Khawaja Abaid Ullah

THE STUDY READING TABLE // PAPERS & PREPRINTS

Selected Manuscripts & Preprints

Research manuscripts on margin-aware inference-time antidistillation defenses against model extraction (ADS-C and matched-cost benchmark review) and theory-grounded evaluation across 34 LLMs and 198 human participants.

STUDY MANUSCRIPT // SOCIAL-ATTRACTION

How Do Large Language Models Judge Social Attraction? Evidence from Theory-Grounded Persona Ratings Across Multiple LLMs and Humans

Hasan Mahmud, Khawaja Abaid Ullah, Mohammad Javad Khojasteh, Jamison Heard, Prabu David

34 LLMs · 198 Humans
evaluated across theory-grounded persona ratings — Controlled analysis of human-LLM alignment and gender presentation

Evaluated social-judgment reliability and human-LLM alignment across 34 large language models and 198 human participants using theory-grounded persona ratings, including controlled analyses of gender presentation.

arXiv:2608.09717 · Under review
Paper (arXiv) Study Code
Open BibTeX Citation for How Do Large Language Models Judge Social Attraction? Evidence from Theory-Grounded Persona Ratings Across Multiple LLMs and Humans
@misc{mahmud2026socialattraction,
  title = {How Do Large Language Models Judge Social Attraction? Evidence from Theory-Grounded Persona Ratings Across Multiple LLMs and Humans},
  author = {Hasan Mahmud and Khawaja Abaid Ullah and Mohammad Javad Khojasteh and Jamison Heard and Prabu David},
  year = {2026},
  eprint = {2608.09717},
  archivePrefix = {arXiv},
  primaryClass = {cs.CL},
  url = {https://arxiv.org/abs/2608.09717}
}
STUDY MANUSCRIPT // REVERSING-INCENTIVE

Reversing the Incentive to Distill: A Review of Test-Time Antidistillation Defenses for Classification

Khawaja Abaid Ullah, Mohammad Javad Khojasteh

−34.5 pp / −47.7 pp
below hard-label baseline on CIFAR-10 and CIFAR-100 — Unified matched-cost benchmark & method-agnostic top-1 wrapper

Built a unified matched-cost benchmark and a method-agnostic top-1-preservation wrapper; directed perturbations drove distilled students 34.5 pp (CIFAR-10) and 47.7 pp (CIFAR-100) below the hard-label baseline. Reimplemented prior test-time defenses under a common threat model and matched-cost evaluation protocol to isolate which perturbation mechanisms genuinely suppress student learning.

Manuscript under review (2026) · Under review
Related ADS-C Folio
Open BibTeX Citation for Reversing the Incentive to Distill: A Review of Test-Time Antidistillation Defenses for Classification
@unpublished{ullah2026reversing,
  title = {Reversing the Incentive to Distill: A Review of Test-Time Antidistillation Defenses for Classification},
  author = {Khawaja Abaid Ullah and Mohammad Javad Khojasteh},
  year = {2026},
  note = {Under review}
}
STUDY MANUSCRIPT // ADS-C

ADS-C: Antidistillation Sampling for Classification

Khawaja Abaid Ullah, Mohammad Javad Khojasteh

−29.6 pp
CIFAR-10 student accuracy (−17.4 pp CIFAR-100 · −13.3 pp Tiny-ImageNet) — 0 pp served top-1 accuracy cost

Proposed a margin-aware inference-time defense that preserves every served top-1 prediction while reducing distilled-student accuracy by 29.6 pp on CIFAR-10, 17.4 pp on CIFAR-100, and 13.3 pp on Tiny-ImageNet. Released with an open-source PyTorch evaluation harness across all three benchmarks and multiple student architectures.

arXiv:2607.15467 · Under review · IEEE OJCS
Paper (arXiv) PyTorch Code Research Folio
Open BibTeX Citation for ADS-C: Antidistillation Sampling for Classification
@misc{ullah2026adsc,
  title = {ADS-C: Antidistillation Sampling for Classification},
  author = {Khawaja Abaid Ullah and Mohammad Javad Khojasteh},
  year = {2026},
  eprint = {2607.15467},
  archivePrefix = {arXiv},
  primaryClass = {cs.LG},
  url = {https://arxiv.org/abs/2607.15467}
}