arXiv subjects

Yi Wang

Publications and source records attributed to Yi Wang.

At least 19 recordsLinked to original sources

arXiv2026

Keepace.ai: A Technical Report on a Large Language Model for Sports and Health

Large language models hold promise for delivering safe, scientifically grounded, and personalized exercise guidance at scale, yet the sports-and-health domain still lacks curated training corpora and rigorous evaluation benchmarks. Keepace.ai is a domain-specialized LLM built on Qwen3.5-27B through post-training, currently optimized for Chinese-language exercise guidance across three tasks: workout plan generation, sports-and-health knowledge QA, and personal sports data analysis. In internal rubric-based evaluations designed with domain experts, Keepace.ai outperforms its base model and larger general-purpose baselines, with the most pronounced improvements in exercise safety, boundary compliance, structured output completeness, and scientific accuracy. This report summarizes the current design, data pipeline, safety-first evaluation framework, and evaluation results of Keepace.ai.

cs.CL↗

arXiv2026

FTP-1: A Generalist Foundation Tactile Policy Across Tactile Sensors for Contact-Rich Manipulation

Despite the success of vision-based generalist robotic policies, existing tactile-based policies remain tied to fixed embodiments and sensor setups. This is because tactile signals are highly heterogeneous across hardware, making cross-sensor generalization difficult. We present FTP-1,the first generalist foundation tactile policy pretrained to acquire transferable tactile manipulation abilities across diverse sensors and embodiments. FTP-1 supports varied tactile inputs, including image-, array-, and state-based signals, by using heterogeneous encoders to project them into unified morphology-aware latent tokens that are jointly modeled by a shared tactile Transformer expert. Pretrained on around 3,000 hours of tactile manipulation data aggregated from 26 data sources, spanning human and robot demonstrations across 21 sensors, FTP-1 learns tactile skills that transfer beyond the sensors seen during pretraining. Across downstream finetuning experiments spanning 5 hardware configurations, FTP-1 improves contact-rich manipulation on seen sensor setups by +17.2% and, surprisingly, transfers to two previously unseen tactile-sensor setups, achieving a +31% gain in success rate. FTP-1 establishes the first unified foundation baseline for tactile manipulation, providing future tactile policies with a shared model-level starting point. Pretrained models, datasets, training code and more visualization at https://ftp1-policy.github.io.

cs.RO↗

arXiv2026

A complete solution to the Boots-Royle/Cao-Vince conjecture

Boots and Royle, and independently Cao and Vince, conjectured that the join of an edge and a path on $n-2$ vertices is the unique planar graph of maximum adjacency spectral radius for $n\geq 9$. Tait and Tobin (JCTB, 2017) proved the conjecture for sufficiently large order. In this paper, we completely resolved the Boots-Royle/Cao-Vince conjecture.

math.CO↗

arXiv2026

Recurrent Self-Improvement: Dynamic Cross-Loop On-Policy Distillation for Looped Language Models

Looped Language Models (LoopLMs) offer a parameter efficient approach to scaling reasoning by reusing shared parameters across recurrent computation steps. Despite their promise, effective post-training of LoopLMs remains challenging. Existing approaches either provide reward based supervision that is sparse or costly to extend across loops, or rely on external teachers or privileged information, leading to limited teacher availability or teacher-student context mismatch. To address these limitations, we introduce LoopOPD, a cross-loop on-policy distillation framework that uses additional recurrent computation within a LoopLM as its own source of supervision. LoopOPD uses a frozen terminal loop policy as a compute privileged teacher for an intermediate loop student on student generated rollouts, providing dense supervision without an external teacher or privileged information. We further propose Dynamic LoopOPD (D-LoopOPD), which continually refreshes the terminal loop teacher as the shared model parameters are updated, enabling recurrent self-improvement. We characterize how distillation updates propagate across loop depths and derive sufficient conditions under which a single update yields simultaneous local improvement at both loop depths. Experiments on Ouro-Thinking models show that LoopOPD improves mathematical reasoning, while D-LoopOPD yields further gains through dynamic teacher updates. Despite being trained only on mathematical data, the resulting models also improve on general reasoning and code generation benchmarks, demonstrating that recurrent computation can serve as an effective source of supervision for LoopLMs. Our code and model checkpoints will be released upon acceptance.

