Searcharxiv⌕ Search

arXiv subjects

Yuxi Liu

Publications and source records attributed to Yuxi Liu.

At least 19 recordsLinked to original sources
arXiv2026

CrossDistill: Balancing Quality and Diversity via Trajectory-Level Hybrid Few-Step Distillation

Few-step distillation accelerates diffusion models but must balance diversity and fidelity: trajectory-based distillation preserves mode coverage, while distribution matching sharpens samples but can reduce diversity. We show that this tension can be exploited in a noise-regime-dependent way: high-noise steps largely determine global modes, whereas low-noise steps refine local details. We propose CrossDistill, a trajectory-level hybrid distillation framework that splits the sampling trajectory at a crossover point, applies a trajectory-preserving objective on the high-noise interval and a distribution-matching objective on the low-noise interval, and couples the two stages through the crossover state. In contrast to loss-level mixing, and complementarily to training-time two-stage recipes, CrossDistill explicitly assigns complementary objectives along the noise axis, so that global branching is preserved before local statistics are sharpened. CrossDistill is a noise-level scheduling policy: PCM and DMD are plug-in instantiations, while the noise partition, crossover coupling, and objective ordering are the key design elements. Experiments on text-to-video diffusion models and qualitative image-to-video results show that CrossDistill expands the few-step quality-diversity frontier, retaining seed-level variation while achieving competitive visual fidelity.

cs.CV↗
arXiv2026

RoLA: Rotary-Positioned Low-Rank Linear Attention for Efficient Diffusion Transformers

Diffusion Transformers (DiTs) achieve strong video generation quality, but their dense spatiotemporal self-attention scales quadratically with sequence length and quickly becomes the dominant inference bottleneck. Sparse low-rank hybrids alleviate this cost by combining a local sparse branch with a global compressed branch. In video DiTs equipped with 3D Rotary Position Embeddings (RoPE), the global branch faces a structural compatibility issue: when RoPE is applied before a nonlinear feature map, the rotation and nonlinearity generally do not commute, making it difficult to keep a query-independent linear summary while preserving relative rotary geometry. Existing work often sidesteps this issue by replacing genuine cross-token global aggregation with coordinate-conditioned surrogates or learnable absolute positional modules. These compromises can be effective, but they approximate relative decay from absolute coordinates and introduce extra positional parameters. We propose \textbf{RoLA}, a rotary-positioned low-rank linear-attention branch that keeps genuine cross-token aggregation while remaining compatible with a reusable linear summary. The design applies RoPE \emph{outside} the nonlinear low-rank feature map and reuses a truncated subset of the pre-trained rotary schedule matched to the low-rank bottleneck. This yields a linear-time low-rank global branch with relative positional behavior by design and no additional positional parameters; the full sparse--low-rank module still includes the fixed-sparsity sparse branch. Experiments on open-source video DiTs show that the resulting method remains competitive in generation quality at 90\% sparsity while achieving 2.63$\times$ end-to-end inference speedup on Wan2.1-14B (720p, 81 frames, measured on an NVIDIA H100 GPU).

cs.CV↗
arXiv2026

World-Model-Augmented Visual Locomotion for Humanoids on Foothold-Constrained Terrain

Foothold-constrained terrain is characterized by sparse, discontinuous, or geometrically restricted feasible foot contacts, as encountered on stepping stones, across gaps, and on narrow stair treads. On such terrain, a single misstep often leaves little room to recover, so policies that base foot-placement decisions primarily on the immediately visible terrain are prone to failure. We ask whether a learned predictive summary of near-future observations and rewards can provide the anticipatory information required in such settings. We present World-Model-Augmented Visual Locomotion (WM-LOCO), which jointly trains a recurrent world model and a PPO policy. Conditioned on proprioception and a single onboard depth image, the world model produces a predictive recurrent feature that guides the policy, without explicit foothold labels. In simulation, WM-LOCO succeeds on gaps and stepping stones where a matched baseline fails completely, and matches the baseline's success rate on stairs while improving stride efficiency and reducing pelvis acceleration. We deploy the same policy onboard a physical Unitree G1 humanoid using onboard proprioception and a single depth stream; it traverses all three terrain classes with an average success rate of 93.3%.

cs.RO↗
arXiv2026

Real-time Whole-Body Motion Planning for Mobile Manipulators Carrying Arbitrarily Shaped Payloads via Kinematically-Coupled SVSDF

