Generative recommendation (GR) has emerged as a widely adopted paradigm in industrial sequential recommendation. Current GR systems follow a similar pipeline: tokenization for item indexing, next-token prediction as the training objective and auto-regressive decoding for next-item generation. However, existing GR research mainly focuses on architecture design and empirical performance optimization, with few rigorous theoretical explanations for the working mechanism of auto-regressive next-token prediction in recommendation scenarios. In this work, we formally prove that \textbf{the k-token auto-regressive next-token prediction (AR-NTP) paradigm is strictly mathematically equivalent to full-item-vocabulary maximum likelihood estimation (FV-MLE)}, under the core premise of a bijective mapping between items and their corresponding k-token sequences. We further show that this equivalence holds for both cascaded and parallel tokenizations, the two most widely used schemes in industrial GR systems. Our result provides the first formal theoretical foundation for the dominant industrial GR paradigm, and offers principled guidance for future GR system optimization.

Jet quenching in heavy-ion collisions probes parton energy loss in the quark--gluon plasma (QGP), but the extracted transport properties may not be universally constrained across centrality, beam energy, and observable class. In this work, we perform an analysis of the compatibility and predictive transferability of Bayesian constraints obtained from a six-parameter JETSCAPE effective energy-loss model across these subsets. The model is calibrated to charged-hadron and inclusive-jet data from ALICE, ATLAS, and CMS in PbPb collisions at $\sqrt{s_{\mathrm{NN}}}=5.02$ and $2.76$ TeV. We find that centrality-dependent posteriors are largely compatible, whereas beam-energy and observable-class splits exhibit moderate shifts within overlapping credible regions, indicating that posterior overlap alone does not guarantee predictive universality. This is further examined by propagating subset posteriors to complementary datasets without refitting, where predictive performance varies across subsets. These results indicate that different observables probe distinct aspects of jet--medium interactions and motivate leading-hadron-selected jet observables to bridge hadron-biased and jet-inclusive constraints.

Bubble N68 in the G35 complex shows clear cloud-cloud collision (CCC) signatures. Its semi-ring-like morphology harbors many significant massive star formation tracers: 6 HII regions, 4 6.7 GHz masers, 5 Midcourse Space Experiment sources, 9 radio peaks, and nearly 10 O/B-type stars. We also identified 163 young stellar objects (45 Class I, 5 Flat, 113 Class II), indicating active star formation toward N68. Our molecular study with CO reveals two distinct molecular clouds (N68a: 47-56 km s$^{-1}$; N68b: 56-64 km s$^{-1}$), with broad bridge features and complementary distributions at their borders, indicating an ongoing CCC. Star formation in N68 is collectively driven by collect-and-collapse (CC), radiation-driven implosion (RDI), and CCC mechanisms. However, compared with the CC and RDI mechanisms, the CCC mechanism does not enhance the star formation efficiency; instead, it tends to trigger the formation of massive stars. N68, along with bubbles N65 and N61, constructs a $\sim100$ pc scale CCC system in the G35 complex.

Low earth orbit (LEO) satellites are a key technology to enable connectivity for rural and remote users. Communication satellites in LEO can provide coverage to much larger areas than terrestrial or aerial systems, while offering improved data rates when compared with geostationary systems. However, a major challenge with LEO satellite communications is the high mobility of the satellite, which results in a rapidly changing communication channel. Due to this, it is challenging to fairly allocate communication resources to multiple users in the system. This work proposes an Adaptive Power Allocation and Scheduling Scheme (APASS) to ensure user fairness in the downlink of a LEO satellite system serving mobile ground users. First, a novel channel and transmission model is introduced to capture the variability in channel statistics due to the satellite's trajectory. Then, a non-convex optimization problem is formulated to maximize the minimum rate across all ground users over a fixed set of time slots. To solve this problem, the proposed APASS dynamically allocates power and schedules transmissions based on predicted future channel gains. Numerical results show that APASS achieves strong performance even with substantial prediction errors, faring close to an upper bound that assumes perfect future channel knowledge. Furthermore, it improves the minimum user rate by a factor of 2.98 compared to equal-power allocation and maintains user fairness with a Jain's fairness index of well above 0.99.

