accepted for publication in IEEE 103rd Vehicular Technology Conference (VTC2026-Spring). Congratulations to Peng-kun and kudos to the entire team!
In this work, we propose TEV-IDS, an ID-agnostic intrusion detection framework for in-vehicle CAN networks. Existing systems are largely limited to binary benign/attack decisions and depend on vehicle-specific CAN ID semantics, which hinders both fine-grained attack diagnosis and scalable deployment across heterogeneous platforms. TEV-IDS overcomes these challenges by constructing a Temporal–Entropy–Variation (TEV) feature space from ID-independent cues (inter-arrival intervals, payload entropy, and bit-level variation), and projecting them via a Tri-Perspective View (TPV) into compact 2D feature maps for a lightweight classifier. A stratified few-shot adaptation protocol further enables calibration to unseen vehicles with minimal target data. On the CAN-MIRGU dataset, TEV-IDS achieves a macro-F1 above 0.99 for 17-class attack classification, while cross-domain evaluation on the ROAD dataset yields a macro-AUC above 0.924 with limited calibration data, all under real-time CPU inference latency.
Shuo Peng, Zhihe Zhang, Go Tsuruoka, and Tatsuya Mori. "TEV-IDS: CAN Intrusion Detection System via Spatial Temporal-Entropy-Variation Fingerprinting." In Proc. of IEEE 103rd Vehicular Technology Conference (VTC2026-Spring), Nice, France, Jun 2026.
森が第22回情報セキュリティ文化賞記念講演会（日経メッセ2026）において「新興技術時代のオフェンシブセキュリティ研究と人材育成の課題」の招待講演をしました。