arXiv subjects

Yin Li

Publications and source records attributed to Yin Li.

At least 19 recordsLinked to original sources

arXiv2026

Raising the Optical Depth to Reionization with Dark Matter Decay

DESI DR2 baryon acoustic oscillation (BAO) measurements have sharpened a discrepancy with cosmic microwave background (CMB) inferences that can be recast as a preference for $τ_{\rm reio}\simeq0.09$, above the $\simeq0.06$ inferred from large-scale CMB polarization. We test whether electromagnetic energy injection from dark matter (DM) decaying through $χ\to e^+e^-$ can supply this additional optical depth. Adopting an asymmetric, simulation-calibrated Gompertzian model for astrophysical reionization, we derive $τ_{\rm reio}$ self-consistently from the free-electron history including DM decay. We jointly constrain cosmological and reionization parameters and the decay rate using ACT+Planck CMB temperature and polarization anisotropies, CMB lensing, BAO, and quasar damping-wing observations. For $m_χ=1$ GeV, DM decay raises the marginalized mean optical depth from 0.064 to 0.072 under a flat decay-rate prior, without significantly improving the fit. Reweighting the chains with log-uniform priors lowers the mean optical depth, indicating prior dependence. The additional optical depth arises primarily from a broad ionization tail through the cosmic dark ages, rather than an earlier onset of astrophysical reionization. For DM masses from 2 MeV to 7 GeV, the 68% credible intervals remain below $τ_{\rm reio}\simeq0.08$, a benchmark previously found sufficient to bring both the neutrino-mass tension and the exclusion of $Λ$CDM below $2σ$. The DM lifetime constraints complement photon, cosmic-ray, and gas-heating bounds and are competitive with previous CMB limits. DM decay therefore shifts the inferred optical depth upward, but the posterior-favored increase remains insufficient to substantially alleviate the CMB--BAO discrepancy.

astro-ph.CO↗

arXiv2026

STAR: Steering Action Representations during Inference

Vision-language-action models (VLAs) map language instructions and visual observations to robot actions, showing strong capabilities across robotic manipulation tasks. However, in our probing experiment, we find that action representations near the action output do not fully preserve task-relevant visual grounding. Motivated by this late-layer grounding gap, we introduce STAR, for STeering Action Representations, a training-free inference method for pre-trained VLAs. STAR applies an exponential moving average over late-layer residual updates, allowing later action representations to inherit object-aware update directions from earlier layers without model updates or architectural changes. It further uses recent proprioceptive states to adapt the mixing strength, using conservative mixing when execution dynamics change rapidly. Experiments on robot manipulation benchmarks show that STAR improves late-layer object awareness and increases average task success rates by 4.0 percentage points. Our code is available at https://github.com/PKU-VaLuE-Lab/STAR.

cs.RO↗

arXiv2026

DRIFT: A Residual Flow Adapter for Decoding Continuous Outputs in Vision-Language Models

Many modern vision-language models (VLMs) build on autoregressive decoding of discrete tokens. While text-based output interfaces enable scalable pretraining and strong zero-shot generalization across diverse tasks, they are poorly suited for problems that require precise continuous outputs, such as localizing temporal boundaries of events or generating robotic control actions. To address this challenge, we propose DRIFT, a general framework for adapting pretrained VLMs to continuous decoding tasks. DRIFT combines a base predictor, which provides a coarse estimate of the target output, with a generative refinement module based on flow matching that iteratively improves the prediction. This residual formulation transforms the generative modeling problem from learning a global output distribution to modeling a localized residual distribution around a strong prior, substantially simplifying optimization. We evaluate DRIFT on both perception and planning tasks, including visual grounding and robotic control. Across multiple tasks and architectures spanning MLLMs, VLAs, and WAMs, DRIFT consistently outperforms a strong set of regression- and generative-based solutions.

cs.CV↗

arXiv2026

CausalAlign: Causal Alignment of Video Generation Models for Consistent Ego-Motion

