arXiv subjects

Bohan Wang

Publications and source records attributed to Bohan Wang.

At least 19 recordsLinked to original sources

arXiv2026

Trajectory Forcing: Exploiting Diffusion Trajectories for Autoregressive Long Video Generation

Autoregressive (AR) modeling is arguably the only way to extend a fixed-length video diffusion model to longer or even streaming videos. However, existing AR diffusion methods use only the previously predicted chunk as context for the next, discarding the intermediate predictions that constitute the chunk's generation history. In contrast, conventional AR models such as LLMs condition on the entire history of preceding tokens. To this end, we propose Trajectory Forcing (TraF), an AR diffusion method that exploits the denoising trajectories of previous video chunks as context. TraF operates under two training regimes: (1) with ground-truth videos, where pseudo-trajectories are constructed from Gaussian interpolations and re-predicted by the model, and (2) with a teacher model, where real trajectories are obtained from the student's own denoising rollout. The first regime naturally provides a strong initialization for the second. On the 5-second VBench benchmark, TraF achieves a Total score of 85.15, outperforming Self Forcing (+0.84) and Causal Forcing (+1.11), with Quality and Semantic improving simultaneously. On 30-second generation---6$\times$ the training horizon---TraF achieves the highest overall quality among all compared methods.

cs.CV↗

arXiv2026

From Single-Actor Environments to Multi-Actor Event Worlds: Can LLMs Be Event World Models?

World models help agents assess the consequences of their actions before acting. Many real systems, however, involve multiple actors who make their own decisions, can evolve even when the agent takes no action, and are only partially observed through event logs. We formalize these systems as event worlds and ask whether large language models can predict future observations from the log and the agent's proposed actions. We build complementary benchmarks from human Diplomacy games and real Polymarket trades. Our hierarchical framework compares prediction with and without other actors' actions, then tests multi-step rollouts, predictions for actions never taken, and usefulness to agents. Fine-tuning and more data improve prediction when all actors' recorded actions are given. With only the agent's action, models struggle when other actors strongly affect the outcome, and larger models and longer histories help little. This gap also limits multi-step prediction and agent decisions, with models overestimating plan success. Yet imagined experience and initialization from world models can still improve policy learning. Anticipating other actors' actions remains a central challenge for event world modeling.

cs.AI↗

arXiv2026

Attainment of the sharp stability constant of Beckner conformal log-Sobolev inequality

The existence of minimizer for stability of the fractional Sobolev inequality and the stability of Beckner conformal Log-Sobolev inequality (the endpoint case of fractional Sobolev inequality) have been established by König and Chen-Lu-Tang. The existence of minimizer for stability of the Log-Sobolev inequality is unsolved. In this paper, we solve this problem. For every $n\geq2$, we prove that the sharp $L^2$-stability constant of Beckner conformal log-Sobolev inequality on $\mathbb S^n$ is attained and is strictly less than $\frac{2\log2}{n}$. Moreover, the normalized stability constant of the fractional Sobolev inequality converges to the stability constant of the log-Sobolev inequality as $s\to0^+$. After Hersch normalization, every sequence of $L^{p_{s_k}}$-normalized minimizers for stability of the fractional Sobolev inequality, with $s_k\to0^+$, has a subsequence converging strongly in $L^2(\mathbb S^n)$ to an $L^2$-normalized minimizer of stability of the log-Sobolev inequality.

math.AP↗

arXiv2026

Fourier-Latent Diffusion for Constrained Generation of Triply Periodic Minimal Surfaces

We present a generative framework for the constrained design of $D_{2h}$-symmetric triply periodic minimal surfaces (TPMS) with low residual mean curvature. Existing TPMS design pipelines either explore a limited set of canonical analytical families or rely on computationally expensive numerical procedures, making it difficult to efficiently generate diverse candidates under geometric and/or mechanical requirements. Our approach learns a distribution of solver-generated TPMS in a compact Fourier space where periodicity and $D_{2h}$ symmetry are guaranteed by construction. This representation eliminates the need for a learned geometric decoder and enables the training of an effective diffusion model that generates low-mean-curvature candidates conditioned on sparse geometric constraints, selected homogenized elastic properties, or their combination. A subsequent coefficient-space refinement further reduces the residual mean curvature. Experiments show that our approach outperforms trigonometric, SDF-based, and standard Fourier-space baselines and enables controllable, high-quality generation under geometric and low-dimensional mechanical constraints. Overall, the proposed framework provides a compact and efficient design space for generating near-minimal periodic structures under user-specified requirements.

