BEGIN:VCALENDAR
VERSION:2.0
PRODID:Data::ICal 0.24
BEGIN:VEVENT
DESCRIPTION:   'Title: LSTM Autoencoder Ensemble for NMEA 2000 Intrusion De
 tection on\n   Vessel Networks\n   Tags: Maritime Hacking Village | Creato
 r Talk/Panel\n   When: Saturday\, Aug 8\, 16:30 - 16:59 PDT\n   Where: LVC
 CW Level 1 Hall 3 801 (Creator Stage 2) - [1]Map\n\n   Description:\n\n   
 Maritime vessels increasingly depend on NMEA 2000 (CAN bus) networks\n   t
 o interconnect navigation\, propulsion\, and safety-critical systems\,\n  
  yet the underlying protocol provides no authentication\, any device can\n
    transmit any message ID. As maritime cyber incidents such as GPS\n   sp
 oofing\, engine manipulation\, and AIS attacks become more frequent\,\n   
 existing automotive CAN intrusion detection systems fall short because\n  
  they fail to model maritime-specific traffic behavior\, including\n   var
 iable RPM\, long duty cycles\, and sensor drift. We present a\n   multi-mo
 dal LSTM autoencoder ensemble in which four specialized models\n   indepen
 dently monitor message timing\, payload content\, frequency\, and\n   devi
 ce-level behavior. Each autoencoder learns normal operating\n   patterns a
 nd flags anomalies via reconstruction error\, with a\n   majority-voting a
 ggregator raising an alert when at least two of four\n   components agree.
  Detection thresholds are set automatically at the\n   99th percentile of 
 validation error\, requiring no labeled attack data.\n   The system was ev
 aluated on a 24-hour continuous capture from an\n   operating vessel compr
 ising 9.16 million messages across 24 PGNs and\n   three source devices\, 
 spanning a full operational cycle and six\n   simulated attack types. Usin
 g 5-fold temporal cross-validation\, the\n   ensemble achieved an F1 score
  of 0.964 (95% CI: 0.952–0.976) and\n   ROC-AUC of 0.991\, while reducin
 g variance 28× relative to a\n   single-model baseline. The unsupervised 
 threshold matched supervised\n   tuning (F1=0.965 vs. 0.971) without attac
 k labels\, and a novel network\n   drift detection module identified unaut
 horized hardware changes with\n   perfect accuracy (F1=1.000) across 120 t
 rials. With 38.76 ms inference\n   latency\, the approach is suitable for 
 real-time onboard monitoring.\n   Future work includes voltage fingerprint
 ing\, live deployment\, and\n   transfer learning.\n\n   Speakers:Anissa E
 lias\,James Campbell\n\n   SpeakerBio:  Anissa Elias\, University of Rhode
  Island\n\n   Anissa Elias is a Research Engineer and Ph.D. researcher spe
 cializing\n   in maritime cyber-physical system security\, autonomous moni
 toring\, and\n   on-edge AI. Her work focuses on securing underwater and m
 aritime\n   systems through real-time anomaly detection\, embedded intelli
 gence\,\n   and resilient network architectures. She has deep experience w
 ith\n   maritime network analysis\, digital twins\, and deployable AI for\
 n   resource constrained environments. Anissa brings a strong background\n
    in defense-focused research\, autonomous system security\, and AI trust
 \n   and compliance for naval and maritime operations.\n\n   SpeakerBio:  
 James Campbell\, Independent Researcher\n\n   James (Soups) Campbell is a 
 Red Team Operator for the US Coast Guard\,\n   specializing in malware res
 earch\, offensive tool development\, and OT\n   cybersecurity. In his free
  time\, he can be found building autonomous\n   vessels\, hacking on boats
 \, and playing with his very energetic puppy.\n\n   '\n\n   1. #LVCCW_Leve
 l1_Hall3\n\n\n
DTEND:20260808T235900Z
DTSTART:20260808T233000Z
LOCATION:Maritime Hacking Village - LVCCW Level 1 Hall 3 801 (Creator Stage
  2)
SUMMARY:LSTM Autoencoder Ensemble for NMEA 2000 Intrusion Detection on Vess
 el Networks
END:VEVENT
END:VCALENDAR