arXiv subjects

Junxi Wang

Publications and source records attributed to Junxi Wang.

At least 19 recordsLinked to original sources

arXiv2026

MemForest: Efficient Agent Memory Management via EventTree Partitioning and Progressive Merging

Agent memory systems have demonstrated significant potential in long-term dialogue, personalized assistants, and video understanding. However, continuously accumulated memory introduces substantial storage and retrieval costs during inference. To address this issue, we propose \textbf{MemForest}, a general memory compression framework adaptable to various agent memory systems. Specifically, MemForest partitions historical memory into event-centric units by leveraging global semantic similarity and local temporal continuity. For each unit, it constructs a maximum spanning tree, termed an EventTree, and progressively merges redundant memory nodes by selecting high-weight edges, reducing storage overhead. Furthermore, we introduce an anchor-guided propagation retrieval mechanism that retrieves relevant memory nodes from the temporal neighborhoods of key nodes, improving retrieval accuracy. Extensive experiments demonstrate the effectiveness of MemForest. Under the unimodal Mem0 framework, MemForest retains \textbf{97.1%} of the original performance while compressing \textbf{50%} of historical memory across three benchmarks (LoCoMo, LongMemEval, and PersonaMem), achieving a \textbf{1.89x} retrieval speedup. Under the multimodal M3-Agent framework, it preserves \textbf{99.7%} of the original performance with a \textbf{50%} compression ratio across two benchmarks (M3-Bench-robot and M3-Bench-web), achieving a \textbf{2.24x} retrieval speedup. \textcolor{RoyalBlue}{\textit{Our code is available at [https://github.com/Celina-love-sweet/MemForest.}}](https://github.com/Celina-love-sweet/MemForest.}})

cs.AI↗

arXiv2026

Look Ahead Before You Distill: Future Trajectory Validation of Teacher Guidance for Agentic On-Policy Distillation

On-policy distillation (OPD) provides teacher supervision on states visited by the student, reducing the distribution gap between training and inference. However, in multi-turn agentic tasks, student deviations may accumulate over time, gradually moving the trajectory away from states where teacher guidance remains effective. Our quantitative analysis further shows that high-disagreement states offer promising opportunities for teacher guidance, but determining whether such guidance is beneficial requires examining its effect on subsequent student trajectories. We propose FutureBridge-OPD (FTB), which executes a short teacher bridge at a high disagreement state and uses the resulting student continuation to assess whether the bridge increases the density of positive distillation signals relative to the teacher. On ALFWorld, WebShop, and ScienceWorld, under the main Qwen3-32B teacher to Qwen3-1.7B student setting, FTB outperforms vanilla OPD and TCOD by an average of 16.6 and 7.6 points, respectively, and remains effective across student scales and teacher settings. Our code is publicly available at https://github.com/ChenChiShui/FutureBridge-OPD.

cs.CL↗

arXiv2026

Multimodal Knowledge Edit-Scoped Generalization for Online Recursive MLLM Editing

Online multimodal knowledge editing requires injecting a continual stream of visual-textual corrections into multimodal large language models (MLLMs) with bounded overhead and minimal disruption to unrelated behaviors. Existing editors mainly emphasize edit reliability and long-horizon stability, but rarely control the semantic boundary of each edit. Our pilot analyses of post-edit behaviors and internal neuronal activities reveal a scope gap behind reliable edits: instance-level success neither guarantees transfer to valid cross-modal variants nor prevents leakage to unrelated inputs, while edit-related cross-modal responses concentrate in deeper semantic layers. Therefore, we formulate Edit-Scoped Generalization, reframing online MLLM editing from merely correcting an instance to controlling the propagation boundary of each edit. To this end, we propose ScopeEdit, a scope-aware online editor that decomposes each update into a modality-local absorption branch and an evidence-gated shared generalization branch. The local branch supports stable edit absorption, whereas the shared branch enables cross-modal propagation only when visual and textual evidence are sufficiently aligned. Both branches perform scope-separated write geometries in orthogonal low-rank spaces and maintain branch-wise preconditioners via Sherman--Morrison recursions, yielding constant per-edit overhead. Extensive experiments across diverse benchmarks, long-horizon edit streams, MLLM backbones, real-world VLKEB scenarios, and complex vision-language architectures show that ScopeEdit consistently improves the trade-off between in-scope cross-modal transfer and out-of-scope locality, while preserving edit reliability, stability and online efficiency. Our code is available at https://github.com/lab-klc/ScopeEdit.