cs.LG↗

arXiv2026

Safe at One Loop, Risky at Another: Aligning Safety Across Recurrent Depths in Looped Language Models

Looped Language Models (LoopLMs) provide a parameter efficient approach to scaling model capabilities through repeated use of shared parameters across recurrent steps. Since each recurrent depth can be read out independently, a single LoopLM exposes a broader output space across inference depths, raising an important question: whether safety is preserved throughout recurrent computation. Prior evaluations suggest that deeper recurrence can improve safety on harmful queries, but robustness under jailbreak attacks remains unclear. We therefore conduct a comprehensive safety evaluation of LoopLMs under jailbreak attacks targeting different recurrent depths. We find that attack success can increase at deeper inference depths, the same query can elicit different safety behaviors across depths, and attacks constructed against one depth can transfer to others. Moreover, SFT and preference alignment do not eliminate these safety gaps, motivating an alignment method designed for LoopLMs. We introduce SafeBridge, which combines lightweight depth specific control of shared recurrent layers, selective state bridging, and joint safety supervision across recurrent depths. Across model scales, multiple attack methods, and safety benchmarks, SafeBridge substantially reduces attack success for both matched-depth and cross-depth attacks. It also improves general utility over the vanilla models while maintaining comparable over-refusal behavior. Our results show that the safety of a LoopLM cannot be inferred from a single recurrent depth, motivating safety alignment across recurrent computation. Our code and model checkpoints will be released upon acceptance.

cs.CR↗

arXiv2026

Coanalytic subfields and bases of prescribed dimension

Assuming $V=L$, we construct proper coanalytic subfields of $\mathbb{R}$ and coanalytic Hamel and transcendence bases of every possible Hausdorff dimension, uniformly in the relevant parameters. There is a single lightface $Π^1_1$ set $G\subseteq[0,1]\times\mathbb{R}$ whose sections $A_α$ are proper subfields with $\dim_{\mathrm H}A_α=α$, form a strictly increasing chain, and satisfy $A_α=\bigcup_{β<α}A_β$ for $α>0$. For every $n\ge1$, every countable subfield $K\subseteq\mathbb{R}$ and every $s\in[0,n]$, we construct a coanalytic Hamel basis of $\mathbb{R}^n$ over $K$ of Hausdorff dimension $s$. Likewise, for every countable subfield $K\subseteq\mathbb{C}$ and every $s\in[0,2]$, we construct a coanalytic transcendence basis of $\mathbb{C}$ over $K$ of Hausdorff dimension $s$. In each case, the bases are sections of a single lightface $Π^1_1$ relation, uniformly in codes for the coefficient field and in the prescribed dimension, as well as in $n$ for Hamel bases. More generally, we prove a uniform theorem for suitably coded families of pregeometries on Euclidean spaces. The constructions combine Cantor sets of controlled dimension with hyperarithmetic coding and coanalytic transfinite recursion. The complexity is optimal: proper subfields of $\mathbb{R}$ of positive Hausdorff dimension and bases of the stated kinds cannot be analytic.

math.LO↗

arXiv2026

Bistable Dynamics and Complete Classifications of A Seasonal Competition Model Between Aedes aegypti and Aedes albopictus

The bistable dynamics and a complete classification of global dynamics are established for a competition model between Aedes aegypti and Aedes albopictus subject to favorable-unfavorable seasonal shifts, originally proposed by Zheng et al. (2025). More precisely, by employing Floquet theory, we prove that the unique positive periodic solution must be unstable. Consequently, bistable dynamics naturally emerge, which provides a rigorous theoretical validation of the numerical simulation results reported in the aforementioned work. Building on this, we further derive a full classification of the model's global dynamics. In addition, we carry out a comprehensive analysis of how seasonal switching mechanisms govern the competitive outcomes between these two mosquito species, thereby offering insights into their ecological coexistence and potential displacement.

math.DS↗

arXiv2026

IndusAgent: Agentic Reinforcement Learning for Cross-Category Industrial Anomaly Detection

