arXiv subjects

Ziming Liu

Publications and source records attributed to Ziming Liu.

At least 19 recordsLinked to original sources

arXiv2026

A Very Big Video Reasoning Suite

Rapid progress in video models has largely focused on visual quality, leaving their reasoning capabilities underexplored. Video reasoning grounds intelligence in spatiotemporally consistent visual environments that go beyond what text can naturally capture, enabling intuitive reasoning over spatiotemporal structure such as continuity, interaction, and causality. However, systematically studying video reasoning and its scaling behavior is hindered by the lack of large-scale training data. To address this gap, we introduce the Very Big Video Reasoning (VBVR) Dataset, an unprecedentedly large-scale resource spanning 200 curated reasoning tasks following a principled taxonomy and over one million video clips, approximately three orders of magnitude larger than existing datasets. We further present VBVR-Bench, a verifiable evaluation framework that moves beyond model-based judging by incorporating rule-based, human-aligned scorers, enabling reproducible and interpretable diagnosis of video reasoning capabilities. Leveraging the VBVR suite, we conduct one of the first large-scale scaling studies of video reasoning and observe early signs of emergent generalization to unseen reasoning tasks. Together, VBVR lays a foundation for the next stage of research in generalizable video reasoning. The data, benchmark toolkit, and models are publicly available at https://video-reason.com/?v=vbvr .

cs.CV↗

arXiv2026

Beyond Quadratic Loss: The Stability Phase Diagram of Adam

Loss spikes are recurrent instabilities in neural-network training and can arise from multiple mechanisms. For Adam in particular, macroscopic loss spikes have been linked to optimizer dynamics, yet how its two momentum timescales govern them remains unclear. We investigate this dependence by mapping training dynamics across the $(β_1,β_2)$ plane. Across a range of model--task settings, an approximately linear boundary, $1-β_2=C(1-β_1)$, separates spiky from non-spiky dynamics, whereas a one-dimensional quadratic loss produces approximately cubic slope. A one-dimensional superquadratic loss $L(x)\propto|x|^n$ recovers the near-linear scaling and links the boundary coefficient to the effective loss exponent $n$. We further show that confident cross-entropy losses develop a core--wall landscape comprising a narrow quadratic core followed by a steep wall, which produces effective superquadratic behavior at the scale of an optimizer update. Together, these results connect Adam loss spikes to both the mismatch between momentum timescales and finite-scale superquadratic loss geometry beyond the Hessian.

cs.LG↗

arXiv2026

A Chosen Future Can Still Be Rewritten: Causal Writability in Video Models

When a video model generates physically incorrect motion, did it fail to learn the correct motion, or did it learn it but fail to use it? We show the latter: the correct motion remains available inside the model and can still be made to control the generated video. We train on videos where red masses oscillate slowly and blue masses oscillate quickly, then test a red mass with fast observed motion. Even when the model generates slow motion in this conflicting case, a low-dimensional edit predicted from simple physical variables restores the correct fast motion. We call this ability causal writability. At fixed strength, we find a sharp depth boundary: the same edit changes the video before the boundary but not after it. This closure marks commitment for that write. The motion signal nevertheless remains, and a stronger downstream write can restore physical motion, while excessive gain overshoots. Early causal writability predicts which errors training later corrects: those errors are writable at more network depths than errors that persist. We reproduce both causal writability and its sharp closure in a pretrained 1.3B video model, supporting generality across model scale and training regime.

cs.LG↗

arXiv2026

Clearing the Fog: Towards Installing and Refining Proactive Exploration Capabilities in LLM Agents

We study proactive exploration in LLM agents, i.e., the ability to explore an environment to acquire information that improves future decision-making. In this regard, we first identify two fundamental bottlenecks that hinder this capability and then propose \ours, a novel method designed to instill and refine proactive exploration. Specifically, \ours\ consists of two components: (1) Exploratory Data Construction, which synthesizes exploration-rich trajectories to mitigate the hindsight bias of standard demonstrations; and (2) RL Optimization with Contrastive Signal Guidance, which leverages contrastive trajectory pairs to distinguish productive exploration from redundant wandering. Extensive experiments demonstrate the effectiveness of \ours\ and provide insights into the characteristics of proactive exploration. Our code is available at: https://github.com/GuanZhizhao/SAFARI.