Distributed LLM serving systems optimize per-request latency and throughput. However, under long-context workloads, inference accuracy becomes more variable. When incorrect responses trigger retries, accuracy directly translates into cumulative user-visible delay that is not captured by single-shot latency metrics. In this work, we argue that under long-context serving, \textbf{accuracy becomes speed} through retry dynamics. We introduce \textit{Time-to-Correct-Answer (TTCA)}, a metric that measures the wall-clock time required to obtain the first correct response. Our measurement study shows that prompt characteristics such as length and language amplify accuracy variance, which inflates TTCA. We demonstrate \textit{Lightweight Accuracy-Aware Routing (LAAR)}, a capability-based routing design that reduces TTCA. Our results suggest that in long-context distributed serving, accuracy should be treated as a first-class systems objective.

Blockchain technology enhances transparency by maintaining a distributed ledger among mutually untrusting parties. Despite its advantages, scalability and availability remain critical bottlenecks that hinder widespread adoption. The increasing complexity of blockchain nodes further necessitates robust fault tolerance and high throughput to ensure seamless operations. We present BlockRaFT, a crash-tolerant distributed framework designed to improve both the scalability and reliability of blockchain node operations. BlockRaFT framework utilizes RAFT consensus protocol to elect a leader within a cluster of systems. The elected leader coordinates and distributes workloads across follower nodes, thereby optimizing resource utilization and work load balancing. We analyzed the tasks performed by blockchain nodes and partition them according to their stateful and stateless characteristics. Stateless operations are centralized at the leader, while stateful operations are replicated and coordinated across the cluster to ensure consistency and fault tolerance. We evaluate whether this distributed intra-node architecture provides measurable benefits over traditional single-node execution models in terms of scalability, availability, and performance. Additionally, we introduce a concurrent Merkle tree optimization that decouples smart contract execution from tree updates, significantly reducing one of the significant performance overheads in blockchain systems. Our design philosophy is rooted in utilizing the well-established principles of distributed computing and customizing them for the blockchain domain rather than reinventing them.

The charged current $ν_μ(\barν_μ)$-induced deep inelastic scattering (DIS) from an $^{40}\mathrm{Ar}$ target is studied using a microscopic framework that incorporates nuclear medium effects due to Fermi motion, binding energy, nucleon correlations, mesonic ($π$ and $ρ$) contributions, and nuclear shadowing and antishadowing across the relevant Bjorken-$x$ region. The nuclear structure functions $F_{iA}(x,Q^2)$ $(i=1\text{-}3)$ are evaluated using a relativistic nucleon spectral function ($S_h$) within the local density approximation employing the free nucleon structure functions, $F_{iN}(x,Q^2)$ $(i=1\text{-}3)$. These $F_{iN}(x,Q^2)$ $(i=1\text{-}3)$ are calculated using parton distribution functions (PDFs) from MMHT 2014 parameterization, including higher-order perturbative QCD corrections up to next-to-next-to-leading order (NNLO), along with nonperturbative target mass corrections (TMC). The resulting nuclear structure functions $F_{iA}(x,Q^2)$ $(i=1\text{-}3)$ are subsequently used to compute the differential DIS cross sections for $^{40}Ar$ nucleus. Numerical results are presented for $ν_μ(\barν_μ)$ beam energies $E=4$ GeV and $E=6$ GeV for the differential scattering cross sections $\frac{d^2σ}{dx dy}$ and $\frac{dσ}{dx}$, relevant to ongoing and upcoming liquid-argon neutrino experiments such as DUNE and the Fermilab Short-Baseline Neutrino program.

For the genus-$4$ Heegaard surface in the $3$-sphere, we present a sufficient condition for a non-separating weak reducing pair to be separated by a reducing sphere for the surface. As a consequence, we reduce the connectivity problem in the reducing sphere complex for the surface to the problem of showing that any two vertices, whose representative reducing spheres are disjoint from a fixed non-separating compressing disk for the surface, are connected in the complex.