Diffusion-based video generation models (VGMs) can produce visually plausible videos, yet their camera ego-motion often drifts over long horizons. Camera ego-motion determines how the viewpoint changes between frames and is essential for maintaining consistent geometry and trajectories. To address this problem, we propose CausalAlign, a causal alignment framework with two components. First, we diagnose how precise ego motion is represented, used, and retained within VGMs using linear probes and causal interventions, revealing a rise-and-fall pattern during denoising. Second, we strengthen ego-motion state in VGM by aligning the representations at the identified layers to a 3D foundation model through delta-token space. This space encodes the changes between consecutive frames, allowing the alignment to focus on ego-motion dynamics rather than static scene content. With Wan2.2 fine-tuned on DL3DV and RealEstate10K, CausalAlign can largely improve motion accuracy, geometric consistency, and temporal smoothness, while preserving visual quality. For example, it improves geometric consistency by more than 20% relative to the base model. Moreover, with the fine-tuned VGM, CausalAlign improves the success rate of world action models (WAMs) by 10 points on mobile manipulation tasks of RoboCasa365 benchmark. Overall, CausalAlign improves ego-motion consistency in diffusion-based VGMs and yields noticeable gains in embodied models.

cs.CV↗

arXiv2026

Memory as Weights: Internalizing Long-Term History for Streaming Videos

Streaming vision-language models (VLMs) must continuously integrate information from the past while remaining responsive to current observations and user queries. Existing memory mechanisms typically compress, retrieve, or cache historical observations and reintroduce them into the input context, forcing historical memory and current observations to compete for limited context. We propose MaLoW, a streaming memory framework that instead internalizes historical context into evolving lightweight parameter updates. MaLoW uses a recurrent HyperNetwork to incrementally consolidate incoming video observations into fixed-size LoRA weights, allowing historical information to be carried forward without revisiting the full video stream. In doing so, MaLoW represents history as weights, while keeping the current observation as input. To support training, we further construct MaLoW-Instruct, a large-scale streaming video dataset containing 381K question-answer pairs over 56K videos. We conduct extensive experiments to demonstrate that MaLoW achieves strong performance on OVO-Bench, StreamingBench, and StreamArena. On the backward split of OVO-Bench, MaLoW outperforms prior memory approaches by 3.9% to 11.8% across different video backbones while incurring only modest latency overhead. Beyond streaming video understanding, MaLoW also generalizes to vision-language-action models, improving the action success rate on Simpler WidowX by 9.3%.

cs.CV↗

arXiv2026

Towards Autopilot Labeling for Short Video Content Governance

The explosive growth of short videos has presented significant challenges to content governance. Traditional manual labeling is inefficient, slow to iterate, and costly. Therefore, we explore and develop a multi-stage autopilot labeling framework aimed at progressively increasing the level of automation in short video content annotation. To move beyond the heavy reliance on human expert labeling at the initial L0 stage, we first introduce a discriminative MLLM labeler at the L1 stage, which is fine-tuned with supervised data to directly predict labels for short videos. Advancing to the L2 stage, we shift to a domain-adaptive MLLM pretrained on short video data and use it for zero-shot labeling, enabling flexible generalization without task-specific supervision. In the L3 stage, we propose an agentic annotation workflow, and equip the agent with several key tools, which empower the agent to handle complex, unfamiliar, or vaguely defined cases, significantly reducing the need for human intervention. We conduct extensive experiments across various content issues, and results demonstrate that each stage of the autopilot labeling framework progressively improves labeling accuracy, and has significantly reduced manual labor costs in real-world applications by up to 85%.

cs.CL↗

arXiv2026

Reconstructing the invisible fraction of semivisible jets in ISR-boosted events via neural network regression

Semi-visible jets (SVJs) provide a characteristic collider signature of strongly interacting dark sectors, in which the key model parameter $r_{\mathrm{inv}}$ controls the fraction of dark hadrons decaying to dark matter candidates. In this work, a regression model is developed to reconstruct $r_{\mathrm{inv}}$ in SVJ events produced in association with an energetic photon. The model uses information from high-level physics objects only, and the training procedure is optimized to ensure applicability. The performance is found to be robust against varying signal parameters and $r_{\mathrm{inv}}$ can be reconstructed at a much higher precision, compared to previously developed analytical method. It offers a new approach to conduct SVJ searches that can potentially unify both $s$-channel and $t$-channel productions, enhancing the sensitivities.

hep-ph↗

arXiv2026

Abundance of affine algebraic structures on stabilized cotangent bundles of surfaces

We show that for any $g\geq2$, there exist uncountably many non-isomorphic smooth affine threefolds that are Stein deformation equivalent to $T^\astΣ_g\times\mathbb{C}$, the product of the cotangent bundle of a genus $g$ oriented surface with the complex plane, answering a question of Ivan Smith. In fact, we prove that our family of smooth affine threefolds is parametrized by a $(6g-7)$-dimensional complex orbifold.

