Selected work

Representative Papers

Universal Concept Disruption for SAM3 Image Segmentation

Aug 06, 2026

Existing work has not yet investigated the adversarial robustness of SAM3 in open-vocabulary concept segmentation, nor has it developed a universal attack method targeting its joint concept–image system. This work proposes Universal Concept Disruption (UCD), the first approach enabling end-to-end, universal cross-concept adversarial attacks on SAM3. UCD learns a single bounded perturbation that simultaneously disrupts textual conditioning inputs, visual feature consistency, existence gating scores, and mask spatial validity. Notably, this perturbation generalizes across datasets, model variants (e.g., SAM3.1), and video inference without re-optimization. Experiments demonstrate that UCD substantially degrades model performance, reducing average mask AP from 59.43 to 18.73 and cgF1 from 50.32 to 20.49, thereby confirming its strong transferability and effectiveness.

0 citationsRead paper

LoRe: Adaptive Interaction-Evaluation Routing with Per-Step Interaction Budgets for Iterative Graph Solvers

May 27, 2026

This work addresses the high computational cost and memory bottlenecks in diffusion-based neural combinatorial optimization solvers during inference, which stem from dense edge or factor interactions. To this end, we propose LoRe—a training-free, plug-and-play dynamic interaction selection mechanism that operates at inference time. LoRe introduces, for the first time, the concept of dynamic routing from many-body physics into graph neural solvers, adaptively selecting critical interactions at each step based on conflict and uncertainty metrics while enforcing strict budget constraints. Without requiring any retraining, LoRe substantially enhances scalability: on the maximum independent set problem, it enables over 3× larger instance sizes, 8× faster inference, and a 12× reduction in peak memory; on thousand-node TSP instances, it achieves 15× speedup and 44× less memory usage, all while maintaining competitive solution quality.

0 citationsRead paper

Recent publications

Latest Papers

Universal Concept Disruption for SAM3 Image Segmentation

Aug 06, 2026

Existing work has not yet investigated the adversarial robustness of SAM3 in open-vocabulary concept segmentation, nor has it developed a universal attack method targeting its joint concept–image system. This work proposes Universal Concept Disruption (UCD), the first approach enabling end-to-end, universal cross-concept adversarial attacks on SAM3. UCD learns a single bounded perturbation that simultaneously disrupts textual conditioning inputs, visual feature consistency, existence gating scores, and mask spatial validity. Notably, this perturbation generalizes across datasets, model variants (e.g., SAM3.1), and video inference without re-optimization. Experiments demonstrate that UCD substantially degrades model performance, reducing average mask AP from 59.43 to 18.73 and cgF1 from 50.32 to 20.49, thereby confirming its strong transferability and effectiveness.

0 citationsRead paper

LoRe: Adaptive Interaction-Evaluation Routing with Per-Step Interaction Budgets for Iterative Graph Solvers

May 27, 2026

This work addresses the high computational cost and memory bottlenecks in diffusion-based neural combinatorial optimization solvers during inference, which stem from dense edge or factor interactions. To this end, we propose LoRe—a training-free, plug-and-play dynamic interaction selection mechanism that operates at inference time. LoRe introduces, for the first time, the concept of dynamic routing from many-body physics into graph neural solvers, adaptively selecting critical interactions at each step based on conflict and uncertainty metrics while enforcing strict budget constraints. Without requiring any retraining, LoRe substantially enhances scalability: on the maximum independent set problem, it enables over 3× larger instance sizes, 8× faster inference, and a 12× reduction in peak memory; on thousand-node TSP instances, it achieves 15× speedup and 44× less memory usage, all while maintaining competitive solution quality.

0 citationsRead paper