In this paper, we analyze the double Wick rotation of a rotating BTZ black hole and the entanglement entropy. We derive the transition matrix dual to the double Wick-rotated BTZ black hole, which has the usual shape at an imaginary chemical potential. In the dual gravity side, the double Wick rotated BTZ black hole, which is obtained as a quotient, is equal to a rotating BTZ black hole after the coordinate transformation and the identification of periodicity. The geometric entropy and time-like entanglement entropy are reproduced by the identification. New Lorentzian entanglement growth is defined by the coefficient of linear growth of time-like entanglement entropy.

A central goal of LLM alignment is to balance helpfulness with harmlessness, yet these objectives conflict when the same knowledge serves both legitimate and malicious purposes. This tension is amplified by context-sensitive alignment: we observe that domain-specific contexts (e.g., chemistry) selectively relax defenses for domain-relevant harmful knowledge, while safety-research contexts (e.g., jailbreak studies) trigger broader relaxation spanning all harm categories. To systematically exploit this vulnerability, we propose Jargon, a framework combining safety-research contexts with multi-turn adversarial interactions that achieves attack success rates exceeding 93% across seven frontier models, including GPT-5.2, Claude-4.5, and Gemini-3, substantially outperforming existing methods. Activation space analysis reveals that Jargon queries occupy an intermediate region between benign and harmful inputs, a gray zone where refusal decisions become unreliable. To mitigate this vulnerability, we design a policy-guided safeguard that steers models toward helpful yet harmless responses, and internalize this capability through alignment fine-tuning, reducing attack success rates while preserving helpfulness.

Biochemical signalling cascades transduce extracellular stimuli into cellular responses through sequences of discrete, node-to-node activations. While signal fidelity depends critically on local interaction kinetics, the mechanisms governing information propagation in realistic, highly variable kinetic contexts remain poorly understood. In this paper, we develop a mathematical framework for travelling waves in canonical feed-forward pathways governed by nonlinear Michaelis-Menten-type kinetics. For uniform pathways, we characterise the complete steady-state landscape and demonstrate that activation bias (the contribution of the binary states of each node to downstream activation) between connected nodes acts as a key bifurcation parameter dictating wave existence. Extending this framework to heterogeneous networks, we show how parameter gradients and random kinetic variations distort wavefronts and induce heavy fluctuations in propagation speed. To recover predictable signal transmission, we introduce a novel reciprocal-velocity spatial rescaling technique. We demonstrate that this coordinate transformation inherently absorbs local kinetic variations, effectively smoothing wave velocities and preserving wavefront profiles without requiring bespoke parameter tuning or continuous limits. Finally, by testing the framework's limits against extreme parameter variability, we reveal how severe kinetic bottlenecks lead to functional pathway fragmentation, offering a mathematically justified basis for rational model reduction in complex biochemical networks.

Disentanglement techniques used in collaborative filtering uncover interaction intents between nodes, improving the interpretability of node representations and enhancing recommendation performance. However, existing disentanglement methods still face two problems. First, they focus on local structural features derived from direct node interactions and overlook the comprehensive graph structure, which limits disentanglement accuracy. Second, the disentanglement process depends on backpropagation signals derived from recommendation tasks and lacks direct supervision, which may lead to biases and overfitting. To address these issues, we propose the Intent Propagation Contrastive Collaborative Filtering (IPCCF) algorithm. Specifically, we design a double helix message propagation framework to more effectively extract the deep semantic information of nodes, thereby improving the model's understanding of interactions between nodes. We also develop an intent message propagation method that incorporates graph structure information into the disentanglement process, thereby expanding the consideration scope of disentanglement. In addition, contrastive learning techniques are employed to align node representations derived from structure and intents, providing direct supervision for the disentanglement process, mitigating biases, and enhancing the model's robustness to overfitting. Experiments on three real data graphs illustrate the superiority of the proposed approach.