cs.GR↗

arXiv2026

HiThink Turn: An Intent-Aware Turn-Taking Control Module for Full-Duplex Dialogue

Full-duplex dialogue requires timely yet selective interruption handling, which end-of-turn prediction alone cannot achieve: complete utterances may need no response, while unfinished requests may warrant interruption. To address this challenge, we propose HiThink Turn, an intent-aware streaming turn-state predictor that separates response intent from semantic completeness and conditions decisions on system playback state. A key contribution is minimal intent-sufficient prefix supervision, constructed through LLM judgments and speech alignment, while training on audio truncated at chunk boundaries improves robustness to partial speech. These components support streaming inference with 240-ms audio chunks, enabling low-latency, accurate full-duplex turn control. Experiments show that HiThink Turn leads the compared methods in Easy Turn macro accuracy, Full-Duplex-Bench average interaction rate score (0.933), and non-target-speech average playback resume rate (0.735). Additionally, intent-prefix triggering raises interruption success from 89\% to 98\% and reduces mean stop latency by 60.9\%.

cs.HC↗

arXiv2026

HLE-Verified: A Systematic Verification and Structured Revision of Humanity's Last Exam

Humanity's Last Exam (HLE) has become a widely used benchmark for evaluating frontier large language models on challenging, multi-domain questions. However, community-led analyses have raised concerns that HLE contains a non-trivial number of noisy items, which can bias evaluation results and distort cross-model comparisons. To address this challenge, we introduce HLE-Verified, a verified and revised version of HLE with a transparent verification protocol and fine-grained error taxonomy. Our construction follows a two-stage validation-and-repair workflow resulting in a certified benchmark. In Stage I, each item undergoes binary validation of the problem and final answer through domain-expert review and model-based cross-checks, yielding 668 verified items. In Stage II, flawed but fixable items are revised under strict constraints preserving the original evaluation intent, through dual independent expert repairs, model-assisted auditing, and final adjudication, resulting in 1,143 revised-and-certified items. The remaining 689 items are released as a documented uncertain set with explicit uncertainty sources and expertise tags for future refinement. We evaluate eight state-of-the-art language models on HLE and HLE-Verified, observing an average absolute accuracy gain of 7--10 percentage points on HLE-Verified. The improvement is particularly pronounced on items where the original problem statement and/or reference answer is erroneous, with gains of 30--40 percentage points. Our analyses further reveal a strong association between model confidence and the presence of errors in the problem statement or reference answer, supporting the effectiveness of our revisions. Overall, HLE-Verified improves HLE-style evaluations by reducing annotation noise and enabling more faithful measurement of model capabilities. Data is available at: https://huggingface.co/datasets/skylenage/HLE-Verified

cs.CL↗

arXiv2026

Your Data Manifold is Secretly a Reward Model: Shell-LCC for Text-to-Video Generation

Recent text-to-video (T2V) diffusion models rely heavily on auxiliary reward signals (e.g., via reward models or DPO) to align generated content with human aesthetics and improve realism. These signals, however, incur substantial computational overhead, require costly human annotations, and often yield limited improvement in fine-grained local details. In this paper, we argue that your data manifold is secretly a reward model. By explicitly modeling the manifold structure of high-quality Supervised Fine-Tuning (SFT) data and encouraging video latents to lie on this manifold, we derive dense, differentiable, and nearly cost-free reward signals that significantly improve video quality, particularly in mitigating low-level distortions. Our modeling builds upon Local Coordinate Coding (LCC), which captures the `skeleton' of the manifold. However, directly applying LCC suffers from mean regression, pulling latents toward the geometric mean and losing high-frequency details. We therefore extend it to Shell Local Coordinate Coding (Shell-LCC), which models the manifold `surface' as an isotropic shell to align with the true high-density region. Experiments demonstrate that our approach improves realism, enhances high-frequency details, reduces over-smoothing artifacts, and alleviates motion blur.