cs.AI↗

arXiv2026

Variational Continuation for Double Pendulum Periodic Orbits

We present a Hessian-based approach to numerically continue periodic orbits in dynamical systems. A loop (periodic orbit candidate) is parametrized as a Fourier series; a loss function is defined based on the deviation of the loop from the physical differential equations. Unlike previous work relying on hand-derived Jacobians, our method automates the process by leveraging automatic differentiation, a common machine learning technique. The continuation direction can be determined by the flat directions of the loss landscapes (directions with zero eigenvalues), making the search of periodic orbits efficient and guided. Our method is integrator-free, precisely initializes oscillations around unstable fixed points, and efficiently detects orbit family intersections and subharmonic bifurcations. As a demonstration, we present full continuations of periodic double pendulum oscillations from fixed points, showing bifurcations along orbit families and categorizing branches of periodic orbits. In particular, we find periodic orbits where both pendulum masses are never simultaneously at rest, which to our knowledge has been missing in the literature.

cs.LG↗

arXiv2026

Beyond Local Surprise: Grounded Dialogue as Selective Belief Revision under Referential Uncertainty

When a speaker refers to a scene that the listener cannot directly see, the listener must decide whether to preserve its current understanding or revise it as new utterances arrive. Many language systems treat local mismatch as a cue for updating: divergence from the current understanding encourages adjustment. Yet conversational understanding may be more conservative, interpreting mismatching evidence relative to prior understanding rather than immediately revising it. We introduce a controlled, data-driven framework for turn-by-turn preserve/revise decisions in dialogue, where competing revision policies are learned under otherwise identical conditions. We compare four theory-driven revision strategies, each reflecting a different assumption about when listeners should preserve or revise. Two findings stand out. First, a mismatch-driven policy that updates solely based on local divergence reacts strongly to mismatch but destabilizes grounding and degrades retrieval. Second, an uncertainty-sensitive policy extends mismatch-based updating with accumulated evidence, preserving coherent understanding while maintaining strong retrieval performance. Surprisingly, coherent understanding emerges from a counterintuitive pattern: local mismatch promotes preservation, whereas accumulated uncertainty promotes revision, suggesting that listeners maintain prior understanding despite local mismatch and revise only when uncertainty sufficiently accumulates. This pattern is consistent with conceptual pact theory.

cs.CL↗

arXiv2026

Universal One-third Time Scaling in Learning Peaked Distributions

Training large language models (LLMs) is computationally expensive, partly because the loss exhibits slow power-law convergence whose origin remains debatable. Through systematic analysis of toy models and empirical evaluation of LLMs, we show that this behavior can arise intrinsically from the use of softmax and cross-entropy. When learning peaked probability distributions, e.g., next-token distributions, these components generically yield power-law vanishing losses and gradients, regardless of many microscopic details, creating a fundamental optimization bottleneck. This ultimately leads to power-law time scaling of the loss with a universal exponent of $1/3$. Our results provide a mechanistic explanation for observed neural scaling and suggest new directions for improving LLM training efficiency.

cs.LG↗

arXiv2026

A Robust Two-Stage Protocol for STAR-RIS-Aided ISAC Networks: Joint Beamforming and Mode Optimization

This paper investigates the robust design of integrated sensing and communication (ISAC) systems assisted by simultaneously transmitting and reflecting reconfigurable intelligent surfaces (STAR-RISs), acting as programmable metasurfaces enabling concurrent sensing and communication over the full space. To exploit the dual transmission-reflection capability of STAR-RISs, we propose a two-stage ISAC protocol: a preparation phase jointly performs direction-of-arrival (DoA) estimation for outdoor users and downlink communication to all users, while a subsequent communication phase leverages the acquired angular information to enhance downlink transmission. To capture sensing uncertainty and imperfect channel knowledge, the DoAs of outdoor users are modeled as Gaussian random variables, and the non-line-of-sight (NLoS) channel components of outdoor links are characterized through their spatial covariance statistics, enabling a robust design that incorporates average communication performance into the optimization. We formulate a performance-balanced optimization problem that maximizes the communication sum-rate while guaranteeing sensing accuracy, jointly determining the beamforming vectors, the STAR-RIS transmission and reflection coefficients in both stages, and the metasurface partition between energy-splitting and transmit-only modes. To address the resulting non-convex mixed discrete-continuous problem, we develop a tailored alternating optimization framework with proven monotonic convergence. Numerical results demonstrate approximately 15% throughput gain over the most competitive benchmark neglecting NLoS statistical characterization, with robustness maintained under DoA estimation errors and imperfect NLoS channel knowledge.