We derive explicit strong-field asymptotics for the normalized differential resistance in Hall field-induced resistance oscillations (HIRO) within the Vavilov-Aleiner-Glazman kinetic framework. For a single-harmonic density of states, the leading oscillation amplitude is set by the full backscattering rate $1/τ(π)$. Extending the theory to a two-harmonic density of states, we show that the off-diagonal mixed kernel $γ_{12}$ admits an exact single-integral representation, from which the strong-field asymptotics follow directly. The resulting odd harmonics, notably $m=1$ and $m=3$, have coefficients determined by combinations of $1/τ(0)$ and $1/τ(π)$, while the leading $m=2$ amplitude remains unchanged. On exact-kernel mock data generated and fit within the same model, with $τ_{\rm tr}$ and $τ_{\rm in}$ held fixed, the resulting extraction protocol recovers $τ_q$, $τ(π)$, and -- when the $m=1,3$ harmonics are resolved -- $τ(0)$ to sub-percent accuracy, with $τ(0)$ providing a consistency check on the disorder description.

We propose a testing and estimation methodology for univariate and bivariate symmatric $α$-stable distributions using a modified version of the Greenwood statistic. Originally designed for positive-valued random variables, the Greenwood statistic, and its modified version tailored for symmetric distributions, have been predominantly applied to univariate random samples. In this paper, we extend the modified Greenwood statistic to a bivariate setting and examine its probabilistic properties within the class of $α$-stable distributions, with a focus on the sub-Gaussian case. Additionally, we introduce a novel testing approach that considers two variations of the modified Greenwood statistic as test statistics for the bivariate case. In the univariate setting, we adapt the proposed testing methodology for estimating the stability index. The simulation studies presented demonstrate that our proposed methodology outperforms classical approaches previously used in this context and serves as an effective tool for distinguishing between Gaussian and $α$-stable distributions with a stability index close to 2. The theoretical and simulation results are further illustrated with practical data examples.

We present a comprehensive study of current-driven dynamics, transformations, and instabilities of skyrmioniums in chiral magnetic films, considering both isolated objects and collective states forming skyrmionium-based meta-matter. Using micromagnetic simulations combined with an analytical description based on the generalized Thiele equation, we clarify how the internal structure of skyrmioniums governs their nonequilibrium response to electric currents. Despite having zero total topological charge, skyrmioniums exhibit a finite transverse velocity under applied currents. We show that this skyrmionium Hall effect originates from an imbalance between positive and negative topological contributions of the inner skyrmion and surrounding ring, which typically occupy different areas. Current-induced deformations further enhance this imbalance, yielding Hall angles comparable to those of skyrmions. At higher current densities, skyrmioniums undergo distinct instabilities depending on magnetic field and uniaxial anisotropy, including elongation, collapse into a skyrmion, transformation into a topologically trivial droplet, and expansion into stripe textures. We map these regimes in current--field and current--anisotropy phase diagrams and resolve their microscopic pathways via the evolution of topological charge and local rotational measures. Beyond isolated textures, mixed skyrmion--skyrmionium lattices display rich collective dynamics, including elastic transport, polymorphic transitions, soliton exchange, and stripe formation. Pulsed currents provide additional control, enabling access to regimes beyond continuous driving. Our results establish skyrmioniums and their meta-matter as tunable nonequilibrium systems probing the topological energy landscape far from equilibrium.

In this paper we consider families of mutually commuting endomorphisms of the generalized tangent bundle. We identify natural tensorial constraints extending the notion of a generalized Kähler structure to endomorphisms that are not necessarily generalized almost complex structures. These tensors form ideals whose generators we explicitly construct and study using Gröbner basis techniques.

In this workshop, we discuss several algorithms for mathematical programs with equilibrium constraints (MPECs). The unifying theme is that MPECs are optimization problems whose feasible set contains a lower-level equilibrium system, often written through complementarity or variational-inequality conditions. This destroys the smooth manifold or convex structure that standard nonlinear programming methods rely on. We focus on four algorithmic viewpoints: (i) the classical penalty interior-point algorithm (PIPA); (ii) a monotone-linear complementarity problem (LCP) variant of PIPA that explicitly controls complementarity decay; (iii) an implicit-programming descent method for variational inequality (VI)-constrained MPECs; (iv) piecewise SQP (PSQP), which applies SQP on locally selected smooth pieces. For each method we explain the model, the search direction subproblem, the globalization mechanism, and the meaning of the convergence result. Particular emphasis is placed on what the assumptions are really doing and on the distinction between an attractive algorithmic idea and a fully valid convergence theorem.