math.AG↗

arXiv2026

HAM-RAG: Hierarchy-Aware Multimodal RAG for Structure-Faithful Interleaved Generation

Existing multimodal RAG methods often flatten structured documents into isolated text and image units, weakening the source organization and local text-image logic needed for faithful evidence selection and placement. We propose HAM-RAG, a Hierarchy-Aware Multimodal RAG framework for structure-faithful interleaved generation. HAM-RAG uses document hierarchy as a grounding signal across retrieval and generation, contextualizing textual and visual evidence and preserving source position and local text-image relations in the prompt. We further introduce HAM-Bench, covering Wukong, Wiki, arXiv, and Recipe across game walkthroughs, web pages, scientific papers, and step-wise recipe documents. Across multiple backbones, HAM-RAG improves the main multimodal average by 17.3% over the strongest non-hierarchical baseline. On Wukong, HAM-RAG improves Img-CBS by 24.2% over the strongest non-hierarchical baseline, demonstrating substantially better local text-image alignment. The main experiments and ablation study together demonstrate that document hierarchy is a key grounding signal for faithful image selection, placement, and local text-image alignment. These findings highlight the value of hierarchy-aware grounding for reliable multimodal assistants that generate answers faithful to the source organization, procedural structure, and local text-image evidence of structured documents, such as technical manuals, maintenance guides, and industrial SOPs. The code is available at https://github.com/MCCodeAI/HAM-RAG.git.

cs.IR↗

arXiv2026

Demonstrating TOFFEE: A Learned System for Synthesizing Data Agent Trajectories at Scale

LLM-powered data agents are playing an increasingly important role in data-driven decision making. However, existing data agents struggle to generalize to unseen data environments and analytical workflows, especially in heterogeneous enterprise settings. This creates a growing need for synthesizing high-quality data agent trajectories that capture complex analytical workflows for given data environments. Such trajectories support two key downstream uses: they can serve as supervised finetuning (SFT) data that adapts data agent models to the target domain, and as in-context learning (ICL) demonstrations to guide general-purpose LLMs in unfamiliar data environments. Thus, we introduce TOFFEE, a system for synthesizing high-quality data agent trajectories from given data environments via Monte Carlo Tree Search (MCTS) with adaptive model selection and cross-task prefix reuse. We show that TOFFEE can effectively generate scalable trajectory data for complex analytical tasks across heterogeneous environments. In this demonstration, we present the system framework of TOFFEE, including its task pool construction, trajectory explorer, and learned cost model. We also introduce the web interface of TOFFEE and its workflow, and demonstrate two end-to-end scenarios: trajectory synthesis for data agent finetuning, and demonstration-augmented data agent reasoning.

cs.AI↗

arXiv2026

402Pilot: An x402 Decision Layer for Autonomous Agent Micropayments

Programmable-payment protocols such as x402 enable per-request micropayments, but they do not determine which payable service an autonomous agent should buy under a finite wallet. We formulate this buyer-side problem as agent-native payment decision-making: contextual provider selection under wallet pressure, chosen-only paid feedback, and changing market conditions. We propose 402Pilot, a protocol-agnostic buyer-side decision layer between autonomous agents and payment execution that implements purchasing policies for selecting among payable providers. We instantiate it with PA-DCT, a payment-aware discounted contextual Thompson-sampling policy that adapts purchasing decisions under wallet pressure while learning from post-payment feedback. To evaluate buyer-side payment policies, we introduce 402Pilot-Bench, a frozen-replay benchmark spanning 823 tasks, five heterogeneous provider pipelines, and three market regimes, each evaluated over 30 paired seeds. PA-DCT achieves the strongest fixed-wallet adaptive trade-off among non-oracle policies: it maintains competitive service quality while spending only 39 to 43 percent of the wallet and reallocates spending as market conditions change. It attains the best non-oracle PA-gap/T under the price shock and the best mean and worst-case ranks across the nine scenario-metric combinations of quality, ROI, and PA-gap/T. Comparisons with learning baselines and component ablations further support the effectiveness and design of the proposed decision policy. These results suggest that programmable payment must be complemented by buyer-side decision-making capable of learning service value and adapting purchasing decisions accordingly.

