Yuv's Portfolio

Research Summary

4 min read

Research Work and Interests

Here you’ll find my research interests, projects, and ongoing work

Maritime Technology and AI

Maritime Anomaly Detection Using Bearing Analysis

  • Date: November 2023 to May 2024
  • Description: Research on focusing on bearing analysis for maritime anomaly detection, addressing AIS spoofing and fake trajectory risks.
  • Details: The maritime industry represents trillions of dollars, making it one of the world’s most important industries. However, current issues in the Automated Identification System (AIS) make it susceptible to spoofing. This study focused on analysis of tanker ships with a normal distribution from the Gulf of Mexico and an anomalous distribution from the Sea of Japan. The anomalous fleet was faking their location in an attempt to illegaly transport oil.
  • Key Findings:
    • Frequency plots showing extreme discrepancies between normal and anomalous bearing changes
    • Average bearing change of anomalous distribution = 51.32 degrees larger than that of normal distribution
    • Further discrepancies shown through the KL Divergence score of 74053.33 calculated using the bearing changes of the two distributions
    • Bearing change shown to be an effective metric in fake trajectory detection
  • Future Work:
    • Usage of AI and ML models to streamline fake trajectory detection
    • LSTMs for long-term anomaly detection
    • Analysis of different types of ships
    • Incorporation of causal data to reduce the bias in anomaly detection

Expanded Applications of Bearing-Based Anomaly Detection through Causal AI

  • Date: January 2025 to present (ongoing)
  • Description: Research on extending bearing analysis to broader AI/ML applications across maritime and logistics domains while providing causal analysis to reduce false classifications of fake trajectories.
  • Details: Currently, the solution of using bearing change for fake trajectory data does not account for external factors and latent confounders. For example, if there was a storm on the ocean, it could cause a ship to change direction in a way that is not anomalous. To reduce bias and increase accuracy in anomaly detection, causal data is necessary.
  • Key Objectives:
    • Utilize Carnegie Mellon’s TETRAD for Fast Causal Inference (FCI) algorithms and charts mapping causality. This allows for the influences of various independent factors to be graphed, and it also includes potential latent confounders that bring bias to the results
    • Align with CMU Professor Kun Zhang’s focus on transfer learning and causation by first training AI/ML models on the current datasets and then expanding. Specifically, the new datasets will include ship trajectories from 3 new locations: the Bay of Bengal, the North Sea, and the Mediterranean Sea. Along with this, the models will be tested on new types of ships → fishing and cargo ships. Expanding the usage of these models will further help in identifying an accurate fake trajectory detection solution for all parts of the maritime industry
    • Adding weather data from the specific locations/times being studied as independent variables that could be confounders into the dataset to be analyzed by using FCI
    • Addressing the philosophical/ethical aspects of fake trajectory detection → by adding concrete causation to fake trajectory detection, it will greatly reduce the risk of innocents being accused of data spoofing and will overall better the security of the maritime industry

Exploring Transfer Learning in Coordination with Causal Analysis through Wildlife Conservation and AI

Wildlife Preservation Applications of Bearing-Based Anomaly Detection

  • Date: February 2025 to present (ongoing)
  • Description: Research on adapting bearing analysis for wildlife preservation and reproductive monitoring.
  • Details: By incorporating transfer learning, the AI models used for bearing-based anomaly detection can be expanded to analyzing wildlife. Specifically, this research will focus on endangered species wearing trackers. These trackers are designed to ensure that the species are safe and moving. However, due to extreme risk of poaching, corpses of some animals might be placed in rivers or on vehicles to make it seem like they are moving. Hence, anomaly detection is required for this field.
  • Key Contributions:
    • Anti-poaching movement pattern analysis
    • Reproductive health monitoring system
    • 37% faster response times to conservation crises
    • Species-specific pattern recognition

Research Areas

  • Artificial Intelligence
  • Machine Learning
  • Maritime Technology
  • Data Science
  • Pattern Recognition
  • Logistics Optimization
  • Cyber-Physical Security
  • Wildlife Conservation
  • Environmental Monitoring

This page will be updated as new research is published.

Graph View