cs.CV↗

arXiv2026

Shellular Metamaterial Design via Compact Electric Potential Parametrization

We present a compact yet highly expressive design space for shellular metamaterials that support both interactive exploration and inverse design. With only a few dozen charges, our representation generates a wide family of periodic shells, spanning from simple planar configurations to complex TPMS-like morphologies. To enable rapid evaluation, we introduce an efficient GPU-based homogenization pipeline that computes the effective elastic tensor of a candidate design in near real time (~ 0.4), making interactive shellular design practical. Across a large set of synthesized structures, our design space exhibits geometric diversity and spans a broad spectrum of mechanical responses, covering a wide range of effective material properties. This fast evaluation further enables inverse design for target macroscopic properties. In the low-solid-volume regime, the resulting shellular structures achieve performance competitive with state-of-the-art shell-based metamaterials in multiple material properties. Finally, we validate manufacturability by fabricating tiled prototypes via additive manufacturing, demonstrating the potential of our approach for real-world engineering applications.

cs.GR↗

arXiv2026

SpecLoR: Spectral Lookahead Rectification for Motion-Coherent Text-to-Video Generation

Flow Matching has enabled robust text-to-video generation via latent ODE sampling. However, velocity approximation and numerical discretization errors inevitably accumulate, causing sampling trajectories to drift. Consequently, generated videos often suffer from severe spatiotemporal inconsistencies. Nevertheless, directly correcting these drifted, noisy latents is challenging: (i) timestep-dependent noise obscures reliable structural cues; (ii) spatial interventions risk disrupting intricate local geometry while incurring heavy computational costs. To address this, we propose Spectral Lookahead Rectification (SpecLoR), a plug-and-play inference method that bypasses noise via lookahead prediction, and circumvents spatiotemporal entanglement by shifting corrections to the frequency domain, where universal statistical priors of natural videos are readily available. First, during early sampling stages, SpecLoR looks ahead to estimate the clean latent $z_{t,0}$ and computes its 3D spatiotemporal spectrum. Next, SpecLoR rectifies the amplitude spectrum to match the prior, leaving the phase intact. Finally, the corrected state is re-noised to resume ODE integration. Experiments on Wan2.2 demonstrate that SpecLoR significantly reduces physical artifacts and enhances motion coherence across multiple benchmarks with minimal computational overhead (4 additional NFEs).

cs.CV↗

arXiv2026

Exposing Vulnerabilities in Explanation for Time Series Classifiers via Dual-Target Attacks

Interpretable time series deep learning systems are often assessed by checking temporal consistency on explanations, implicitly treating this as evidence of robustness. We show that this assumption can fail: Predictions and explanations can be adversarially decoupled, enabling targeted misclassification while the explanation remains plausible and consistent with a chosen reference rationale. We propose TSEF (Time Series Explanation Fooler), a dual-target attack that jointly manipulates the classifier and explainer outputs. In contrast to single-objective misclassification attacks that disrupt explanation and spread attribution mass broadly, TSEF achieves targeted prediction changes while keeping explanations consistent with the reference. Across multiple datasets and explainer backbones, our results consistently reveal that explanation stability is a misleading proxy for decision robustness and motivate coupling-aware robustness evaluations for trustworthy time series tasks.

cs.LG↗

arXiv2026

Extra-Merge: Tracing the Rank-1 Subspace of Model Merging in Language Model Pre-Training

Model merging has emerged as a lightweight paradigm for enhancing Large Language Models (LLMs), yet its underlying mechanisms remain poorly understood. In this work, we analyze late-stage pre-training trajectories and uncover a \textbf{Rank-1 Subspace} phenomenon: while raw optimization steps oscillate violently, consecutive \emph{merged} checkpoints collapse onto a stable, approximately one-dimensional linear manifold. We theoretically ground this observation in a \emph{river-valley} landscape analysis: averaging acts as a geometric low-pass filter that dampens high-curvature noise to reveal the optimal descent direction. Capitalizing on this insight, we propose \textbf{Extra-Merge}, a training-free strategy that extrapolates along this subspace to minimize loss without additional gradient updates. Extensive experiments across GPT-2 and LLaMA families (124M to 2B) demonstrate that Extra-Merge consistently outperforms standard merging baselines. Notably, it yields consistent zero-shot accuracy gains on Pythia-12B downstream tasks and generalizes effectively to the Muon optimizer \citep{jordan2024muon}.

