Benchmarking and evaluation resources for trust, robustness, and reliability analysis in graph-based AI systems.
-
Updated
Oct 24, 2026 - Python
Benchmarking and evaluation resources for trust, robustness, and reliability analysis in graph-based AI systems.
(ECCV 2026) Official repository for MedQ-Deg, a benchmark evaluating medical VLM robustness to image degradation. Features 24,894 QA pairs across 7 imaging modalities, 18 degradation types, and 30 clinical capabilities, with multi-severity evaluation and confidence calibration analysis.
Official Code for "Can These Views Be One Scene?"
GUI-first transfer learning studio for image classification, with custom model generation, PEFT fine-tuning strategies, experiment tracking, robustness evaluation, Grad-CAM inspection, and report-ready visual comparison.
PyTorch + Hugging Face implementation of GIB-200K, a 200,000-item benchmark testing whether vision–language models trust retrieved text or their own eyes. Accuracy collapses from 92.5–94.2% to 41.3–59.9% under stress; up to 71.4% of correct answers abandoned when the document lies; LLaVA over-weights text up to 1.32×; every number auditable.
Computational Stability of Cubical Homology: Orthogonal Controls for Fault Isolation in Retinal Diagnostics
Diffusion-based adversarial purification framework with configurable noise mixing and evaluation on CIFAR-10 and ImageNet.
Traitor-Tracer is an invisible image watermarking system for source tracking and traitor identification, with built-in evaluation against common image attacks such as cropping, resizing, noise, and compression.
Code for FusionAudit: Pathway-Conditioned Robustness Auditing for Native Multimodal Models (NeurIPS 2026)
Reproducible artifact for the ICE2CT-2026 paper Dual-Stream Phase-Aware Inception-Time CNN for adversarially robust V2X spectrum sensing: CPU-only dual-stream CNN (86k params), FGSM/PGD/APGD/FAB/Square evaluation, corrected metrics (robust accuracy + conditional ASR), one-command reproduction, and a full honesty audit of legacy results.
AI in Education research bundle auditing an enrollment-time dropout-risk model with calibration, subgroup evidence, robustness, uncertainty, and governance boundaries.
Official implementation of the CMF-Lite lightweight cross-cue fusion detector for deepfake video detection accepted at MIWAI2026
TriAudit: An empirical audit of adversarially trained image classifiers through clean utility, adversarial robustness, and confidence calibration.
To associate your repository with the robustness-evaluation topic, visit your repo's landing page and select "manage topics."