Small defects in unfamiliar product categories can be difficult to distinguish from benign appearance variations in a global view. Additional reasoning over the same view may leave this ambiguity unresolved. We introduce IndusAgent, a Qwen3-VL-8B inspection policy trained through supervised fine-tuning and tool-augmented reinforcement learning for cross-category industrial anomaly detection. The policy selects among six tools for cropping, enhancement, measurement, within-image comparison, texture analysis, and external visual opinions from Qwen3-VL-Max, then integrates returned observations into a structured diagnosis. Indus-CoT provides 3,000 trajectories from 26 Real-IAD categories, generated with actual tool feedback and screened against annotations after teacher generation. GRPO optimizes an accuracy-gated objective that conditions localization, type, and tool rewards, including call costs, on final diagnostic correctness while retaining an independent format reward. Across five benchmarks, IndusAgent achieves 83.5% mean balanced accuracy, compared with 79.4% for direct Max inference under the common protocol. Across three training seeds, the five-benchmark mean balanced accuracy is 80.5\% for a weight-matched ungated control and 83.5% for the gated policy. Evaluation uses disjoint product categories and a fixed defect vocabulary.

cs.CV↗

arXiv2026

RoboPocket: Improve Robot Policies Instantly with Your Phone

Scaling robot learning is bottlenecked not only by data volume, but also by whether the data targets the policy's failures. Existing robot-free handheld interfaces enable portable in-the-wild data collection, yet remain open-loop: users record demonstrations without feedback from the current policy. Interactive methods such as DAgger provide such feedback by collecting corrective on-policy data, but require deploying imperfect policies on physical robots, making them costly, unsafe, and difficult to scale. We present RoboPocket, a smartphone-based system for robot-free policy iteration. By combining on-device sensing, display, and communication with cloud-side policy inference, RoboPocket provides real-time collection guidance and visualizes the policy's predicted trajectory in augmented reality as AR Visual Foresight. This allows users to identify policy weaknesses and collect targeted corrective data without deploying a physical robot. An asynchronous finetuning pipeline continuously incorporates new corrections and updates the policy within minutes, realizing a DAgger-inspired robot-free correction loop. Across diverse real-world manipulation tasks, we validate RoboPocket as a scalable data engine and show that online corrective data collection improves sample efficiency by up to $2\times$ over scaling offline demonstrations. Project page and videos: https://robo-pocket.github.io.

cs.RO↗

arXiv2026

Dynamic Hotspot Migration and Self-Induced Resonances in Ultrafast Laser-Irradiated Silicon Thin Films

The spatial distribution of energy in silicon thin films under femtosecond laser irradiation is demonstrated to be influenced by the interaction between carrier-induced refractive index modulation and self-induced Fabry-Pérot resonances. We resolve three characteristic excitation regions such as interface-localized, binodal and multi-nodal regimes considering the analysis of silicon-on-insulator (SOI) configurations across thicknesses of 100, 230, and 350 nm. We report a nonlinear hotspot migration effect where the transient reduction of the refractive index stretches the internal standing wave, translating intensity antinodes toward the film volume during the pulse. Numerical results demonstrate that at higher fluences, plasma-induced extinction suppresses internal reflections, triggering a transition from resonance-mediated absorption to plasma shielding. These findings establish a physical basis for depth-resolved modification in laser nanomanufacturing, providing a dual-capacity for simultaneous surface patterning and depth-resolved subsurface modification with direct applications in laser-assisted interfacial separation and 3D integrated photonic fabrication.

physics.optics↗

arXiv2026

Continuous-Phase Frequency-Hopping MIMO Signaling and Processing for ISAC

Frequency-hopping MIMO integrated sensing and communications (FH-MIMO-ISAC) supports agile multidimensional signaling but faces challenges in phase-continuous transmission, demodulation without channel-phase information, and coherent processing of multi-frequency echoes. We propose a continuous-phase frequency-hopping (CP-FH) MIMO ISAC waveform with matched communication and radar processing. Information is conveyed jointly by frequency combinations and continuous-phase-modulation symbols. CPM phase accumulation and frequency-hopping linear-frequency-modulation phase compensation maintain phase continuity and reduce spectral leakage. We derive exact inter-element orthogonality conditions, residual cross-correlation bounds, and a bound on averaged transmit-beampattern ripple. For communication, projection-energy demodulation requires no channel-phase information and approximates least-squares detection under weak inter-branch correlation. It separates frequency-combination search from CPM-symbol decisions, reducing complexity while retaining direct joint-search decisions. For radar, structure-matched coherent processing combines space-time-frequency matched filtering, velocity-consistent Doppler focusing, and frequency-aperture focusing to form a three-dimensional coarse-range--velocity--fine-range response. It reduces Doppler-mismatch loss and extra detections from periodic range grating lobes in conventional pulse-Doppler processing. We further derive the joint range response, output signal and noise powers, and receive beampattern. The analysis characterizes target-discrimination limits under periodic ambiguity and shows that the proposed processor preserves frequency-aperture range resolution and full-aperture angular resolution, with equal output noise power under the stated conditions.