cs.LG↗

arXiv2026

Generalization Bounds of Stochastic Gradient Descent in Homogeneous Neural Networks

Algorithmic stability is among the most potent techniques in generalization analysis. However, its derivation usually requires a stepsize $η_t = \mathcal{O}(1/t)$ under non-convex training regimes, where $t$ denotes iterations. This rigid decay of the stepsize potentially impedes optimization and may not align with practical scenarios. In this paper, we derive the generalization bounds under the homogeneous neural network regimes, proving that this regime enables slower stepsize decay of order $Ω(1/\sqrt{t})$ under mild assumptions. We further extend the theoretical results from several aspects, e.g., non-Lipschitz regimes. This finding is broadly applicable, as homogeneous neural networks encompass fully-connected and convolutional neural networks with ReLU and LeakyReLU activations.

cs.LG↗

arXiv2026

Pre-training Epidemic Time Series Forecasters with Compartmental Prototypes

Accurate epidemic forecasting is crucial for outbreak preparedness, but existing data-driven models are often brittle. Typically trained on a single pathogen, they struggle with data scarcity during new outbreaks and fail under distribution shifts caused by viral evolution or interventions. However, decades of surveillance data and the design of various compartmental models from diverse diseases offer an untapped source of transferable knowledge. To leverage the collective lessons from history, we propose CAPE, the first open-source pre-trained model for epidemic forecasting. Unlike existing time series foundation models that overlook epidemiological challenges, CAPE models epidemic dynamics as mixtures of latent compartmental population states, termed \textit{compartmental prototypes}. It models a flexible dictionary of compartment prototypes directly from a large collection of simulation data, enabling each outbreak to be expressed as a time-varying mixture that links observed infections to latent population states. To promote robust generalization, CAPE adopts the next-token-prediction paradigm during pre-training with lightweight epidemic-aware regularization that aligns the learned prototypes with epidemiological semantics. On a comprehensive benchmark spanning 17 diseases, CAPE significantly outperforms strong baselines with zero-shot forecasting. This work represents a principled step toward pre-trained epidemic models that are both transferable and epidemiologically grounded. We provide our code in: https://github.com/nuuuh/CAPE.

cs.LG↗

arXiv2026

Quantitative properties of the Hardy-type mean field equation

In this paper, we consider the following Hardy-type mean field equation \[ \left\{ {\begin{array}{*{20}{c}} { - Δu-\frac{1}{(1-|x|^2)^2} u = λe^u}, & {\rm in} \ \ B_1,\\ {\ \ \ \ u = 0,} &\ {\rm on}\ \partial B_1, \end{array}} \right. \] \[\] where $λ>0$ is small and $B_1$ is the standard unit disc of $\mathbb{R}^2$. Applying the moving plane method of hyperbolic space and the accurate expansion of heat kernel on hyperbolic space, we establish the radial symmetry and Brezis-Merle lemma for solutions of Hardy-type mean field equation. Meanwhile, we also derive the quantitative results for solutions of Hardy-type mean field equation, which improves significantly the compactness results for classical mean-field equation obtained by Brezis-Merle and Li-Shafrir. Furthermore, applying the local Pohozaev identity from scaling, blow-up analysis and a contradiction argument, we prove that the solutions are unique when $λ$ is sufficiently close to 0.

math.AP↗

arXiv2025

BSFA: Leveraging the Subspace Dichotomy to Accelerate Neural Network Training

Recent studies \citep{gur2018gradient,song2024does, wen2024understanding} highlight a fundamental dichotomy in deep learning optimization: Although parameter updates along the top eigendirections of the loss Hessian (Dom-space) capture most of the update magnitude, they often contribute minimally to loss reduction. In contrast, updates in the orthogonal component (Bulk-space) have smaller magnitudes but drive most learning progress. In this work, we further advance the understanding of this phenomenon and introduce the \textbf{Bulk-Space-Filtration-Accelerator (BSFA)}, a novel plug-and-play framework. BSFA accelerates training by differentially scaling update components projected onto these distinct subspaces, simultaneously enhancing stability by moderating updates in the dominant subspace and boosting convergence speed by amplifying those in the bulk-space. To ensure BSFA is both practical and scalable for contemporary large models, we introduce two key innovations: an efficient estimator using Principal Component Analysis (PCA) on historical updates for fast subspace estimation, and a block-wise strategy that applies this estimation on a per-parameter-block basis. These designs make BSFA computationally tractable and highly effective. We demonstrate BSFA's acceleration across various tasks, notably achieving approximately 2$\times$ speedup when pre-training LLaMA-72M on WikiText-103 and LLaMA-134M on OpenWebText compared to vanilla AdamW.