Radiation-driven winds of massive stars can be described within the modified CAK theory, which parametrises the radiation force through three key quantities: $α$, $δ$, and $k$. Different combinations of these parameters, together with rotation, result in three types of stationary solutions, namely fast (or classical), $δ$-slow, and $Ω$-slow solutions. The primary objective of this work is to model radiation-driven winds inside the gap region between the fast and $δ$-slow regimes, where stationary solutions have proven elusive. In addition, we compute synthetic line profiles of H I, He I, and Si IV to illustrate the morphology of different wind regimes. We employ the time-dependent hydrodynamic code ZEUS-3D, capable of obtaining stationary solutions by progressing through an initial solution. Then we compute the line profiles solving the transfer equation for an expanding atmosphere, assuming spherical symmetry in the comoving frame, under non-local thermodynamic equilibrium (NLTE) conditions. We found new stationary solutions in the gap region, alongside their corresponding line profiles, for a typical B supergiant star model. In this model, the new solutions are stable, and some of them present a kink in the velocity profile at a fixed distance from the star, depending on the $δ$ value. Perturbations in the wind ionisation may trigger transitions between different hydrodynamic regimes and offer a plausible explanation for structured and variable winds. A systematic investigation of these effects will be the subject of future work. Furthermore, we investigate the resulting line profiles from different hydrodynamic solutions and compare them with those predicted by a velocity profile given by a $β$-law using the same global wind parameters.

This paper introduces a high-precision indoor positioning and tracking method that utilizes multi-site single-input single-output (SISO) radar systems. We propose a novel velocity synthesis-assisted (VSA) localization algorithm that iteratively refines target position estimates within range bins by fusing radial velocity measurements from multiple radars. This approach ensures enhanced accuracy in both velocity and position estimation. Moreover, the inherent geometric constraints introduced by velocity synthesis enable the proposed algorithm to remain robust under low signal-to-noise ratio (SNR), severe multipath propagation, and large synchronization latency. Notably, our method eliminates the use of multiple-input-multiple-output (MIMO) configurations and stringent phase synchronization requirements, substantially reducing hardware complexity while maintaining high positioning accuracy. We define standardized reference trajectories to facilitate a comprehensive and reproducible performance evaluation. Extensive simulations and experimental validations demonstrate that our multi-site radar systems achieve centimeter-level tracking accuracy for human subjects, outperforming existing methods in complex trajectory tracking.

This paper addresses the classic problem of parameter estimation (PE) in multimachine power system models. Such models are typically described by a set of nonlinear differential-algebraic equations (DAE), where generator physics and network power flow equations are coupled. DAE models are well established in classic power system textbooks, but parameter identification and estimation of generator inertia and damping together with network branch resistances and reactances for these models remain relatively underexplored. In contrast to prior approaches that rely on ODE approximations, this paper develops a joint Bayesian inference framework to perform PE of generator and network parameters while exploiting grid DAE models. It further combines physics-aware statistical modeling with computationally efficient posterior sampling to make joint Bayesian calibration practical. Results on the IEEE 9-bus system show accurate parameter recovery with well-behaved posterior uncertainty, while a short 39-bus study provides evidence that the framework remains effective on a materially larger joint-estimation problem. These results are obtained without requiring overly conservative priors.

