arXiv subjects

Jingyu Li

Publications and source records attributed to Jingyu Li.

At least 19 recordsLinked to original sources

arXiv2026

Two-Dimensional Magnetic Materials from Ferromagnetism to Altermagnetism: Fabrication, Symmetry, Evidence, and Device Readiness

Atomically thin magnets are governed by composition, thickness, stacking, strain, defects, substrates, ferroic order, and interfaces. This Review organizes two-dimensional (2D) magnetism using net magnetization and momentum-space spin splitting or spin-polarization patterns as primary coordinates, supplemented by sublattice structure and the symmetry relations between opposite-spin sublattices. Ferromagnets generally carry a net moment; ferrimagnets can be fully compensated. Conventional antiferromagnets may retain spin-degenerate bands, while altermagnets permit nonrelativistic alternating spin splitting under dimension-specific symmetry constraints. M1-M4 reporting classes introduced here distinguish layer-resolved atomically thin forms, non-layer-resolved thin flakes or films, bulk benchmarks, and theoretical candidates or design routes; the evidence category and ordering-temperature provenance are reported independently. Current experimental support for altermagnetism comes mainly from M2 thin-film and M3 bulk benchmarks, leaving a fully verified atomically thin M1 altermagnetic phase as an open experimental goal. Most proposed monolayer and few-layer systems therefore fall within M4, and contested cases clarify the measurements needed for phase assignment. Device readiness is compared qualitatively using operating temperature, readout, switching, domain control, stability, scalability, interface sensitivity, and integration difficulty. Reliable phase assignment and device assessment require complementary measurements from the same material form under matched conditions, ideally on the same specimen or in situ.

cond-mat.mtrl-sci↗

arXiv2026

Natural Hard-Negative Mining from Positives for Instruction-Following Retrieval

Instruction-following information retrieval (IFIR) requires retrievers to rank documents by both a query and its instance-specific instructions, yet instruction-conditioned contrastive supervision remains scarce. Existing synthetic methods typically fix a positive passage and generate additional negatives, overlooking the hard-negative potential of the original positive. We propose Natural Hard-Negative Mining (NHNM), which promotes convertible negatives under query-compatible instructions and repurposes original positives as natural instruction hard negatives. Its automated Candidate Cleaning, Reversal Construction, and Geometry Filtering pipeline produces compact, high-contrast supervision aligned with the pretrained retrieval space. Applied to MSMARCO, NHNM yields NHNM-MARCO, an IFIR training set of nearly 7K instances. On InstructIR and FollowIR, models with 568M to 4B parameters fine-tuned on NHNM-MARCO achieve strong instruction-following performance and match or surpass substantially larger retrievers in several settings. Controlled analyses and ablations attribute these gains to natural hard-negative supervision rather than model scale or training-data volume.

cs.AI↗

arXiv2026

SMR: Learning streaming motion retargeting for humanoids from human semantic

Retargeting human motion to humanoids requires preserving motion semantics while accounting for the robot's physical constraints. Kinematic pipelines prescribe a robot trajectory through geometric matching, without establishing whether it can be executed under the robot's dynamics. Prior physics-supervised neural methods train bidirectional full-sequence models on physics-refined kinematic trajectories, remaining tied to prescribed kinematic solutions and requiring noncausal context. We propose SMR, a streaming neural motion retargeter learned from semantic-conditioned physics experts. We convert human motion into robot-aware task-space targets for body landmark positions, orientations, and velocities. Physics experts track these targets directly in simulation under contact, dynamics, and actuator constraints. We train a single strictly causal Transformer on their executed state trajectories that complete the motion and pass semantic and physical validity checks. At each incoming frame, the student uses only current and past human observations to generate a robot motion reference for a downstream whole-body controller. Experiments demonstrate that semantic-conditioned physics experts generate motion with higher semantic fidelity and physical validity than direct optimization-based retargeting, while the resulting streaming retargeter achieves improved closed-loop performance over kinematic and physics-refined baselines under a common whole-body controller.

cs.RO↗

arXiv2026

Asymptotic stability of strong rarefaction waves to a parabolic-hyperbolic system arising from chemotaxis