cs.AI↗

arXiv2026

Look Less, Think Faster: Joint Token-Compute Adaptation for Multimodal LLMs

Multimodal Large Language Models (MLLMs) have recently demonstrated strong performance across vision-language tasks. However, their high inference cost, arising from both the large number of input visual tokens and the heavy computation of the large language model (LLM), remains a key barrier to practical deployment. Recent work attempts to reduce the cost by adaptively optimizing individual dimensions, e.g., pruning redundant visual tokens or skipping LLM layers and heads. Nonetheless, prior approaches typically treat these dimensions independently and overlook a fundamental coupling: the available compute resources must be dynamically allocated across all dimensions based on the input content. To bridge the gap, we propose SmartVL, a unified adaptive inference framework that jointly controls vision token number and model compute capability in response to varying input contents and compute budgets. SmartVL introduces a vision-side token controller that dynamically selects informative visual tokens and an LLM-side compute controller that adaptively adjusts LLM computation. Importantly, these controllers are trained to coordinate with each other so that the overall inference cost satisfies a target budget. To allow this joint scheduling, we connect the controllers using a shared budget encoding and leverage a differentiable latency estimator for end-to-end training. This design enables SmartVL to learn cross-stage allocation strategies that adapt to both input complexity and runtime compute constraints. Experiments across multiple MLLM benchmarks demonstrate that, with joint scheduling, SmartVL consistently outperforms prior adaptive methods and achieves superior accuracy-efficiency Pareto frontiers. Project page: https://www.schaterji.io/publications/2026/jointtokencompute.

cs.CV↗

arXiv2026

Lagrangian capacity and chain level string topology

We derive upper bounds for the Lagrangian capacities of Liouville domains with finite Gutt--Hutchings capacities and show that the Lagrangian capacity of a convex or concave toric domain of arbitrary dimension equals its diagonal. In particular, this completely settles the conjecture of Cieliebak-Mohnke on the Lagrangian capacity of ellipsoids. Our proof is based on an $S^1$-equivariant variant of the techniques of Fukaya and Irie, and does not use holomorphic curves with local tangency constraints, which would inevitably cause transversality issues. Moreover, we show that any extremal Lagrangian torus in an $2n$-dimensional ellipsoid must lie on the boundary. Applications of our results and techniques include new upper bounds on the Lagrangian width for aspherical Lagrangians in Liouville manifolds and the first computations of the Lagrangian capacities for many non-subcritical Weinstein domains in dimensions 4 and 6.

math.SG↗

arXiv2026

NeMo: Needle in a Montage for Video-Language Understanding

Recent advances in video large language models (VideoLLMs) call for new evaluation protocols and benchmarks for video-language understanding. Inspired by the needle in a haystack test widely used by LLMs, we introduce a novel task of Needle in a Montage (NeMo), which is designed to assess the temporal understanding capabilities of advanced VideoLLMs. Specifically, the proposed task focuses on two fundamental abilities critical for temporal understanding, i.e., retrieval-style long-context recall and temporal grounding. To generate video question answering data for our task, we develop a scalable automated data generation pipeline that facilitates high-quality data synthesis. Built upon the proposed pipeline, we present NeMoBench, a video-language benchmark centered on our task. Specifically, our full set of NeMoBench features 31,378 automatically generated question-answer (QA) pairs from 13,486 videos with various durations ranging from seconds to hours. Experiments demonstrate that our pipeline can reliably and automatically generate high-quality evaluation data, enabling NeMoBench to be continuously updated with the latest videos. We evaluate 20 state-of-the-art models on our benchmark, providing extensive results and key insights into their capabilities and limitations. Our project page is available at: https://lavi-lab.github.io/NeMoBench.

cs.CV↗

arXiv2026

Five parameters are all you need (in $Λ$CDM)

The standard cosmological model, with its six independent parameters, successfully describes our observable Universe. One of these parameters, the optical depth to reionization $τ_\mathrm{reio}$, represents the scatterings that Cosmic Microwave Background (CMB) photons will experience after decoupling from the primordial plasma as the intergalactic medium transitions from neutral to ionized. $τ_\mathrm{reio}$ depends on the neutral hydrogen fraction $x_\mathrm{HI}$, which, in turn, should theoretically depend on cosmology. We present a novel method to establish the missing link between cosmology and reionization timeline using symbolic regression. We discover the timeline has a universal shape well described by the Gompertz mortality law, applicable to any cosmology within our simulated data. Unlike the conventional tanh prescription, our model is asymmetric in time and a good fit to astrophysical constraints on $x_\mathrm{HI}$. By combining CMB with astrophysical data and marginalizing over astrophysics, we treat $τ_\mathrm{reio}$ as a derived parameter, tightening its constraint to $<3\%$. This approach reduces the error on the amplitude of the primordial fluctuations by a factor of 2.3 compared to Planck's PR3 constraint and provides a commanding constraint on the ionization efficiency $ζ_\mathrm{UV} = 26.9^{+2.1}_{-2.5}$. We expect further improvements in the near term as reionization constraints increase and our understanding of reionization advances.