Mobile manipulators are increasingly tasked with transporting large, non-convex payloads through cluttered environments, yet existing planners either oversimplify the payload geometry or fail to handle the kinematic coupling between manipulator links, leading to lost feasible space or stalled optimization. This letter presents a real-time whole-body motion planning framework for mobile manipulators carrying arbitrarily shaped payloads. The front-end employs a chain-decomposed kernel-based collision check that preserves the true geometry of the robot and payload, with compact storage and fast bit-level queries. A mid-end preprocessing stage converts the front-end path into a continuous trajectory enforcing smoothness and feasibility, and executes it directly when collision-free to bypass the costly back-end. When refinement is required, the back-end performs trajectory optimization built on a Kinematically-Coupled SVSDF (KC-SVSDF), which propagates collision-avoidance gradients along the kinematic chain to produce coherent whole-body escape directions. Ablation studies, comparative benchmarks against state-of-the-art baselines, and real-world experiments on a differential-drive mobile manipulator demonstrate that the proposed framework reliably transports large, non-convex payloads through tight passages and cluttered environments.

cs.RO↗
arXiv2026

RhinoVLA Technical Report

Vision-Language-Action (VLA) models have shown strong potential for robotic manipulation, but real-time deployment on edge hardware remains challenging. In this work, we identify VLM visual and context tokens as a major source of deployment latency: for GEMM-dominated projection operators, computation grows linearly with the number of input tokens when model dimensions are fixed. Motivated by this observation, we propose RhinoVLA, a deployment-oriented VLA model co-designed with the Huixi R1 edge SoC. RhinoVLA adopts a token-efficient Qwen3-VL backbone and a continuous Action Expert, reducing the VLM-side token and computation burden while preserving pretrained multimodal capability. To support cross-robot learning, RhinoVLA further introduces a unified interface that combines View Registry, 72D physical state-action slot space, and robotinstance LoRA, allowing heterogeneous robot observations and action schemas to be aligned under a shared policy. On the deployment side, RhinoVLA is optimized through hardware-aware compilation, mixed-precision execution, and parallel visual encoding. Experiments show that RhinoVLA achieves downstream performance comparable to π0.5 at a similar parameter scale, while reaching 11.69 Hz end-to-end inference on Huixi R1, meeting the 10 Hz real-time closedloop control target. The project will be open-sourced at https://github.com/HuixiAI/RhinoVLA.

cs.RO↗
arXiv2026

The Seriality Gap in Video Diffusion Models

When one ball strikes another, then another, video models should predict the consequences of each bounce. In controlled experiments on multi-ball hard-sphere dynamics, we find that the performance of standard bidirectional video diffusion degrades as the causal chain lengthens, even when provided more denoising steps. In a length-matched single-ball control, where ball-ball interactions are absent, the degradation largely disappears, isolating dependent-event structure rather than video length as the cause. Across intervention studies, methods that increase effective serial computation improve performance disproportionately, including autoregressive/blockwise generation and architectural depth. We identify this pattern as the seriality gap: a mismatch between tasks requiring growing serial computation and video diffusion models whose denoising loop does not provide scalable serial compute. We then prove that, for deterministic video prediction, denoising steps do not add serial computation beyond the backbone, indicating a structural obstacle for video diffusion on serial reasoning and simulation tasks.

cs.LG↗
arXiv2026

BaCon: Efficient Batch Processing of Counting Queries [Full Version]

Counting queries are ubiquitous in database systems, particularly for driving internal system optimization. Learned models for cardinality estimation rely heavily on large-scale training data, yet generating such data by executing massive batches of counting queries is expensive. We propose BaCon, an efficient algorithm for batch evaluation of counting queries on top of a database system, without modifying its internals. BaCon integrates the idea of factorized databases with a workload-aware domain quantization strategy, allowing it to evaluate batches of counting queries using compact data structures rather than materializing massive join results. BaCon's design is compatible with most database management system, and we have implemented it as a client-side application on PostgreSQL with a lightweight C-language UDF (user-defined function). This implementation delivers speedups between 2$\times$ and 178$\times$ over baselines and good performance across various workloads, making training and maintenance of learned cardinality estimation models significantly more practical.

cs.DB↗
arXiv2026

Mixture of Distributions Matters: Dynamic Sparse Attention for Efficient Video Diffusion Transformers