cs.AI↗

arXiv2026

Large Language Models Do Not Always Need Readable Language

Large language models (LLMs) are commonly prompted and interfaced with human-readable natural language, even when the intended reader is another model. This paper investigates whether semantic information can be encoded in compact, non-standard textual forms that sacrifice human readability while remaining recoverable by LLMs. We refer to this class of model-centric textual representations as BabelTele, approached here not as a fixed protocol but as an empirical probe into LLMs' capacity to generate and interpret such representations. Through readability diagnostics, model likelihood measures, human questionnaires, and downstream task evaluations, we find that BabelTele can substantially depart from ordinary natural language while preserving core semantics for instruction-tuned LLMs. As a task-agnostic representational paradigm, BabelTele demonstrates high information density, maintaining 99.5% semantic fidelity even when the text volume is condensed to 27.9% of its original length. We further evaluate its semantic robustness in cross-model transfer, agent memory, and multi-agent communication. Results suggest that BabelTele can reduce context overhead while generally maintaining reliable downstream performance, although its effectiveness depends on the compressor-reader pair and task setting. These findings indicate that human readability, natural-language typicality, and model-side semantic recoverability can be partially decoupled, opening a path toward model-native representations in future exploration of LLM systems.

cs.CL↗

arXiv2026

Skill or Skip? Learning Selective Skill Invocation in Agentic Tasks via Dual-Granularity Preference Learning

Agent skills are callable procedural modules that provide reusable knowledge and execution policies for complex agentic tasks. However, existing methods mainly focus on selecting relevant skills or improving the skills themselves, while overlooking whether a relevant skill should actually be invoked at the current decision point. Unhelpful invocations may introduce irrelevant context and disrupt an otherwise correct execution process. To address this issue, we propose SelSkill, a dual-granularity preference-learning framework for selective skill invocation. SelSkill formulates skill use as a skill-or-skip decision, uses predictive uncertainty to prioritize candidate decision points, and constructs controlled invoke-skip preference pairs from shared trajectory prefixes. It further combines episode-level outcome preferences with step-level invocation preferences to capture both overall trajectory quality and the local effectiveness of skill invocation. On ALFWorld with Qwen3-8B, SelSkill improves task success by 10.9 points over the skill-enabled baseline and 7.8 over No-Skill, with 29.1-point higher execution precision. On BFCL, task success and execution precision improve by 5.7 and 29.5 points, respectively. Zero-shot results on Tau-bench and PopQA suggest partial transfer to unseen domains and skills.

cs.CL↗

arXiv2026

EvoStreaming: Your Offline Video Model Is a Natively Streaming Assistant

Streaming video understanding demands more than watching longer videos: assistants must decide when to speak in real time, balancing responsiveness against verbosity. Yet most video-language models (VideoLLMs) are trained for offline inference, and existing streaming benchmarks externalize this timing decision to the evaluator. We address this gap with RealStreamEval, a frame-level multi-turn evaluation protocol that exposes models to sequential observations and penalizes unnecessary responses. Under this protocol, we observed that strong offline VideoLLMs retain useful visual understanding but lack an interaction policy for deciding when to respond. Motivated by this observation, we propose EvoStreaming, a self-evolved streaming adaptation framework in which the base model itself acts as data generator, relevance annotator, and roll-out policy to synthesize streaming trajectories without external supervision. With only $1{,}000$ self-generated samples ($139\times$ less than the leading streaming instruction-tuning approach) and no architectural changes, EvoStreaming consistently improves the overall RealStreamEval score by up to $10.8$ points across five open VideoLLM backbones (Qwen2/2.5/3-VL, InternVL-3.5, MiniCPM-V4.5) while largely preserving offline video performance. These results suggest that data-efficient interaction tuning is a practical path for adapting existing VideoLLMs to streaming assistants.

cs.CV↗

arXiv2026

StreamMeCo: Long-Term Agent Memory Compression for Efficient Streaming Video Understanding

Vision agent memory has shown remarkable effectiveness in streaming video understanding. However, storing such memory for videos incurs substantial memory overhead, leading to high costs in both storage and computation. To address this issue, we propose StreamMeCo, an efficient Stream Agent Memory Compression framework. Specifically, based on the connectivity of the memory graph, StreamMeCo introduces edge-free minmax sampling for the isolated nodes and an edge-aware weight pruning for connected nodes, evicting the redundant memory nodes while maintaining the accuracy. In addition, we introduce a time-decay memory retrieval mechanism to further eliminate the performance degradation caused by memory compression. Extensive experiments on three challenging benchmark datasets (M3-Bench-robot, M3-Bench-web and Video-MME-Long) demonstrate that under 70% memory graph compression, StreamMeCo achieves a 1.87* speedup in memory retrieval while delivering an average accuracy improvement of 1.0%. Our code is available at https://github.com/Celina-love-sweet/StreamMeCo.