Following our recent numerical study [arXiv:2601.16299 (2026)], we investigate vibrational excitation induced by transient optical driving in molecular ensembles strongly coupled to a cavity mode using the field-driven Holstein--Tavis--Cummings model. We analyze how pulsed excitation redistributes energy among electronic, photonic, and vibrational degrees of freedom in molecular polaritons. Vibrational dynamics are examined over a broad range of pulse durations and intensities within both the single-excitation approximation and a mean-field description of collective light--matter coupling. Despite their distinct formulations and microscopic descriptions, these two approaches yield consistent scaling relations for vibrational excitation. In particular, we disentangle linear and nonlinear contributions to vibrational excitation, which are reflected in distinct quadratic and quartic scaling behaviors with respect to the driving field amplitude (that is, linear and quadratic dependence on the incident pulse intensity). The microscopic origin of the nonlinear component is identified as a polariton-mediated intrapulse stimulated Raman-like process, enabled by a pulse spectral bandwidth large enough to overlap both upper and lower polaritons (rather than a conventional multi-pulse scheme). These results establish a unified framework for understanding vibrational excitation under pulsed polariton driving and provide guidance for the interpretation and control of ultrafast polariton experiments. Discrepancies between the mean-field and single-excitation approaches under certain pulsed conditions are identified and analyzed.

Motivated by recent theoretical and experimental studies on the role of flatbands in the thermoelectric properties of Ni$_3$In$_{1-x}$Sn$_x$ compounds, we investigate electron transport in two minimal one-dimensional flatband models, the sawtooth and diamond chains, which differ in a crucial aspect: the flatband is separated from the dispersive band by a finite gap in the former, while the two bands touch in the latter. Using a non-equilibrium Green function framework with interactions treated at the Hartree-Fock and GW levels, we compute the full set of thermoelectric coefficients and the figure of merit $zT$ as functions of gate voltage and temperature. We show that, contrary to naive expectation, a perfectly isolated flat-band is a physically ill-founded thermoelectric: the electrical conductivity vanishes as the chemical potential enters the flat-band, rendering the large Seebeck coefficient and the apparent violation of the Wiedemann-Franz law physically meaningless. Optimal thermoelectric performance is instead achieved just below the flat-band edge, where the transmission function varies most rapidly with energy, consistent with the Mahan-Sofo picture, and requires a finite broadening of the flat-band through hybridization with dispersive states. We further show that electron-electron interactions renormalize the flat-band structure itself, inducing an interaction-driven narrowing of the bandwidth and, in the diamond chain, a correlation-induced opening of a gap between the flat-band and the dispersive band near half-filling. Mean-field treatments are found to systematically overestimate \(zT\), highlighting the importance of beyond-mean-field correlations for quantitatively reliable predictions in flat-band thermoelectrics.

We study a finite time horizon Markov decision process (MDP) consisting of several groups of multi-action finite-state restless bandit processes, which are identical within each group. The bandit processes into different groups can be rather different. The bandit processes are subject to multiple weakly coupled constraints on their state and action variables. In contrast to prior studies that considered only a few specific policies/algorithms, here, we study the behaviours of the general stochastic process and, most importantly, the design of policies that guarantee its convergence to an ideal trajectory as the problem size increases. We prove that, for any policy in a rather general class, the resulting stochastic process converges in probability to a deterministic process as the system size (measured by the number of bandits) tends to infinity, at an exponential rate. Unlike the previous proofs, our exponential convergence does not rely on any non-degenerate assumptions. It follows that the chosen policy asymptotically approaches optimality (with exponentially diminishing suboptimality) in the size dimension if and only if the deterministic process coincides with optimality. We further propose a policy and prove that, in general, it converges in probability to optimality exponentially fast in the system size.

Fungal protein materials exhibit inherently anisotropic microstructures formed by networks of hyphae, which suggest a natural pathway to replicate the fibrous texture of animal meat. We probe whether this structural anisotropy translates into macroscopic mechanical and sensory anisotropy. Using orthogonal tension, compression, and shear experiments on three fungi-based materials, we identify distinct symmetry classes that range from strongly anisotropic to effectively isotropic behavior. Automated model discovery reveals that fiber-dependent invariants emerge only when mechanically relevant, and enables direct identification of material symmetry from data. These results demonstrate that microstructural anisotropy does not universally imply anisotropic mechanics or perception and establish a data-driven framework to infer symmetry in complex soft materials.