We are interested in the asymptotic behavior of solutions toward strong rarefaction waves for a parabolic-hyperbolic system arising from chemotaxis. Suppose that the Riemann problem to the corresponding inviscid system admits rarefaction waves. We show that if the initial data is a small perturbation of an approximate rarefaction wave, then the Cauchy problem has a unique global solution that converges to the rarefaction wave asymptotically in time. The stability results hold regardless of the wave strengths. The waves can be either a single rarefaction wave or a composite of two rarefaction waves. In the composite case, we separate the initial centers of the two approximate profiles to make their interaction error term small while retaining convergence to the prescribed Riemann wave. The proofs are based on the energy method, where the key observations are the monotonicity of the approximate rarefaction waves with respect to both space and time, and the appropriate replacement of spatial derivative of the hyperbolic component with the parabolic component in proper forms.

math.AP↗

arXiv2026

Real-Time Joint Audio-Video Generation by Parallel Adapter Composition

Deploying a joint audio-video diffusion transformer for real-time, interactive generation normally requires two essential modifications: block-autoregressive attention, so frames can be emitted before the whole clip is finished, and few-step sampling, so each block is cheap. Conventionally, the streaming video literature obtains both capabilities from a chained pipeline. It first distills a bidirectional teacher into a causal student, then into a few-step one, or proceeds in reverse order. Each stage of such a chain fine-tunes the weights the previous one produced, so a later objective can undo an earlier capability. Following the idea of model merging, we show that on a packed audio-video backbone the two capabilities can be acquired in parallel. A causal adapter is trained against the frozen backbone, and an off-the-shelf few-step adapter provides the few-step capability. As the two edit different functional axes, we predict, and then verify, that their weight-update directions are near-orthogonal, without any explicit orthogonality constraint during training. Orthogonal updates should combine without interfering, so parallel composition is a direct sum. The two adapters are simply added at inference, with no joint training, yielding few-step, streaming audio-video whose image quality tracks the bidirectional teacher. Compared to the chained baselines, the composed model matches or beats them on most metrics, making parallel composition a practical approach. The resulting streaming system generates joint audio-video in real time, $\approx$26 fps at $480\times832$ without quantization, and sustains 30 s of continuous generation with stable image quality. Demo Page: https://pac-demo-2027.github.io/demo/

cs.CV↗

arXiv2026

Spatial Latent Reasoning for Embodied Reference Understanding

Pointing-gesture visual grounding requires connecting hand geometry with the visual identity and extent of a referred object. A central challenge for continuous latent reasoning is how to organize these complementary cues into useful intermediate supervision. We propose Spatial Latent Reasoning (SLR), a framework that structures this supervision around an ordered sequence of geometric and visual states. A spatial ray state is supervised by fingertip position and pointing direction, followed by four states aligned with target-region features. To construct the visual targets, we introduce parity pooling, which applies polyphase grouping to average region tokens on four interleaved spatial supports. All states are generated recurrently during training and inference; auxiliary annotations are required only during training. On EgoPoint-Ground, the framework improves mIoU over same-backbone supervised fine-tuning by 2.8, 17.5, and 21.1 percentage points on Qwen3.5-4B, Qwen2.5-VL-7B, and Qwen3-VL-8B, respectively, with improvements on both hard subsets. On YouRefIt, it achieves 77.6% precision at IoU 0.5, a numerical margin of 5.2 percentage points over the reported state of the art under differing evaluation protocols. Ablations support joint geometric and visual supervision on the standard and similar-object sets, and favor parity over three alternative pooling operators on the standard set. These results support task-structured supervision for continuous pointing grounding. We will release the code and supporting materials.

cs.CV↗

arXiv2026

LongEmo: Towards Emotion Understanding and Reasoning in Long Videos

While recent Multimodal Large Language Models (MLLMs) have shown promise in affective computing, their reasoning capabilities are largely confined to short video clips with limited interactions. However, real-world emotions are not merely isolated instantaneous reactions but dynamic and cumulative processes deeply shaped by past experiences and ongoing events. To bridge this gap, we introduce LongEmoBench, a benchmark dedicated to emotion understanding and reasoning in long videos. It assesses progressive capabilities scaling from continuous scene interactions to complex episodic developments. Furthermore, we propose LongEmo, a novel memory-augmented agentic framework designed to tackle the immense challenges of long-range affective reasoning. LongEmo processes continuous video streams to construct an Event Memory Graph, explicitly modeling long-range dependencies and capturing emotional dynamics across discrete events. Given a question, the agent retrieves a query-relevant event stream from the graph, iteratively integrating multimodal memories and relational dependencies to deduce the final answer. Extensive evaluations of 17 representative methods reveal that they struggle significantly with emotion understanding and reasoning in long videos. In contrast, LongEmo achieves state-of-the-art performance, demonstrating the efficacy of its event-centric memory architecture.