cs.CV↗

arXiv2026

Overcoming sensitivity-bandwidth trade-off in mid-infrared spectroscopy by a microresonator-anchored swept laser

Optical frequency combs have revolutionized high-precision spectroscopy, yet an intrinsic trade-off between spectroscopic signal-to-noise ratio (sSNR) and measurement bandwidth ($B$) fundamentally constrains sensitive, broadband measurements. While broadband swept lasers offer a potential solution, generating broadband, ultrafast and linearly sweeping lasers with a narrow linewidth remains a significant challenge, particularly in the fingerprint mid-infrared (mid-IR) band. Here we overcome this limitation by using a microresonator-anchored ultrafast sweeping Fourier domain mode-locked (FDML) laser for mid-IR spectroscopy. We introduce a dual-microresonator-anchor approach: a microcomb provides frequency calibration and a high-Q microresonator resolves the instantaneous FDML lasing lineshape. The strategy enables accurate correction of the FDML laser's sweep nonlinearity and broad linewidth in the near-IR, allowing the FDML laser to function as a high-fidelity mid-IR light via difference frequency generation. The system achieves a record sSNR$\times$$B$ of 1.3$\times$10$^5$ THz$\cdot \sqrt{\rm Hz}$ and methane sensing precision of 9 ppb$\cdot$m$\cdot$$\sqrt{\rm s}$, while retaining GHz resolution to distinguish methane isotope. We further demonstrate broadband, coherent swept laser phase spectroscopy in the mid-IR, tolerating losses up to 78 dB. This work leverages advances in integrated photonics to overcome the fundamental limitations of precision spectroscopy, paving the way for next-generation, broadband, and ultra-sensitive mid-IR spectroscopic sensing systems.

physics.optics↗

arXiv2026

High-Q AlN microresonators for nonlinear near-infrared and near-visible photonics

High Q-factors of microresonators are crucial for nonlinear integrated photonics, as many nonlinear dynamics have quadratic or even cubic dependence on Q-factors. The unique material properties make AlN microresonators invaluable for microcomb generation, Raman lasing and visible integrated photonics. However, the loss level of AlN falls behind other integrated platforms. By optimizing the fabrication, we demonstrate record Q-factors of 5.4$\times$10$^6$ and 2.2$\times$10$^6$ for AlN microresonators in the near-infrared and near-visible, respectively. Polarized-mode-interaction was used to create anomalous dispersion to support bright AlN Dirac solitons. Measurement of polarization-dependent spectra reveals the polarization hybridization of the Dirac soliton. In a microresonator with normal dispersion, Raman assisted four-wave-mixing (RFWM) was observed to initiate platicon formation, adding an approach to generate normal dispersion microcombs. A design of width-varying waveguides was used to ensure both efficient coupling and high Q-factor for racetrack microresonators at 780 nm. The microresonator was pumped to generate near-visble Raman laser at 820 nm with a fundamental linewidth narrower than 220 Hz. Our work unlocks new opportunities for integrated AlN photonics by improving Q-factors and uncovering nonlinear dynamics in AlN microresonators.

physics.optics↗

arXiv2025

FakeSV-VLM: Taming VLM for Detecting Fake Short-Video News via Progressive Mixture-Of-Experts Adapter

We present FakeSV-VLM in this paper, a new VLM-based framework for detecting fake news on short video platforms. Despite significant efforts to combat this issue due to the severe threat that fake news videos pose to public information security, existing methods still fall short in detection accuracy, often due to lack of knowledge to verify the news is real or not. However, large Vision Language Models (VLMs) have absorbed extensive real-world knowledge from massive multimodal datasets. Motivated by this, we adapt advanced VLMs for fake news detection in short videos. Upon close examination of news samples, we observe that short video samples can be categorized into four distinct scenarios: both video and text are real (for real samples), or both are fake, or either the video or text is fake (for fake samples). Inspired by this insight, we design four experts tailored to handle each scenario and integrate them into VLM via Mixture of Experts. Specifically, we develop the Progressive MoE Adapter (PMOE) module where detection experts first provide an initial analysis, followed by attribution experts for a comprehensive diagnosis, leading to a robust decision. Additionally, we also note the fake news videos often show inconsistency between two modalities. Consequently, we further design the Alignment-driven Event Checking (ADEC) module, which perceives the fake news by capturing the inconsistency between different modalities. Extensive experiments on two benchmark datasets, FakeSV and FakeTT, verify the superiority of our model. It significantly outperforms current state-of-the-art models by +3.32% and +5.02%, establishing a new benchmark in the field.