cs.LG↗

arXiv2025

Score-based Conditional Out-of-Distribution Augmentation for Graph Covariate Shift

Distribution shifts between training and testing datasets significantly impair the model performance on graph learning. A commonly-taken causal view in graph invariant learning suggests that stable predictive features of graphs are causally associated with labels, whereas varying environmental features lead to distribution shifts. In particular, covariate shifts caused by unseen environments in test graphs underscore the critical need for out-of-distribution (OOD) generalization. Existing graph augmentation methods designed to address the covariate shift often disentangle the stable and environmental features in the input space, and selectively perturb or mixup the environmental features. However, such perturbation-based methods heavily rely on an accurate separation of stable and environmental features, and their exploration ability is confined to existing environmental features in the training distribution. To overcome these limitations, we introduce a novel distributional augmentation approach enabled by a tailored score-based conditional graph generation strategies to explore and synthesize unseen environments while preserving the validity and stable features of overall graph patterns. Our comprehensive empirical evaluations demonstrate the enhanced effectiveness of our method in improving graph OOD generalization.

cs.LG↗

arXiv2025

Continuous-time Riemannian SGD and SVRG Flows on Wasserstein Probabilistic Space

Recently, optimization on the Riemannian manifold have provided valuable insights to the optimization community. In this regard, extending these methods to to the Wasserstein space is of particular interest, since optimization on Wasserstein space is closely connected to practical sampling processes. Generally, the standard (continuous) optimization method on Wasserstein space is Riemannian gradient flow (i.e., Langevin dynamics when minimizing KL divergence). In this paper, we aim to enrich the family of continuous optimization methods in the Wasserstein space, by extending the gradient flow on it into the stochastic gradient descent (SGD) flow and stochastic variance reduction gradient (SVRG) flow. By leveraging the property of Wasserstein space, we construct stochastic differential equations (SDEs) to approximate the corresponding discrete Euclidean dynamics of the desired Riemannian stochastic methods. Then, we obtain the flows in Wasserstein space by Fokker-Planck equation. Finally, we establish convergence rates of the proposed stochastic flows, which align with those known in the Euclidean setting.

cs.LG↗

arXiv2025

Quantization property of n-Laplacian mean field equation and sharp Moser-Onofri inequality

In this paper, we are concerned with the following $n$-Laplacian mean field equation \[ \left\{ {\begin{array}{*{20}{c}} { - Δ_n u = λe^u} & {\rm in} \ \ Ω, \\ {\ \ \ \ u = 0} &\ {\rm on}\ \partial Ω, \end{array}} \right. \] \[\] where $Ω$ is a smooth bounded domain of $\mathbb{R}^n \ (n\geq 2)$ and $- Δ_n u =-{\rm div}(|\nabla u|^{n-2}\nabla u)$. We first establish the quantization property of solutions to the above $n$-Laplacian mean field equation. As an application, combining the Pohozaev identity and the capacity estimate, we obtain the sharp constant $C(n)$ of the Moser-Onofri inequality in the $n$-dimensional unit ball $B^n:=B^n(0,1)$, $$\mathop {\inf }\limits_{u \in W_0^{1,n}(B^n)}\frac{1}{ n C_n}\int_{B^n} | \nabla u|^n dx- \ln \int_{B^n} {e^u} dx\geq C(n),$$ which extends the result of Caglioti-Lions-Marchioro-Pulvirenti in \cite{Caglioti} to the case of $n$-dimensional ball. Here $C_n=(\frac{n^2}{n-1})^{n-1} ω_{n-1}$ and $ω_{n-1}$ is the surface measure of $B^n$. For the Moser-Onofri inequality in a general bounded domain of $\mathbb{R}^n$, we apply the technique of $n$-harmonic transplantation to give the optimal concentration level of the Moser-Onofri inequality and obtain the criterion for the existence and non-existence of extremals for the Moser-Onofri inequality.

math.AP↗