cs.CV↗

arXiv2026

Bare-Die Antiferromagnetic Computing

Semiconductor electronic devices are increasingly constrained by fundamental quantum tunneling effects and charge-based mechanisms, which severely limit further miniaturization, write-speed scaling, and environmental robustness of silicon-based technologies. These limitations are particularly prohibitive for deep-space exploration, where extreme temperatures, ultra-strong magnetic fields, and intense radiation rapidly incapacitate conventional electronics without massive shielding. Here, we present an intrinsically resilient, strain-mediated antiferromagnetic MnIr/PMN-PT edge processor that operates reliably as a bare die under temperatures up to 500 K, magnetic fields of 55 T, and radiation doses of 1.5 Mrad. By exploiting an input-modulated in situ self-refreshing encoding mechanism, the device performs nonlinear feature extraction and classification directly from raw analog signals, enabling an analog computing architecture that requires no time-frequency transformation. This architecture achieves 99.8% accuracy in speech recognition without digital preprocessing and 100% accuracy in astronaut visual object recognition. Furthermore, an all-hardware integrated drone vision system demonstrates real-time in situ command execution and autonomous navigation, delivering a terahertz-level response frequency and an ultra-low energy consumption of approximately 0.2 fJ per operation. This work expands the functional scope of antiferromagnetic devices beyond memory and logic, establishing them as a promising materials platform for energy-efficient physical computing and autonomous intelligence in extreme environments.

cond-mat.mes-hall↗

arXiv2026

Spin-Polarized Magnetic Metal Electrodes for Magnetic Tunnel Junctions

Magnetic tunnel junctions are foundational components of spintronic memory, sensing, and computing, and their performance depends critically on the magnetic metallic electrodes. This Perspective examines electrode materials through the lens of magnetic order and the distinct microscopic mechanisms that generate spin-selective tunneling. Conventional ferromagnets, including CoFeB and half-metallic Heusler alloys, support exchange-split electronic states and symmetry-filtered tunneling, whereas compensated collinear and noncollinear antiferromagnets exploit sublattice selectivity, spin-orbit anisotropy, vector spin textures, and magnetic multipoles. Altermagnets provide a collinear, zero-net-moment route based on symmetry-allowed momentum-dependent spin splitting. Across these material classes, we compare the origins of tunneling polarization, the roles of barrier evanescent states and interface termination, and strategies for electrically writing and reading the relevant magnetic order. This comparison reveals a broader design principle: effective electrode polarization is not a scalar bulk quantity, but a momentum-, orbital-, symmetry-, and interface-resolved transport property. Beyond the pursuit of ever larger tunneling magnetoresistance, future progress will depend on converting the unconventional spin polarization of emerging magnetic metals into electrically addressable, thermally robust, and reproducible tunneling functionality at realistic interfaces.

cond-mat.mes-hall↗

arXiv2026

Stratification for Nonlinear Semidefinite Programming

This paper introduces a stratification framework for nonlinear semidefinite programming (NLSDP) that reveals and utilizes the geometry behind the nonsmooth KKT system. Based on the index stratification of $\mathbb{S}^n$ and its lift to the primal-dual space, a stratified variational analysis is developed. Specifically, we define the stratum-restricted regularity property, characterize it by the verifiable weak second order condition (W-SOC) and weak strict Robinson constraint qualification (W-SRCQ), and interpret the W-SRCQ geometrically via transversality, with stability along strata. The interactions of these properties across neighboring strata are further examined, leading to the conclusion that classical strong-form regularity conditions correspond to the local uniform validity of stratum-restricted counterparts. On the algorithmic side, a stratified Gauss--Newton method with normal steps and a correction mechanism is proposed for globally solving the KKT equation through a least-squares merit function. We demonstrate that the algorithm converges globally to directional stationary points. Moreover, under the second order sufficient condition (SOSC) and the strict Robinson constraint qualification (SRCQ) at an accumulation point, with a suitable correction threshold, the whole sequence converges superlinearly to this point, which is a KKT pair, and eventually identifies the active stratum. The rate is quadratic if the problem data are additionally of class $LC^2$ near the solution.

math.OC↗

arXiv2026

Nonlinear stability of composite waves of traveling wave and rarefaction wave for a parabolic-hyperbolic system arising from chemotaxis