eess.SP↗

arXiv2026

Region-Adaptive Sampling for Diffusion Transformers

Diffusion models (DMs) have become the leading choice for generative tasks across diverse domains. However, their reliance on multiple sequential forward passes significantly limits real-time performance. Previous acceleration methods have primarily focused on reducing the number of sampling steps or reusing intermediate results, failing to leverage variations across spatial regions within the image due to the constraints of convolutional U-Net structures. By harnessing the flexibility of Diffusion Transformers (DiTs) in handling variable number of tokens, we introduce RAS, a novel, training-free sampling strategy that dynamically assigns different sampling ratios to regions within an image based on the focus of the DiT model. Our key observation is that during each sampling step, the model concentrates on semantically meaningful regions, and these areas of focus exhibit strong continuity across consecutive steps. Leveraging this insight, RAS updates only the regions currently in focus, while other regions are updated using cached noise from the previous step. The model's focus is determined based on the output from the preceding step, capitalizing on the temporal consistency we observed. We evaluate RAS on Stable Diffusion 3 and Lumina-Next-T2I, achieving speedups up to 2.36x and 2.51x, respectively, with minimal degradation in generation quality. Additionally, a user study reveals that RAS delivers comparable qualities under human evaluation while achieving a 1.6x speedup. Our approach makes a significant step towards more efficient diffusion transformers, enhancing their potential for real-time applications.

cs.CV↗

arXiv2026

Gridless Full-Space DOA Estimation for STAR-RIS-Assisted Wireless Systems

Simultaneously transmitting and reflecting reconfigurable intelligent surfaces (STAR-RIS) enable full-space ($0^\circ$--$360^\circ$) signal coverage, making them a compelling platform for integrated sensing and communication in next-generation wireless networks. In this paper, we investigate gridless direction-of-arrival (DOA) estimation across the full spatial domain in STAR-RIS-assisted systems operating with a single RF sensing chain. We show that the coupled reflection-transmission mechanism of STAR-RIS induces a multichannel finite-rate-of-innovation (FRI) structure in the received signal, which enables casting DOA estimation as a structured low-rank recovery problem without angular grid discretization. Building on this observation, we develop a proximal gradient descent algorithm with alternating projections onto a block-Hankel matrix set, enabling robust angle retrieval from limited measurements. Two practically relevant STAR-RIS configurations are addressed: element-wise uniform and nonuniform energy-splitting designs, each handled through a dedicated lifting strategy that preserves the underlying algebraic structure. A Ziv-Zakai bound is derived for the coupled full-space sensing model as a performance benchmark across the full SNR range. Numerical results show that the proposed methods consistently outperform grid-based baselines, achieving sub-degree accuracy within $\pm 60^\circ$ of boresight at comparable or lower computational cost.

eess.SP↗

arXiv2026

Inverse Depth Scaling From Most Layers Being Similar

Neural scaling laws relate loss to model size in large language models (LLMs), yet depth and width may contribute to performance differently, requiring more detailed studies. Here, we quantify how depth affects loss via analysis of LLMs and toy residual networks. We find loss scales inversely proportional to depth in LLMs, probably due to functionally similar layers reducing error through ensemble averaging rather than compositional learning or discretizing smooth dynamics. This regime is inefficient yet robust and may arise from the architectural bias of residual networks and target functions incompatible with smooth dynamics. The findings suggest that improving LLM efficiency may require architectural innovations to encourage compositional use of depth.

cs.LG↗

arXiv2026

HiFAST: An HI data calibration and imaging pipeline for the FAST IV: The stray-radiation correction