While Diffusion Transformers (DiTs) have achieved notable progress in video generation, this long-sequence generation task remains constrained by the quadratic complexity inherent to self-attention mechanisms, creating significant barriers to practical deployment. Although sparse attention methods attempt to address this challenge, existing approaches either rely on oversimplified static patterns or require computationally expensive sampling operations to achieve dynamic sparsity, resulting in inaccurate pattern predictions and degraded generation quality. To overcome these limitations, we propose a \underline{\textbf{M}}ixture-\underline{\textbf{O}}f-\underline{\textbf{D}}istribution \textbf{DiT} (\textbf{MOD-DiT}), a novel sampling-free dynamic attention framework that accurately models evolving attention patterns through a two-stage process. First, MOD-DiT leverages prior information from early denoising steps and adopts a {distributed mixing approach} to model an efficient linear approximation model, which is then used to predict mask patterns for a specific denoising interval. Second, an online block masking strategy dynamically applies these predicted masks while maintaining historical sparsity information, eliminating the need for repetitive sampling operations. Extensive evaluations demonstrate consistent acceleration and quality improvements across multiple benchmarks and model architectures, validating MOD-DiT's effectiveness for efficient, high-quality video generation while overcoming the computational limitations of traditional sparse attention approaches.

cs.CV↗
arXiv2026

Weighted Envy-Freeness Revisited: Indivisible Resource and House Allocations

Envy-Freeness is one of the most fundamental and important concepts in fair allocation. Some recent studies have focused on the concept of weighted envy-freeness. Under this concept, each agent is assigned a weight, and their valuations are divided by their weights when assessing fairness. This concept can promote more fairness in some scenarios. But on the other hand, experimental research has shown that this weighted envy-freeness significantly reduces the likelihood of fair allocations. When we must allocate the resources, we may propose fairness concepts with lower requirements that are potentially more feasible to implement. In this paper, we revisit weighted envy-freeness and propose a new concept called SumAvg-envy-freeness, which substantially increases the existence of fair allocations. This new concept can be seen as a complement of the normal weighted envy-fairness. Furthermore, we systematically study the computational complexity of finding fair allocations under the old and new weighted fairness concepts in two types of classic problems: Indivisible Resource Allocation and House Allocation. Our study provides a comprehensive characterization of various properties of weighted envy-freeness.

cs.GT↗
arXiv2026

RoPeSLR: 3D RoPE-driven Sparse-LowRank Attention for Efficient Diffusion Transformers

Diffusion Transformers (DiTs) have revolutionized high-fidelity video generation, yet their $\mathcal{O}(L^2)$ attention complexity poses a formidable bottleneck for long-sequence synthesis. While recent sparse-linear attention hybrids aim to mitigate this, their performance severely degrades at extreme sparsity due to the "RoPE Dilemma": standard linear attention fails to preserve the orthogonal relative-position structure of 3D Rotary Position Embeddings (RoPE), neutralizing vital distance awareness. To address this, we propose \textbf{RoPeSLR}, a 3D RoPE-guided Sparse-LowRank attention framework. We establish that under empirically validated assumptions, the DiT attention manifold admits a decoupling into a high-frequency semantic spike set (bounded by $\mathcal{O}(L^{3/2})$ sparsity) and an extreme low-rank ($\mathcal{O}(d_h \log L)$) background continuum. Guided by this structural prior, RoPeSLR eschews standard linear attention for a head-wise low-rank parameterization equipped with a learnable 3D Absolute Positional Embedding (PE) injection, seamlessly synthesizing long-range relative distance decay. By guaranteeing sub-quadratic sparsity and sub-linear rank growth, RoPeSLR is exceptionally suited for scaling to ultra-long video inference. Extensive evaluations validate this scalable superiority: at 90\% sparsity, RoPeSLR achieves up to $10\times$ fewer FLOPs on Wan2.1-1.3B and delivers a $2.26\times$ end-to-end inference speedup on the ultra-long 100K+ token sequences of HunyuanVideo-13B, all while maintaining near-lossless generation fidelity (less than 1.3\% average VBench degradation).