eess.SP↗

arXiv2026

Imagine Before You Predict: Interleaved Latent Visual Reasoning for Video Event Prediction

Video event prediction (VEP) requires models to infer unobserved future states from partial video evidence. Existing video MLLMs usually verbalize intermediate future reasoning in text space: once visual evidence is verbalized, fine-grained motion, geometry, and interaction cues can be lost, leading to plausible but visually ungrounded hallucinations. We introduce Future-L1, an interleaved latent visual reasoning framework that lets an MLLM alternate between language tokens and continuous latent visual spans during autoregressive decoding. To train this capability, we construct Future-L1-50K by selecting examples where future visual hints help prediction and align latent states to future-frame embeddings, then further optimize sampled latent trajectories with LA-DAPO, a latent-aware RL objective with outcome-contrastive and temporal-diversity rewards. Future-L1 achieves new state-of-the-art results on both benchmarks: on FutureBench, it improves Qwen3-VL-8B from 61.0 to 85.4 and exceeds the previous best Video-CoE by 10.4 points; on TwiFF-Bench, it improves the average score from 2.44 to 3.04. These results suggest that future-oriented video reasoning benefits from preserving intermediate visual semantics in latent space rather than translating every reasoning step into text.

cs.CV↗

arXiv2026

A Zero--One Law for Subrings of the Reals Closed under Binary Interleaving

Erdős Problem 1154 asks whether every number in $[0,1]$ occurs as the Hausdorff dimension of a subring or subfield of $\mathbb{R}$. We prove that a subring of $\mathbb{R}$ closed under binary interleaving has Hausdorff dimension either zero or one. Consequently, every Turing ideal and its associated subfield satisfy the same dichotomy.

math.LO↗

arXiv2026

Magnetic Reconnection Governs the Entire Lifecycle of Solar Eruptive Magnetic Flux Rope

While magnetic flux ropes (MFRs) are widely accepted to exist prior to many solar eruptions, the physical mechanism responsible for triggering and subsequently driving their eruptions remains debated. Classical models commonly attribute eruption onset to ideal magnetohydrodynamic instabilities, particularly the torus instability, while magnetic reconnection is often regarded as a secondary process. Using high-resolution three-dimensional MHD simulations, we investigate the eruption of an MFR that forms self-consistently through flux cancellation driven by converging motions and photospheric diffusion following an initial shearing phase. The pre-eruption MFR rises quasi-statically and reaches a height where the decay index exceeds the classical torus instability threshold, yet the system continues to evolve quasi-statically without eruption. Instead, eruption begins only when the current sheet beneath the MFR becomes sufficiently thin for magnetic reconnection to set in. Increasing numerical resolution systematically delays the eruption onset, whereas the decay index at eruption onset remains nearly unchanged, demonstrating that the onset is governed by reconnection rather than torus instability. During the eruption, magnetic reconnection continuously converts surrounding sheared arcades into newly reconnected twisted flux that merges with the pre-eruption MFR and continuously to build the erupting MFR. A component-resolved analysis of the Lorentz force work further shows that the newly reconnection flux contributes predominantly to the main acceleration phase. These results demonstrate that magnetic reconnection plays a central role throughout the entire lifecycle of the eruptive MFR: it builds the MFR, directly triggers its eruption, and provides the dominant driving of the eruption.

astro-ph.SR↗

arXiv2026

Designing Textual Interfaces To Enhance Llm Reasoning On Audio-Visual Question Answering

Automatic prompt optimization (APO) has shown strong effectiveness for text-based LLMs and LLM systems, but its application to audio-visual understanding systems requires suitable textual interfaces through which multimodal reasoning can be exposed to text-based optimizers. We investigate how the design of textual interfaces affects prompt optimization for multimodal reasoning, on the audio-visual question answering task. We compare three interface designs, including a direct inference system consisting of a single omni-modal large language model (OmniLLM), a captioner-reasoner system, and an agentic architecture in which a primary agent orchestrates subagent OmniLLMs. Experiments with Qwen3-Omni and Gemini show that direct prompt optimization provides model-dependent gains; the captioner-reasoner system and the agentic system expose more interpretable optimization interfaces but are harder to optimize, where error analysis of cases affected by optimization is conducted. Our study highlights textual interface design as an important factor in making prompt optimization effective for omni-modal reasoning systems. Our implementation details are specified in GitHub repo.