cs.MM↗

arXiv2025

Consistency-aware Fake Videos Detection on Short Video Platforms

This paper focuses to detect the fake news on the short video platforms. While significant research efforts have been devoted to this task with notable progress in recent years, current detection accuracy remains suboptimal due to the rapid evolution of content manipulation and generation technologies. Existing approaches typically employ a cross-modal fusion strategy that directly combines raw video data with metadata inputs before applying a classification layer. However, our empirical observations reveal a critical oversight: manipulated content frequently exhibits inter-modal inconsistencies that could serve as valuable discriminative features, yet remain underutilized in contemporary detection frameworks. Motivated by this insight, we propose a novel detection paradigm that explicitly identifies and leverages cross-modal contradictions as discriminative cues. Our approach consists of two core modules: Cross-modal Consistency Learning (CMCL) and Multi-modal Collaborative Diagnosis (MMCD). CMCL includes Pseudo-label Generation (PLG) and Cross-modal Consistency Diagnosis (CMCD). In PLG, a Multimodal Large Language Model is used to generate pseudo-labels for evaluating cross-modal semantic consistency. Then, CMCD extracts [CLS] tokens and computes cosine loss to quantify cross-modal inconsistencies. MMCD further integrates multimodal features through Multimodal Feature Fusion (MFF) and Probability Scores Fusion (PSF). MFF employs a co-attention mechanism to enhance semantic interactions across different modalities, while a Transformer is utilized for comprehensive feature fusion. Meanwhile, PSF further integrates the fake news probability scores obtained in the previous step. Extensive experiments on established benchmarks (FakeSV and FakeTT) demonstrate our model exhibits outstanding performance in Fake videos detection.

cs.CV↗

arXiv2025

3D Heterogeneous Integration of Silicon Nitride and Aluminum Nitride on Sapphire toward Ultra-wideband Photonics Integrated Circuits

Extending two-dimensional photonic integrated circuits (PICs) to three-dimensional (3D) configurations promises great potential for scaling up integration, enhancing functionality, and improving performance of PICs. Silicon-based 3D PICs have made substantial progress due to CMOS compatibility. However, the narrow bandgap of silicon (1.1 eV) limits their use in short-wavelength applications, such as chemical and biological sensing, underwater optical communications, and optical atomic clocks. In this work, we developed a 3D photonics platform by heterogeneously integrating silicon nitride (SiN) and aluminum nitride (AlN) PICs on sapphire (Al$_2$O$_3$). The broadband transparency of these materials allow our platform to operate over a multi-octave wavelength ranging from ultraviolet to infrared. Leveraging this platform, we demonstrated efficient optical nonlinearity in an AlN microcavity, low-loss and tunable SiN waveguide-based optical components, and optical linking between AlN and SiN PICs layers in the visible and near-infrared spectrum, hinting at potential applications in integrated quantum systems. Our work presents an ultra-wideband 3D PICs platform, providing new opportunities for broadband and short-wavelength applications of PICs.

physics.optics↗

arXiv2025

Self-injection locking dynamics with Raman actions in AlN microresonators

Self-injection locking (SIL) of semiconductor lasers to on-chip microcavities enables significant laser noise purification and diverse nonlinear optical actions. Realizing nonlinear SIL in new material platforms is essential for advancing photonic integrated circuits. Here, we demonstrate nonlinear SIL in AlN microcavities that generates stimulated Raman lasers (SRLs) and microcombs. We achieve SRL emission with an output power exceeding 10 mW and a fundamental linewidth below 70 Hz in the 1750 nm band. The Kerr effect further mediates stimulated emissions at the 2nd-Stokes and anti-Stokes frequencies. Multi-time-scale thermal relaxations during turnkey SIL enable GHz-level frequency sweeps of the SRL and pump. Raman actions also render a Stokes platicon microcomb state with co-emission in the pump and Stokes bands. Hybrid-integrated crystalline microresonators can be a versatile platform to investigate nonlinear photon-phonon interactions.

physics.optics↗

arXiv2021

On-chip parallel processing of quantum frequency combs for high-dimensional hyper-entanglement generation

High-dimensional encoding and hyper-entanglement are unique features that distinguish optical photons from other quantum information carriers, leading to improved system efficiency and novel quantum functions. However, the disparate requirements to control different optical degrees of freedom have prevent the development of complete integrated platforms that is capable of leveraging the complementary benefits of high-dimensional encoding and hyper-entanglement at the same time. Here we demonstrate the chip-scale solution to the generation and manipulation of high-dimensional hyper-entanglement. This is achieved by the parallel processing of multiple quantum frequency combs in the path domain. Cavity-enhanced parametric down-conversion with Sagnac configuration is implemented to ensure the spectral indistinguishability. Simultaneous entanglement in path and frequency is realized with high dimensions. On-chip reconfiguration of the entanglement structure is also demonstrated. We further present quantum interference in both entanglement degrees of freedom with high visibility. Our work provides the critical step for the efficient and parallel processing of quantum information with integrated photonics.