cs.CV↗
arXiv2026

Linear Kernels for $l$-Exact Component Order Connectivity

The \textsc{$l$-Exact Component Order Connectivity} problem asks whether, given an input graph $G$ and an integer $k$, there exists a vertex subset $S\subseteq V(G)$ of size at most $k$ such that every connected component in $G - S$ has exactly $l$ vertices. In this paper, we present an $O(kl)$-vertex kernel for this problem, computable in $|V(G)|^{O(l)}$ time. This is the first known linear kernel for each fixed $l\geq 3$. For $l=1$, this problem reduces to the classical \textsc{Vertex Cover}, and our result matches the best-known $2k$-vertex kernel. For $l=2$ (known as \textsc{Deletion to Induced Matching}), we can get a $(3k + 1)$-vertex kernel, improving the previously known result of $6k$ vertices. Our kernelization algorithm is built upon on an extended crown decomposition combined with linear programming and other techniques.

cs.DS↗
arXiv2026

Deterministic Single Exponential Time Algorithms for Co-Path Packing and Co-Path Set Parameterized by Treewidth

The \textsc{Co-Path Packing} (resp., \textsc{Co-Path Set}) problem asks whether a given graph can be edited to a collection of induced paths by deleting at most $k$ vertices (resp., $k$ edges). Both are fundamental problems with significant applications in bioinformatics and have been extensively studied within the framework of exact and parameterized algorithms. Currently, the state-of-the-art approach utilizes the randomized ``Cut \& Count'' technique, which solves \textsc{Co-Path Set} in $O^*(4^{\mathbf{tw}})$ time and \textsc{Co-Path Packing} in $O^*(5^{\mathbf{pw}})$ time, where $\mathbf{tw}$ is treewidth and $\mathbf{pw}$ is pathwidth. However, as there is no known method to derandomize the ``Cut \& Count'' technique, the existence of deterministic single exponential time algorithms for these problems parameterized by treewidth has remained an open question. In this paper, we resolve this gap by providing deterministic single exponential time algorithms for both problems when parameterized by treewidth.

cs.DS↗
arXiv2026

CR-Net: Scaling Parameter-Efficient Training with Cross-Layer Low-Rank Structure

Low-rank architectures have become increasingly important for efficient large language model (LLM) pre-training, providing substantial reductions in both parameter complexity and memory/computational demands. Despite these advantages, current low-rank methods face three critical shortcomings: (1) compromised model performance, (2) considerable computational overhead, and (3) limited activation memory savings. To address these limitations, we propose Cross-layer Low-Rank residual Network (CR-Net), an innovative parameter-efficient framework inspired by our discovery that inter-layer activation residuals possess low-rank properties. CR-Net implements this insight through a dual-path architecture that efficiently reconstructs layer activations by combining previous-layer outputs with their low-rank differences, thereby maintaining high-rank information with minimal parameters. We further develop a specialized activation recomputation strategy tailored for CR-Net that dramatically reduces memory requirements. Extensive pre-training experiments across model scales from 60M to 7B parameters demonstrate that CR-Net consistently outperforms state-of-the-art low-rank frameworks while requiring fewer computational resources and less memory.

cs.LG↗
arXiv2026

The Serial Scaling Hypothesis

While machine learning has advanced through massive parallelization, we identify a critical blind spot: some problems are fundamentally sequential. These "inherently serial" problems-from mathematical reasoning to physical simulations to sequential decision-making-require sequentially dependent computational steps that cannot be efficiently parallelized. We formalize this distinction in complexity theory, and demonstrate that current parallel-centric architectures face fundamental limitations on such tasks. Then, we show for first time that diffusion models despite their sequential nature are incapable of solving inherently serial problems. We argue that recognizing the serial nature of computation holds profound implications on machine learning, model design, and hardware development.

cs.LG↗
arXiv2026

Predicting kernel regression learning curves from only raw data statistics

We study kernel regression with common rotation-invariant kernels on real datasets including CIFAR-5m, SVHN, and ImageNet. We give a theoretical framework that predicts learning curves (test risk vs. sample size) from only two measurements: the empirical data covariance matrix and an empirical polynomial decomposition of the target function $f_*$. The key new idea is an analytical approximation of a kernel's eigenvalues and eigenfunctions with respect to an anisotropic data distribution. The eigenfunctions resemble Hermite polynomials of the data, so we call this approximation the Hermite eigenstructure ansatz (HEA). We prove the HEA for Gaussian data, but we find that real image data is often "Gaussian enough" for the HEA to hold well in practice, enabling us to predict learning curves by applying prior results relating kernel eigenstructure to test risk. Extending beyond kernel regression, we empirically find that MLPs in the feature-learning regime learn Hermite polynomials in the order predicted by the HEA. Our HEA framework is a proof of concept that an end-to-end theory of learning which maps dataset structure all the way to model performance is possible for nontrivial learning algorithms on real datasets.