cs.AI↗

arXiv2026

LoopArena: Benchmarking Models as Runtime Controllers for Loop Engineering

Loop Engineering is emerging as a practice for organizing development work around coding agents. Instead of writing each prompt by hand, practitioners design loops that monitor progress, assign work, run checks, and decide what the agent should do next. Even with a capable coding agent, a loop may trust a stale progress note, skip needed verification, spend its budget in the wrong direction, or stop before the task is safe to submit. Yet the final outcome of one end-to-end run cannot tell whether success or failure reflects the loop's guidance or the coding agent's ability to carry out the task. We introduce LoopArena, a benchmark for evaluating how well one model can guide a separate coding agent through a long-running task. The model under evaluation is the Controller: after each coding round, it receives a structured summary of the run and instructs a separate, fixed coding agent, the Worker, on what to do or verify next, or decides whether to stop. LoopArena evaluates this ability in three complementary settings that differ in execution scope and cost. Type I consists of 160 low-cost control questions; its execution cost is incurred once during benchmark construction, so each new Controller is evaluated through a single four-way choice with no code execution. Type II evaluates Controller guidance on 80 selected task slices, while Type III evaluates the corresponding 80 full tasks from their original states. On full tasks, the best observed Strict Success Rate is 36.25%, leaving substantial room for improvement in long-horizon loop control. Across Controllers, the paired reduction in estimated inference cost averages 60.6%, and Type II produces a positively correlated ordering under the main Core criterion (Spearman's rho = 0.7353). We release the benchmark data and evaluation code at https://github.com/AMAP-ML/LoopArena.

cs.AI↗

arXiv2026

Physics-Aware Object-Centric SE(3) Forecasting for Real-Time Proactive Dynamic Grasping

Proactive dynamic grasping requires robots to predict the future state at which physical contact will occur, rather than reacting to the currently observed object pose. But what future must the robot predict, and at what level of abstraction? We argue that contact-relevant rigid-body state in SE(3) is a compact task-sufficient abstraction for this problem: it directly determines where and when the robot should establish contact, without requiring visually complete future scenes. This insight leads to a lightweight object-centric forecast-then-select framework that predicts future rigid-body trajectories from short pose histories and uses a reachability-aware interception selector to choose feasible contact states under kinematic and timing constraints. Trained with randomized physics simulation over primitive shapes, the forecaster transfers to real objects once their metric geometry and SE(3) state histories are recovered. Experiments across simulation and real-world dynamic grasping tasks show that our framework improves the average success rate from $24.5\%$ to $62.0\%$ in simulation, and from $15.3\%$ to $67.3\%$ in real-world catching, conveyor, and turntable interception. The SE(3) forecaster runs in $15\,\mathrm{ms}$, orders of magnitude faster than dense visual future prediction, enabling real-time contact decision making. Our results suggest that effective proactive grasping does not require imagining the entire future scene. Instead, predicting the compact physical future that determines contact provides an effective and practical abstraction for real-time dynamic grasping.

cs.RO↗

arXiv2026

Urbex: Agentic Spatial Grounding for City-Scale 3D Scenes

City-scale 3D grounding is commonly formulated as candidate ranking, which requires precomputed instance proposals and becomes inefficient in large outdoor scenes. We propose Urbex, an agentic vision-language framework that instead casts grounding as active spatial search. Urbex represents a city-scale 3D scene as an interactive multi-view environment, allowing a VLM agent to query landmarks, zoom into local regions, render oblique views, and commit a grounded bounding box. We optimize this tool-use policy with reinforcement learning on CityRefer, using smooth localization rewards and lightweight shaping rewards for efficient exploration. Experiments show that Urbex achieves the best Hit@1 and IoU on CityRefer, supports candidate-free grounding without ground-truth boxes, and is substantially faster than the strongest city-scale ranking baseline. Zero-shot results on STPLS3D-Refer further indicate improved cross-scene robustness.

cs.CV↗