arXiv subjects

Yuhao Lei

Publications and source records attributed to Yuhao Lei.

8 recordsLinked to original sources

arXiv2026

Online carpooling against an oblivious adversary

In the online carpooling problem, edges on $n$ vertices arrive one at a time and must be oriented immediately, so as to keep the difference between the indegree and the outdegree of every vertex small. We give a randomized online algorithm which bounds the largest absolute difference over the first $T$ edges by $O((\log T)^{1/3})$ with high probability on every input sequence fixed in advance. This proves a conjecture of Kulkarni, Reis and Rothvoss and matches the known worst case lower bound. The algorithm keeps the law of the discrepancy vector dominated in the convex order by a Gaussian measure, and forces the endpoint with the smaller discrepancy to receive the edge whenever the two discrepancies are far apart. In the analysis we show that a vertex of discrepancy $D$ is preceded by a set of $O(D)$ vertices with total discrepancy $Ω(D^2)$.

cs.DS↗

arXiv2026

Event-Aligned Visual Action Reasoning for World Action Models

World-Action Models (WAMs) utilize future visual prediction as an intermediate reasoning process to guide action generation. However, existing WAMs typically structure visual imagination according to predefined temporal intervals, without explicitly accounting for the different roles of task-critical interactions and connecting transitions. We argue that effective visual foresight should align directly with task-relevant interactions and their corresponding reasoning demands. To this end, we introduce an event-aligned visual action reasoning framework that organizes visual-action prediction around interaction events. Through event-aligned visual-action supervision, WAM learns to generate event-aligned visual context in each imagined rollout, placing greater emphasis on critical state changes that inform action generation. This shapes the visual reasoning granularity according to the underlying interaction dynamics, with detailed reasoning around task-critical events and coarser progression through connecting transitions. Furthermore, we introduce an execution validity head that identifies the valid portion of each predicted action sequence, avoiding redundant actions during chunked inference. Experiments demonstrate a 10.26 percentage point improvement in DOMINO success rate over baseline and competitive performance on RoboTwin 2.0. It also transfers from DOMINO Level 1 to Levels 2 and 3 without target-level adaptation.

cs.CV↗

arXiv2026

Q-WAM: 4-Bit Quantization of World Action Models with Action-Subspace Protection

World Action Models (WAMs) jointly generate video and robot actions through iterative diffusion and perform strongly in robotic manipulation. However, their prohibitive compute and memory costs pose substantial deployment challenges. Post-training quantization (PTQ) can reduce these costs, but existing PTQ methods such as smoothing and rotation are insufficient to maintain the precision of action generation. To overcome this limitation, we propose Q-WAM, a new 4-bit weight-activation quantization for WAMs that preserves the actions the model generates. Specifically, we introduce the \textit{Action Observability Gramian (AOG)}, which measures how much rounding errors in each weighted combination of a layer's input channels change the final action through all denoising steps. We also develop Action-Subspace Protection (ASP), which keeps the few most action-sensitive channel combinations in a tiny 16-bit low-rank branch and quantizes the complementary weights and activations to 4 bits, both as dense matrix multiplications that run efficiently on GPUs. Finally, to preserve action quality with minimal overhead, we identify the experts that matter most for the generated action by aggregating the AOG-derived action mass across the layers of each expert and apply ASP only to those experts. We evaluate Q-WAM on three WAMs, both in simulation and in real-world deployment. On the RoboTwin 2.0 benchmark, it reaches 89.6--93.0\% average success rate, within 1.1 percentage points of the 16-bit models, while reducing the memory of the quantized blocks by 3.1--3.4$\times$. Our method outperforms the strongest baseline, SVDQuant, by 2.5--8.7 percentage points. On a Unitree G1 humanoid and a bimanual UR3 robot, it improves success over SVDQuant by 12.8-17.6 percentage points.

cs.RO↗

arXiv2026

Vis-Poison: Poisoning Visual Knowledge in Multimodal Retrieval-Augmented Generation

While multimodal retrieval-augmented generation (RAG) systems increasingly rely on images as external knowledge sources, the introduction of poisoned visual evidence can severely compromise multimodal large language model (MLLM) generation. Unlike prior attacks that rely on altering textual metadata, we introduce Vis-Poison, a novel visual knowledge poisoning attack where the poisoned image itself is the attacker-controlled payload, without manipulating captions, summaries, metadata, or other associated text. Specifically, this attack is instantiated through an automated multi-agent method that constructs visually plausible poisoned images. To assess its impact, we evaluate Vis-Poison across two representative multimodal RAG pipelines, four embedding models, and six generation models. Empirically, Vis-Poison achieves an end-to-end attack success rate of 40.16% to 65.40% against 30k-entry multimodal knowledge bases in \emph{black-box} settings. Moreover, Vis-Poison remains effective against various MLLMs that can answer correctly from parametric knowledge alone, with an average success rate above 60%. Code and data are available at https://github.com/SWUFE-DB-Group/Vis-Poison.