cs.LG↗
arXiv2026

Fairness-Aware Fine-Tuning of Vision-Language Models for Medical Glaucoma Diagnosis

Vision-language models achieve expert-level performance on medical imaging tasks but exhibit significant diagnostic accuracy disparities across demographic groups. We introduce fairness-aware Low-Rank Adaptation for medical VLMs, combining parameter efficiency with explicit fairness optimization. Our key algorithmic contribution is a differentiable MaxAccGap loss that enables end-to-end optimization of accuracy parity across demographic groups. We propose three methods: FR-LoRA integrates MaxAccGap regularization into the training objective, GR-LoRA applies inverse frequency weighting to balance gradient contributions, and Hybrid-LoRA combines both mechanisms. Evaluated on 10,000 glaucoma fundus images, GR-LoRA reduces diagnostic accuracy disparities by 69% while maintaining 53.15% overall accuracy. Ablation studies reveal that strong regularization strength achieves optimal fairness with minimal accuracy trade-off, and race-specific optimization yields 60% disparity reduction. Our approach requires only 0.24% trainable parameters, enabling practical deployment of fair medical AI in resource-constrained healthcare settings.

cs.CV↗
arXiv2026

Dynamic topological exciton-polaritons enabling ultrafast logic operations

Topological active materials have emerged as powerful paradigm bridging the discovery of exotic topological phases of matter with the development of functional topological devices. The recent extension of these material systems into dynamic regime, where topological properties can be actively manipulated at ultrafast timescales, promises unprecedented control over topological states and their functionalities. However, translating the static topological lasing signals into high-performance logic functions remain highly challenging, which imposes a far more stringent set of materials attributes. Here, leveraging the strong nonlinearity and pronounced spectral isolation of perovskite exciton-polaritons embedded in a Dirac vortex microcavity, we experimentally demonstrate the dynamic topological Majorana-like state polariton condensation with its ultrafast logic operations at room temperature. By actively coordinating pump and control beams in both spectral and temporal domain, we dynamically steer the topological polariton condensation process and demonstrate AND and NOT logic operations, achieving record extinction ratio (~20 dB), extremely low control fluence (~0.2 nJ/cm2) and sub-picosecond response time (~500 fs). Our results expand the frontier of dynamic topology and establish a novel pathway towards robust, ultrafast, and reconfigurable on-chip polaritonic logic circuits.

physics.optics↗
arXiv2026

Faster Parameterized Vertex Multicut

In the {\sc Vertex Multicut} problem the input consists of a graph $G$, integer $k$, and a set $\mathbf{T} = \{(s_1, t_1), \ldots, (s_p, t_p)\}$ of pairs of vertices of $G$. The task is to find a set $X$ of at most $k$ vertices such that, for every $(s_i, t_i) \in \mathbf{T}$, there is no path from $s_i$ to $t_i$ in $G - X$. Marx and Razgon [STOC 2011 and SICOMP 2014] and Bousquet, Daligault, and Thomassé [STOC 2011 and SICOMP 2018] independently and simultaneously gave the first algorithms for {\sc Vertex Multicut} with running time $f(k)n^{O(1)}$. The running time of their algorithms is $2^{O(k^3)}n^{O(1)}$ and $2^{O(k^{O(1)})}n^{O(1)}$, respectively. As part of their result, Marx and Razgon introduce the {\em shadow removal} technique, which was subsequently applied in algorithms for several parameterized cut and separation problems. The shadow removal step is the only step of the algorithm of Marx and Razgon which requires $2^{O(k^3)}n^{O(1)}$ time. Chitnis et al. [TALG 2015] gave an improved version of the shadow removal step, which, among other results, led to a $k^{O(k^2)}n^{O(1)}$ time algorithm for {\sc Vertex Multicut}. We give a faster algorithm for the {\sc Vertex Multicut} problem with running time $k^{O(k)}n^{O(1)}$. Our main technical contribution is a refined shadow removal step for vertex separation problems that only introduces an overhead of $k^{O(k)}\log n$ time. The new shadow removal step implies a $k^{O(k^2)}n^{O(1)}$ time algorithm for {\sc Directed Subset Feedback Vertex Set} and a $k^{O(k)}n^{O(1)}$ time algorithm for {\sc Directed Multiway Cut}, improving over the previously best known algorithms of Chitnis et al. [TALG 2015].

cs.DS↗