We study the nonlinear stability of a composite wave consisting of a traveling wave and a rarefaction wave for a parabolic-hyperbolic system arising from chemotaxis. We prove that if the initial value is a small $H^1$-type perturbation of composite wave, then the system admits a global solution that converges toward the composite wave with an absolutely continuous shift. The proof combines the weighted relative-entropy mechanism for viscous shocks with the energy structure of rarefaction waves. A key ingredient is the inclusion of a rarefaction modulation factor in the weighted relative entropy; its derivatives combine with the terms generated by the rarefaction profile to produce a rarefaction dissipation. Moreover, the spatial separation of the two waves yields time-integrable interaction errors caused by the non-exact superposition. Since diffusion acts only on the density component, the full $H^1$ estimate is closed by exploiting the coupling structure of the system to recover the missing dissipation of the hyperbolic component.

math.AP↗

arXiv2026

When Successful Memories Mislead Embodied Agents:Memory Adaption For Task-Conditioned Execution

Experience reuse can reduce repeated exploration in embodied agents, but a trajectory that succeeded previously may be unsuitable for the current execution context. Existing memory systems pri marily optimize construction and retrieval; semantic relevance and historical success therefore remain insufficient when retrieved ex perience contains incompatible actions or an inappropriate level of structure. We introduce Memory Adaptation for Task-Conditioned Execution (MATE), a deterministic post-retrieval procedure that converts trajectories into execution-oriented memory. MATE re moves obsolete control context, extracts condition-action-effect transitions, applies verified action normalization, selects a task dependent representation, and serializes the result under a fixed budget without additional LLM inference. On 134 ALFWorld tasks, MATE achieves task success rates of 81.3% and 93.3% with Qwen2.5-14B and 72B while using approximately one-tenth of the tokens required by raw trajectories. Controlled comparisons show that verified action normalization is the principal mechanism by which MATE restores the utility of retrieved experience, support ing memory adaptation as a distinct stage between retrieval and embodied execution.

cs.CL↗

arXiv2026

Omni-Streaming Thinking

Streaming omni-modal models must decide what and when to answer from the video chunks and synchronized audio observed so far. Visual cues often support an interpretation before an utterance or sound event is complete. If that interpretation enters memory as a fact, later reasoning can keep relaying it even after audio contradicts it. We call this failure premature cross-modal commitment. We propose Omni-Streaming Thinking (OST), which generates structured outputs that include evidence observed so far, forecasts of future evidence, and claims based on this evidence. Each claim is initially marked as pending and linked to a future verification interval. Audio and visual evidence are stored separately, and OST checks a claim against the evidence from the specified modality at the end of the verification interval. When contradictory evidence is detected, a refutation process reduces the influence of the claim and its dependent states, and then guides a state update using the new evidence. An answer gate decides whether the answer-critical claims meet the conditions for giving a response. Using a frozen Qwen3-Omni-30B-A3B-Instruct backbone with lightweight adaptation, OST outperforms the strongest open baselines on five streaming and audio-visual benchmarks by more than 10% relative on average. We also introduce OST-DiagBench, which holds video fixed and edits audio to test agreement, absence, contradiction, coexistence, and subtitle-speech conflict. OST reaches d-prime = 2.95, compared with at most 1.38 for open baselines, while reducing vision-induced auditory hallucinations.

cs.LG↗

arXiv2026

A Large-Scale Probing Analysis of Speaker-Specific Attributes in Self-Supervised Speech Representations

Enhancing explainability in speech self-supervised learning (SSL) is important for understanding and effectively utilising speech SSL representations. This study conducts a large-scale layer-wise probing analysis of 11 speech SSL models, examining speaker identity together with acoustic, prosodic, and paralinguistic attributes. The results confirm a general hierarchy wherein initial layers encode fundamental acoustics and middle layers synthesise abstract traits. The consensus that final layers purely abstract linguistic content is challenged, as larger models unexpectedly recover speaker-discriminative information in their deep layers. Furthermore, the intermediate representations of speech SSL models are found to provide stronger probing performance for several attributes than specialised speaker embeddings. These insights give a systematic empirical characterisation of speaker-related information in SSL models, providing guidelines for selecting representations for downstream tasks.

