arXiv subjects

Yiping Mao

Publications and source records attributed to Yiping Mao.

3 recordsLinked to original sources

arXiv2026

Cross-Evidence Authorization State Learning for Runtime Detection of Broken Object-Level Authorization in Black-Box Web Applications

Broken object-level authorization (BOLA) remains difficult to detect in black-box Web applications because external runtime monitors generally lack access to source code, authorization policies, and object-ownership records. This paper proposes a cross-evidence authorization-state learning method that reconstructs subject--resource--object--action relations from request--response transactions. The method represents authorization relations with confidence-aware states, aggregates positive and negative evidence while accounting for provenance independence, and applies contamination-gated promotion, evidence quarantine and rollback, expiration, and revocation. A Python prototype is evaluated using a deterministic synthetic benchmark comprising 64,400 requests across five independent random seeds. In this benchmark, the proposed method achieves 100.00% recall, an F1-score of 56.63%, and a false-positive rate of 33.65%, while the measured unauthorized-relation pollution rate is 0.00%. Ablation experiments show that removing provenance independence or evidence quarantine increases the pollution rate to 100.00%. The experiments provide an initial evaluation under controlled synthetic conditions.

cs.CR↗

arXiv2026

REALM: An MLLM-Agent Framework for Open World 3D Reasoning Segmentation and Editing on Gaussian Splatting

Bridging the gap between complex human instructions and precise 3D object grounding remains a significant challenge in vision and robotics. Existing 3D segmentation methods often struggle to interpret ambiguous, reasoning-based instructions, while 2D vision-language models that excel at such reasoning lack intrinsic 3D spatial understanding. In this paper, we introduce REALM, an innovative MLLM-agent framework that enables open-world reasoning-based segmentation without requiring extensive 3D-specific post-training. We perform segmentation directly on 3D Gaussian Splatting representations, capitalizing on their ability to render photorealistic novel views that are highly suitable for MLLM comprehension. As directly feeding one or more rendered views to the MLLM can lead to high sensitivity to viewpoint selection, we propose a novel Global-to-Local Spatial Grounding strategy. Specifically, multiple global views are first fed into the MLLM agent in parallel for coarse-level localization, aggregating responses to robustly identify the target object. Then, several close-up novel views of the object are synthesized to perform fine-grained local segmentation, yielding accurate and consistent 3D masks. Extensive experiments show that REALM achieves remarkable performance in interpreting both explicit and implicit instructions across LERF, 3D-OVS, and our newly introduced REALM3D benchmarks. Furthermore, our agent framework seamlessly supports a range of 3D interaction tasks, including object removal, replacement, and style transfer, demonstrating its practical utility and versatility. Project page: https://ChangyueShi.github.io/REALM.

cs.CV↗

arXiv1995

Geodesic Conjugacy in two-step nilmanifolds

Two Riemannian manifolds are said to have $C^k$-conjugate geodesic flows if there exist an $C^k$ diffeomorphism between their unit tangent bundles which intertwines the geodesic flows. We obtain a number of rigidity results for the geodesic flows on compact 2-step Riemannian nilmanifolds: For generic 2-step nilmanifolds the geodesic flow is $C^2$ rigid. For special classes of 2-step nilmanifolds, we show that the geodesic flow is $C^0$ or $C^2$ rigid. In particular, there exist continuous families of 2-step nilmanifolds whose Laplacians are isospectral but whose geodesic flows are not $C^0$ conjugate.

math.DG↗