Stray radiation is a considerable challenge for radio telescopes, requiring careful assessment due to its effects. This is crucial when the strong background flux from side lobes significantly affects the total flux, especially for extended sources. In this study, we introduced the beam pattern of the L-band receiver on the Five-hundred-meter Aperture Spherical Telescope (FAST), covering various frequencies based on recent observations. We discovered that the main beam efficiency of all beams exceeds 90\% throughout the L band frequencies, with efficiency decreasing slowly as frequency increases. Subsequently, we developed a module to mitigate stray radiation effects, incorporating it into FAST's standard \HI data reduction process, referred to as \texttt{HiFAST}. Our analysis shows that side lobe flux's influence, particularly for extended sources with significant surface density gradients, necessitates detailed evaluation. Corrections for the extended M33 galaxy can reach up to 20\%. Moreover, the pattern data presented here is vital for studying HI intensity maps at high redshift. The module, along with HiFAST and beam pattern data across 15 frequency bins, can be accessed at \textrm{https://hifast.readthedocs.io}. The datasets of beam pattern presented in this paper, are openly available at \textrm{https://doi.org/10.57760/sciencedb.j00113.00266} (https://www.scidb.cn/s/bqQRNv).

astro-ph.IM↗

arXiv2026

The FAST Hundred-Deg$^2$ HI Deep (HD$^2$) Survey: Early Results from the Pilot Survey

The Hundred-deg$^2$ HI Deep (HD$^2$) survey carried out with the Five-hundred-meter Aperture Spherical Telescope (FAST) is planned to map a contiguous region within the DESI DR1 footprint, achieving an effective integration time of 20 minutes for each pointing and a uniform detection sensitivity of 0.28 mJy beam$^{-1}$ at 4.8 km s$^{-1}$ resolution. We present early results from the pilot HD$^2$ survey: a 10 deg$^2$ field overlapping with HSC-SSP and the DESI EDR SV3, observed with an integration time of 7.3 minutes per beam and the rms of 0.45 mJy beam$^{-1}$ at 4.8 km s$^{-1}$ resolution. We identify 339 HI sources at $z<0.09$, corresponding to $\sim$34 detections per deg$^2$, nearly six times higher than the detection rate of the wide-field surveys. Optical counterparts are primarily identified using DESI redshifts, yielding a matching rate and correctness exceeding 90% for galaxies with $r<19.5$ mag, a substantial improvement over SDSS. Under the constraint of $r < 17.8$ mag and $0.01 < z < 0.05$, nearly 50% of galaxies in the DESI BGS samples have HI detections in this pilot survey. The optical properties of these HI-detected galaxies span nearly the entire parameter range of the DESI sample. The gas fraction scaling relations versus stellar mass, stellar mass surface density, NUV-r, and specific star formation rate are consistent with previous surveys, e.g., ALFALFA, DINGO, and xGASS. These results justify the feasibility of the full HD$^2$ survey, which will build a high-completeness HI census over a contiguous area to probe the cold gas scaling relations of galaxies over different scales.

astro-ph.GA↗

arXiv2026

PlexRL: Cluster-Level Orchestration of Serviceized LLM Execution for RLVR

Reinforcement learning with verifiable rewards (RLVR) has recently unlocked strong reasoning capabilities in large language models (LLMs), triggering rapid exploration of new algorithms and data. However, RLVR training is notoriously inefficient: long-tailed rollouts, tool-induced stalls, and asymmetric resource requirements between rollout and training introduce substantial idle time that cannot be eliminated by job-local optimizations such as synchronous pipelining, asynchronous rollout, or colocated execution. We argue that this inefficiency is structural. While idle gaps are unavoidable within individual RLVR jobs, they are largely anti-correlated across jobs and therefore exploitable at the cluster level. Leveraging this observation, we present PlexRL, a cluster-level runtime for multiplexing unified LLM services across RLVR jobs. By centrally managing model placement, state transitions, and function-level scheduling under strict affinity constraints, PlexRL time-slices LLM execution across jobs to fill otherwise idle periods without expensive model migration. Our implementation and evaluations demonstrate that PlexRL significantly improves effective cluster capacity and reduces user GPU hour cost by maximum 37.58% while preserving algorithmic flexibility and introducing minimal per-job overhead.

