Publications Google Scholar

  1. Judging by the Cover: Cleaning LLM Truthfulness Benchmarks to Avoid Surface-Level Feature Leakage
    Judging by the Cover: Cleaning LLM Truthfulness Benchmarks to Avoid Surface-Level Feature Leakage
    Foad Namjoo, Remy Ogasawara, Amirali Abdullah, Cullen Anderson, Narmeen Fatimah Oozeer, and Jeff M. Phillips
    arXiv 2026, under review at ICLR 2027
    Summary: A six-feature surface probe that never reads the question separates TruthfulQA’s correct from incorrect answers (AUC 0.715); Audit-Prune removes the leaking pairs, and TruthfulQA-476 keeps the benchmark’s model ranking with the shortcut gone.
  2. Sampling for Region-Aggregated Spatial Scan Statistics
    Sampling for Region-Aggregated Spatial Scan Statistics
    Foad Namjoo, Drew McClelland, Michael Matheny, and Jeff M. Phillips
    In Proceedings of the 34th ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems (SIGSPATIAL), 2026
    Accepted as a full paper (58/248, 23.4%).
    Summary: Replacing each region with 20–50 sampled points improves the statistical power that centroid-collapsing loses — a drop-in fix for point-based spatial scan statistics on aggregated data.
  3. Orthogonal and Parametrized Multi-Attribute Steering of LLMs
    Orthogonal and Parametrized Multi-Attribute Steering of LLMs
    Remy Ogasawara, Cullen Anderson, Narmeen Oozeer, Foad Namjoo, Amirali Abdullah, Jens Kristian Refsgaard Schou, and Jeff M. Phillips
    arXiv 2026, under review at ICLR 2027
    Summary: Steering vectors for related traits point in similar directions, so pushing one drags the others; rescaling each vector to its most steerable range, setting a trait’s level instead of pushing it, and orthogonalizing the directions steer several traits at once without over- or under-steering.
  4. Understanding and Mitigating Dataset Corruption in LLM Steering
    Understanding and Mitigating Dataset Corruption in LLM Steering
    Cullen Anderson, Narmeen Oozeer, Foad Namjoo, Remy Ogasawara, Amirali Abdullah, and Jeff M. Phillips
    Findings of EACL 2027
    Accepted; to appear.
    Summary: How noisy and adversarial corruption in steering datasets degrades LLM control — and robust estimation that mitigates it.
  5. Designing a Secure and Resilient Distributed Smartphone Participant Data Collection System
    Designing a Secure and Resilient Distributed Smartphone Participant Data Collection System
    Foad Namjoo, Neng Wan, Devan Mallory, Yuyi Chang, Nithin Sugavanam, Long Yin Lee, Ning Xiong, Emre Ertin, and Jeff M. Phillips
    In Proceedings of the EAI International Conference on Security and Privacy in Cyber-Physical Systems and Smart Vehicles (SmartSP), 2025
    Summary: The system design behind MotionPI: secure, resilient, privacy-first mobile data collection at scale.
  6. Addressing Data Limitations to Explore Management Strategies and Adaptations Using Stylized Agent-Based Modeling: A Case Study of Socio-Environmental Systems
    Addressing Data Limitations to Explore Management Strategies and Adaptations Using Stylized Agent-Based Modeling: A Case Study of Socio-Environmental Systems
    Mehrsa Pouladi, Parsa Pouladi, Saba Naderian Jahromi, and Foad Namjoo
    Ecological Modelling, 2025
    Summary: Stylized agent-based modeling to explore management strategies for socio-environmental systems under limited data.
  7. Efficient and Stable Multi-Dimensional Kolmogorov–Smirnov Distance
    Efficient and Stable Multi-Dimensional Kolmogorov–Smirnov Distance
    Peter Matthew Jacobs, Foad Namjoo, and Jeff M. Phillips
    arXiv 2025, under review after minor revisions at Foundations of Data Science
    Authors listed alphabetically, following the convention in theory.
    Summary: A multi-dimensional Kolmogorov–Smirnov distance that’s a true metric, with a stable finite-sample two-sample test, near-linear in 2–4D.
  8. A Hyperspectral Change Detection (HCD-Net) Framework Based on Double Stream Convolutional Neural Networks and an Attention Module
    A Hyperspectral Change Detection (HCD-Net) Framework Based on Double Stream Convolutional Neural Networks and an Attention Module
    Seyd Teymoor Seydi, Mahboubeh Boueshagh, Foad Namjoo, Seyed Mohammad Minouei, Zahir Nikraftar, and Meisam Amani
    Remote Sensing, 2024
    Summary: A dual-stream CNN with an attention module for change detection in hyperspectral imagery.
  9. A Machine Learning Approach for Harmful Algal Bloom (Red Tide) Forecasting Using MODIS Level 3 Ocean Colour Products from Google Earth Engine
    Moein Izadi, Foad Namjoo, and Zahir Nikraftar
    In AGU Fall Meeting Abstracts, 2021
    Summary: Forecasting harmful algal blooms (red tide) from MODIS ocean-colour products on Google Earth Engine.
  10. Remote Sensing & Statistical Learning Approach to Harmful Algal Bloom Forecasting Using MODIS Ocean Colour Parameters
    Moein Izadi, Mohamed Sultan, Racha Kadiri, Amin Ghannadi, Zahir Nikraftar, and Foad Namjoo
    In AGU Fall Meeting Abstracts, 2020
    Summary: A statistical-learning approach to harmful algal bloom forecasting from MODIS ocean-colour parameters.
  11. Australian 2020 Bushfire Propagation, Network Science Approach
    Foad Namjoo, Ali Kamandi, and Seyed Mohammad Minouei
    In National Conference on Vision of the Country, 2020
    Summary: Modeling the 2020 Australian bushfire propagation with a network-science approach.