astro-ph.CO↗

arXiv2026

Density-Shear Baryon Acoustic Oscillation as a Cosmological Consistency Check

Tensions often arise between different datasets in cosmology, and consistency tests can serve as a powerful tool for diagnosing potential issues. Density-shear Baryon Acoustic Oscillation (GI BAO) is the imprint of the BAO feature on the shear field induced by the large-scale tidal field. We highlight that GI BAO can provide a robust consistency check for the density BAO, shear measurement, and alignment model. Failure of this check hints at systematics in any of these parts. As an illustration, we present the first GI BAO measurement on photometric data using the Dark Energy Survey Year 3 dataset, achieving a detection significance of $0.86 σ$. We find the GI BAO constraint on the BAO scale dilation parameter $α$ to be $ 0.966 \pm 0.252 $ (1$σ$), in good agreement with the density BAO constraint, $ 0.966 \pm 0.037 $, thereby validating the density BAO, shear measurement, and the linear alignment model. Furthermore, we argue that combining the density BAO with GI BAO yields results that are more resilient to systematic effects. Thanks to the massive data volumes of stage IV surveys, GI BAO will play an even more prominent role as a consistency check.

astro-ph.CO↗

arXiv2026

EgoSAT: A Comprehensive Benchmark of Egocentric Streaming Interaction Understanding

We introduce EgoSAT, the first comprehensive benchmark for egocentric video reasoning in streaming settings, designed to evaluate the capabilities of modern vision-language models (VLMs). The benchmark targets streaming interaction understanding, where video frames arrive sequentially and models must continuously interpret evolving visual context. EgoSAT unifies several previously distinct tasks within a single streaming framework. In this formulation, queries about completed events correspond to retrospective reasoning, queries about ongoing activities require online understanding, and queries about future actions involve prospective anticipation. This unified setting requires models to reason about the past, present, and future while operating under the constraint that only previously observed frames are available. EgoSAT contains 1,997 unique videos spanning 165 hours of egocentric footage and around 4,800 high-quality question-answer pairs, carefully designed to probe reasoning across varying temporal contexts. Using this benchmark, we evaluate a diverse set of both open-weight and closed-weight VLMs, providing a systematic assessment of their ability for streaming interaction understanding. By distinguishing answerability and conducting diagnostics on confidence of models, we find existing models not only struggle with prospective and retrospective modeling, but also exhibit severe mis-calibration: confidence often fails to track inherent answerability, leading to dangerous "confidently wrong" behaviors. Project page: https://leiyj23.github.io/EgoSAT/

cs.CV↗

arXiv2026

L20-Edu-135M: An Auditable Single-GPU Study of Data-Efficient Small Language Modeling

Small language models are cheap to serve and feasible on local hardware, but strong public 135M-class systems are commonly trained with hundreds of billions to trillions of tokens on large clusters. We study a sharply resource-constrained regime: a complete 134.5M-parameter language-model pipeline executed on one NVIDIA L20 GPU. The released checkpoint, L20-Edu-135M, receives approximately 13B pretraining tokens: 10B FineWeb-Edu tokens followed by a 3B-token educational, mathematics, code, and reasoning mixture. We document the architecture, data gates, cross-source MinHash/LSH near-deduplication, segment deduplication, benchmark-overlap removal, throughput optimization, supervised fine-tuning (SFT) with weight interpolation, and reinforcement learning from verifiable rewards (RLVR) on GSM8K. In a self-run zero-shot six-task harness, L20-Edu-135M obtains a mean score of 0.4150. It trails SmolLM-135M (0.4767) and SmolLM2-135M (0.4917), but its mean is 87.1% of SmolLM-135M's while its nominal token count is 2.17% as large. This ratio is descriptive, not evidence of statistical equivalence or a controlled scaling law. The model exceeds several older 100M-160M public baselines under the same harness. Direct GRPO-style RLVR decreases GSM8K exact-match accuracy from 1.82% to 1.59% (192-token completions) and 1.21% (320-token completions). These single-run results identify a concrete failure mode rather than establishing a general lower bound on RLVR. The contribution is an auditable resource-constrained case study, not a state-of-the-art claim.

cs.LG↗