cs.DC↗

arXiv2026

The strong hull property for affine irreducible Coxeter groups of rank 3

A conjecture proposed by Gaetz and Gao asserts that the Cayley graph of any Coxeter group satisfies the strong hull property. In this paper, we prove this conjecture for all affine irreducible Coxeter groups of rank 3. Our approach exploits the geometry of Coxeter complexes to reduce the analysis of convex hulls to finitely many manageable configurations.

math.CO↗

arXiv2026

Are Your Reasoning Models Reasoning or Guessing? A Mechanistic Analysis of Hierarchical Reasoning Models

Hierarchical reasoning model (HRM) achieves extraordinary performance on various reasoning tasks, significantly outperforming large language model-based reasoners. To understand the strengths and potential failure modes of HRM, we conduct a mechanistic study on its reasoning patterns and find three surprising facts: (a) Failure of extremely simple puzzles, e.g., HRM can fail on a puzzle with only one unknown cell. We attribute this failure to the violation of the fixed point property, a fundamental assumption of HRM. (b) "Grokking" dynamics in reasoning steps, i.e., the answer is not improved uniformly, but instead there is a critical reasoning step that suddenly makes the answer correct; (c) Existence of multiple fixed points. HRM "guesses" the first fixed point, which could be incorrect, and gets trapped there for a while or forever. All facts imply that HRM appears to be "guessing" instead of "reasoning". Leveraging this "guessing" picture, we propose three strategies to scale HRM's guesses: data augmentation (scaling the quality of guesses), input perturbation (scaling the number of guesses by leveraging inference randomness), and model bootstrapping (scaling the number of guesses by leveraging training randomness). On the practical side, by combining all methods, we develop Augmented HRM, boosting accuracy on Sudoku-Extreme from 54.5% to 96.9%. On the scientific side, our analysis provides new insights into how reasoning models "reason".

cs.AI↗

arXiv2026

KANO: Kolmogorov-Arnold Neural Operator

We introduce Kolmogorov--Arnold Neural Operator (KANO), a dual-domain neural operator jointly parameterized by both spectral and spatial bases with intrinsic symbolic interpretability. We theoretically demonstrate that KANO overcomes the pure-spectral bottleneck of Fourier Neural Operator (FNO): KANO remains expressive over generic position-dependent dynamics (variable coefficient PDEs) for any physical input, whereas FNO stays practical only for spectrally sparse operators and strictly imposes a fast-decaying input Fourier tail. We verify our claims empirically on position-dependent differential operators, for which KANO robustly generalizes but FNO fails to. In the quantum Hamiltonian learning benchmark, KANO reconstructs ground-truth Hamiltonians in closed-form symbolic representations accurate to the fourth decimal place in coefficients and attains $\approx 6\times10^{-6}$ state infidelity from projective measurement data, substantially outperforming that of the FNO trained with ideal full wave function data, $\approx 1.5\times10^{-2}$, by orders of magnitude.

cs.LG↗

arXiv2026

DM0: An Embodied-Native Vision-Language-Action Model towards Physical AI

Moving beyond the traditional paradigm of adapting internet-pretrained models to physical tasks, we present DM0, an Embodied-Native Vision-Language-Action (VLA) framework designed for Physical AI. Unlike approaches that treat physical grounding as a fine-tuning afterthought, DM0 unifies embodied manipulation and navigation by learning from heterogeneous data sources from the onset. Our methodology follows a comprehensive three-stage pipeline: Pretraining, Mid-Training, and Post-Training. First, we conduct large-scale unified pretraining on the Vision-Language Model (VLM) using diverse corpora--seamlessly integrating web text, autonomous driving scenarios, and embodied interaction logs-to jointly acquire semantic knowledge and physical priors. Subsequently, we build a flow-matching action expert atop the VLM. To reconcile high-level reasoning with low-level control, DM0 employs a hybrid training strategy: for embodied data, gradients from the action expert are not backpropagated to the VLM to preserve generalized representations, while the VLM remains trainable on non-embodied data. Furthermore, we introduce an Embodied Spatial Scaffolding strategy to construct spatial Chain-of-Thought (CoT) reasoning, effectively constraining the action solution space. Experiments on the RoboChallenge benchmark demonstrate that DM0 achieves state-of-the-art performance in both Specialist and Generalist settings on Table30.

cs.RO↗