cs.CV↗

arXiv2026

Common sublattice-pure van Hove singularities in the kagome superconductors $\textit{A}$V$_{3}$Sb$_{5}$ ($\textit{A}$ = K, Rb, Cs)

Kagome materials offer a versatile platform for exploring correlated and topological quantum states, where van Hove singularities (VHSs) play a pivotal role in driving electronic instabilities, exhibiting distinct behaviors depending on electron filling and interaction settings. In the recently discovered kagome superconductors $\textit{A}$V$_{3}$Sb$_{5}$ ($\textit{A}$ = K, Rb, Cs), unconventional charge density wave order, superconductivity, and electronic chirality emerge, yet the nature of VHSs near the Fermi level ($\textit{E}$$_{F}$) and their connection to these exotic orders remain elusive. Here, using high-resolution polarization-dependent angle-resolved photoemission spectroscopy, we uncover a universal electronic structure across $\textit{A}$V$_{3}$Sb$_{5}$ that is distinct from density-functional theory predictions that show noticeable discrepancies. We identify multiple common sublattice-pure VHSs near $\textit{E}$$_{F}$, arising from strong V-$\textit{d}$/Sb-$\textit{p}$ hybridization, which significantly promote bond-order fluctuations and likely drive the observed charge density wave order. These findings provide direct spectroscopic evidence for hybridization-driven VHS formation in kagome metals and establish a unified framework for understanding the intertwined electronic instabilities in $\textit{A}$V$_{3}$Sb$_{5}$.

cond-mat.supr-con↗

arXiv2025

Strain energy enhanced room-temperature magnetocaloric effect in second-order magnetic transition materials

Large magnetic entropy change (deltaSM) can realize a prominent heat transformation under the magnetic field and directly strengthen the efficacy of the magnetocaloric effect, which provides a pioneering environmentally friendly solid-state strategy to improve refrigeration capacities and efficiencies. The second-order magnetic transition (SOMT) materials have broader deltaSM peaks without thermal hysteresis compared with most first-order magnetic transition materials, making them highly attractive in magnetic refrigeration, especially in the room temperature range. Here, we report a significant enhancement of deltaSM at room temperature in single-crystal Mn5Ge3. In this SOMT system, we realize a 60% improvement of -deltaSM from 3.5 J/kgK to 5.6 J/kgK at T = 300K. This considerable enhancement of deltaSM is achieved by intentionally introducing strain energy through high-pressure constrained deformation. Both experimental results and Monte Carlo simulations demonstrate that the enhancement of deltaSM originates from the microscopic strain and lattice deformation induced by strain energy after deformation. This strain energy will reconstruct the energy landscape of this ferromagnetic system and enhance magnetization, resulting in a giant intensity of magnetocaloric responses. Our findings provide an approach to increase magnetic entropy change and may give fresh ideas for exploring advanced magnetocaloric materials.

cond-mat.mtrl-sci↗

arXiv2024

Controlling ultrafast laser writing in silica glass by pulse temporal contrast

Control of laser writing in transparent dielectrics using temporal contrast of light pulses is demonstrated. Anisotropic nanopores in silica glass are produced by high-contrast of 10^7 femtosecond Yb: KGW laser pulses rather than low-contrast of 10^3 Yb fiber laser pulses. Low-contrast pulses are useful for creating lamellar birefringent structures, possibly driven by quadrupole nonlinear current. The difference originates in the fiber laser storing third of its energy in a post-pulse of 200 ps duration. The absorption of this low-intensity fraction of the pulse by laser-induced transient defects with relatively long lifetime and low excitation energy, such as self-trapped holes, drastically changes the kinetics of energy deposition and type of material modification.

physics.optics↗

arXiv2023

A double-clad ytterbium-doped tapered fiber with circular birefringence as a gain medium for structured light

Amplifying radially and azimuthally polarized beams is a significant challenge due to the instability of the complex beam shape and polarization in inhomogeneous environment. In this Letter, we demonstrated experimentally an efficient approach to directly amplify cylindrical-vector beams with axially symmetric polarization and doughnut-shaped intensity profile in a picosecond MOPA system based on a double-clad ytterbium-doped tapered fiber. To prevent polarization and beam shape distortion during amplification, for the first time to the best of our knowledge, we proposed using the spun architecture of the tapered fiber. In contrast to an isotropic fiber architecture, a spun configuration possessing nearly-circular polarization eigenstates supports stable wavefront propagation. Applying this technique, we amplified the cylindrical-vector beam up to 22 W of average power with 10 ps pulses at a central wavelength of 1030 nm and a repetition rate of 15 MHz, maintaining both mode and polarization stability.

physics.optics↗