quant-ph↗

arXiv2020

Soliton dynamics in microresonators with XPM induced negative thermo-optic effect

Optical frequency comb generation in microresonators has attracted significant attention over the past decade, as it offers the promising potential for chip-scale optical frequency synthesis, optical clocks and precise optical spectroscopy. However, accessing temporal dissipative Kerr soliton (DKSs) is known to be severely hampered by thermal effects. Furthermore, due to the degeneracy of soliton existence range with respect to soliton number, deterministically accessing single soliton state is another challenge. Here, we demonstrate stable and deterministic single soliton generation in AlN-on-sapphire platform via auxiliary laser pumping scheme without the requirement of fast control of the pump power and detuning. Moreover, we reveal the underlying physics of soliton switching in a dual-pumped microcomb, which is fully described by the Lugiato - Lefever equation. The switching process is attributed to cross-phase modulation (XPM) induced degeneracy lifting of the soliton existence range, corresponding to an effective negative thermo-optic effect.

physics.optics↗

arXiv2020

BAlN for III-nitride UV light emitting diodes: undoped electron blocking layer

The undoped BAlN electron-blocking layer (EBL) is investigated to replace the conventional AlGaN EBL in light-emitting diodes (LEDs). Numerical studies of the impact of variously doped EBLs on the output characteristics of LEDs demonstrate that the LED performance shows heavy dependence on the p-doping level in the case of the AlGaN EBL, while it shows less dependence on the p-doping level for the BAlN EBL. As a result, we propose an undoped BAlN EBL for LEDs to avoid the p-doping issues, which a major technical challenge in the AlGaN EBL. Without doping, the proposed BAlN EBL structure still possesses a superior capacity in blocking electrons and improving hole injection compared with the AlGaN EBL having high doping. This study provides a feasible route to addressing electron leakage and insufficient hole injection issues when designing UV LED structures.

physics.app-ph↗

arXiv2019

Beyond 100 THz-spanning ultraviolet frequency combs in a non-centrosymmetric crystalline waveguide

Ultraviolet frequency combs enable applications ranging from precision spectroscopy to atomic clocks by addressing the electronic transitions of atoms and molecules. Access to ultraviolet light via integrated nonlinear optics is usually hampered by the strong material dispersion and large waveguide attention in the ultraviolet region. Here we demonstrate a simple route to chip-scale ultraviolet comb generators, simultaneously showing a gap-free frequency span of ~128 terahertz and supercontinuum sourced by an ultrafast fiber laser. The simultaneous cubic and quadratic nonlinear processes are implemented in single-crystalline aluminum nitride thin films, where a chirp-modulated taper waveguide is patterned to ensure a broad phase matching. The heterodyne beatnote characterization suggests that both the near-visible and ultraviolet supercontinuum combs maintain a high degree of coherence. Our approach is also adaptable to other non-centrosymmetric photonic platforms for ultrafast nonlinear optics with scalable bandwidth.

physics.optics↗

arXiv2019

AlN/beta-Ga2O3 based HEMT: a potential pathway to ultimate high power device

Gallium Oxide (Ga2O3) has a huge potential on the power device for its high breakdown filed and good transport properties. beta-Ga2O3 as the thermodynamics stable phase, has been demonstrated to form high electron mobility transistor (HEMT) through delta-doping in the barrier due to its none-polar property. Following the development in III-V HEMT which turns from delta-doping-induced to polarization-induced 2DEG, an alternative method based on III-N materials/beta-Ga2O3 heterostructure is proposed that utilizing the polarization difference on the interface. Further requirements of electric field and conduction band difference show that only nitrogen (N)-polar AlN on beta-Ga2O3 can form the channel and hold large 2DEG concentration on the interface. Compared with conventional metal-polar AlN/GaN HEMT, the proposed N-polar AlN/beta-Ga2O3 HEMT show a much larger 2DEG concentration, the spontaneous-polarization-dominated electric field, better DC output performance, as well as higher breakdown voltage. This study provides a new research approach that shifting from delta-doping-induced to polarization-induced on beta-Ga2O3-based HEMT, which can also be a guideline for community excavating the application potential of Ga2O3.

physics.app-ph↗