eess.AS↗

arXiv2026

CUHK-EE Systems for the vTAD Challenge at NCMMSC 2025

This paper presents the Voice Timbre Attribute Detection (vTAD) systems developed by the Digital Signal Processing & Speech Technology Laboratory (DSP&STL) of the Department of Electronic Engineering (EE) at The Chinese University of Hong Kong (CUHK) for the 20th National Conference on Human-Computer Speech Communication (NCMMSC 2025) vTAD Challenge. The proposed systems leverage WavLM-Large embeddings with attentive statistical pooling (ASTP) to extract robust speaker representations, followed by two variants of Diff-Net, i.e., Feed-Forward Neural Network (FFN) and Squeeze-and-Excitation-enhanced Residual FFN (SE-ResFFN), to compare timbre attribute intensities between utterance pairs. Experimental results demonstrate that the WavLM-Large+FFN system generalises better to unseen speakers, achieving 77.96% accuracy and 21.79% equal error rate (EER), while the WavLM-Large+SE-ResFFN model excels in the 'Seen' setting with 94.42% accuracy and 5.49% EER. These findings highlight a trade-off between model complexity and generalisation, and underscore the importance of architectural choices in fine-grained speaker modelling. Our analysis also reveals the impact of speaker identity, annotation subjectivity, and data imbalance on system performance, pointing to future directions for improving robustness and fairness in timbre attribute detection.

eess.AS↗

arXiv2026

An Investigation of Reprogramming for Cross-Language Adaptation in Speaker Verification Systems

Language mismatch is among the most common and challenging domain mismatches in deploying speaker verification (SV) systems. Adversarial reprogramming has shown promising results in cross-language adaptation for SV. The reprogramming is implemented by padding learnable parameters on the two sides of input speech signals. In this paper, we investigate the relationship between the number of padded parameters and the performance of the reprogrammed models. Sufficient experiments are conducted with different scales of SV models and datasets. The results demonstrate that reprogramming consistently improves the performance of cross-language SV, while the improvement is saturated or even degraded when using larger padding lengths. The performance is mainly determined by the capacity of the original SV models instead of the number of padded parameters. The SV models with larger scales have higher upper bounds in performance and can endure longer padding without performance degradation.

eess.AS↗

arXiv2026

Fourier Neural Operators for Composition-Driven Crystal Structure Discovery

Crystalline materials discovery is essential for energy, electronics, and catalysis, but the vast chemical and structural space makes exhaustive screening infeasible. Existing voxel-based methods are limited by the local receptive fields of three-dimensional convolutional neural networks and the posterior collapse of high-dimensional variational autoencoders. Here, we develop a Fourier Neural Operator (FNO)-based crystal-field solver that maps a prescribed chemical formula and lattice parameters to periodic number-density and electron-density fields. By operating on global Fourier modes, the solver captures long-range correlations in periodic crystal fields beyond conventional local convolutions. Building on this solver, we construct a coupled generation-solving framework in which a conditional variational autoencoder generates diverse candidate lattice parameters in a low-dimensional basis-coefficient space, followed by density-field prediction and atomic reconstruction through peak detection, position optimization, and weight optimization. The reconstructed structures are further screened using voxel-level filtering, machine-learning interatomic-potential relaxation, and first-principle calculations. The framework generates novel structures across 104 chemical formulas with competitive reconstruction accuracy, demonstrating high generative diversity and structural validity. By extending Fourier neural operators to periodic crystal fields and coupling them with composition-conditioned lattice generation, our approach provides a scalable route to crystal structure discovery from prescribed chemical compositions.

cond-mat.mtrl-sci↗

arXiv2026

FutureBridge: Token Selection Beyond Local Preference in Collaborative Decoding

Token-level collaboration allows a large language model (LLM) to assist a small language model (SLM) when their predictions diverge. Existing methods either use LLM-generated intervention tokens or rank candidates with the LLM's next-token probabilities. Both rely on the LLM's local preference, even though an LLM-selected token may be difficult for the SLM to build on. We present FutureBridge, which ranks joint LLM-SLM token candidates according to how well they support the SLM's subsequent reasoning. During training, an answer-verified LLM trajectory supplies a fixed shared future, and a frozen SLM evaluates every candidate under this common context. The resulting counterfactual scores supervise a lightweight token reranker that observes only the current state and candidate token. At inference, FutureBridge uses the LLM only to expand the candidate pool, selects one token, and returns generation to the SLM without generating or appending a future suffix. Across five mathematical reasoning benchmarks, FutureBridge improves the Qwen3-1.7B SLM's Math Avg. by 35.1% relative to greedy SLM decoding. These results indicate that token selection benefits from modeling whether the receiving SLM can use each candidate to continue reasoning, rather than relying on the LLM's local preference alone.

cs.CL↗