With the push toward 6G commercialization, Frequency Range 3 (FR3) bands, specifically 7.125-8.4 GHz and 14.8-15.3 GHz, have become focal points for achieving wide-area, high-capacity coverage. However, practical deployment is often limited by the physical aperture constraints of base station antennas. This study conducts comprehensive measurements in Urban Macro (UMa) scenarios using a unified dual-band sounding platform to evaluate channel characteristics and system performance under the strict constraint of "equal physical array aperture." The results indicate that higher frequency bands exhibit increased sparsity in both delay and spatial domains. Regarding coverage, while the 15 GHz band can theoretically accommodate four times the number of antenna elements (128 elements) within the same area to compensate for path loss, empirical data reveals a residual coverage deficit of approximately 3.0 dB at cell edges compared to the 8 GHz baseline. In contrast, the 15 GHz band excels in capacity; the increased element count effectively overcomes channel sparsity, resulting in spectral efficiency (SE) that significantly outperforms the 8 GHz band. Furthermore, the research demonstrates that for a fixed number of elements, system performance remains largely insensitive to specific array topologies (e.g., 1x32, 2x16, or 4x8). Ultimately, FR3 system performance is dictated by the trade-off between propagation characteristics and hardware-enabled gain. These findings provide a theoretical foundation for spatial-domain design and help address engineering challenges for 6G base station implementation

Website fingerprinting (WF) attacks infer the websites visited by users from encrypted traffic in anonymous networks such as Tor. Existing deep learning methods achieve high accuracy under the single-tab assumption but degrade substantially when users open multiple tabs concurrently, producing interleaved traffic that transforms WF into an implicit demixing problem. We identify three structural requirements for effective multi-tab demixing, namely signal integrity at segment boundaries, multi-scale local modeling, and relative temporal association of dispersed fragments, and show that no prior method satisfies all three simultaneously. We propose DEMUX, a designed framework that addresses these requirements through three tightly coupled components. A Boundary Preserving Aggregation Module employs overlapping window partitioning with joint packet-level and burst-level feature extraction. A Multi-Scale Parallel CNN captures heterogeneous temporal patterns via parallel branches. A two-stage Transformer encoder with Rotary Positional Embedding enables robust cross-window fragment association. The Boundary Preserving Aggregation Module additionally serves as a plug-and-play preprocessor that consistently improves existing baselines without architectural modification. Extensive experiments across closed-world, open-world, defense-augmented, dynamic-tab, and cross-configuration settings demonstrate that DEMUX achieves state-of-the-art performance. In the challenging closed-world 5-tab setting, DEMUX attains a P@5 of 0.943 and MAP@5 of 0.961, outperforming the strongest baseline by 9.2 and 6.2 percentage points respectively, confirming its strong robustness in complex multi-tab demixing scenarios.

Knowledge Graph-based Retrieval-Augmented Generation (KG-RAG) has emerged as a promising paradigm for enhancing LLM reasoning by retrieving multi-hop paths from KGs. However, existing KG-RAG frameworks often underperform in real-world scenarios because the pre-captured knowledge dependencies are not tailored to the downstream task or its evolving requirements. These frameworks struggle to adapt to task-specific requirements and lack mechanisms to filter low-contribution knowledge during generation. We observe that feedback on generated responses offers effective supervision for improving KG quality, as it directly reflects user expectations and provides insights into the correctness and usefulness of the output. However, a key challenge lies in effectively linking response-level feedback to triplet-level contribution evaluation and knowledge updates in the KG. In this work, we propose EvoRAG, a self-evolving KG-RAG framework that leverages the feedback over generated responses to continuously refine the KG and enhance reasoning accuracy. EvoRAG introduces a feedback-driven backpropagation mechanism that attributes feedback to retrieved paths by measuring their utility for response and propagates this utility back to individual triplets, supporting fine-grained KG refinements towards more adaptive and accurate reasoning. Through EvoRAG, we establish a closed loop that couples feedback, LLM, and graph data, continuously enhancing the performance and robustness in real-world scenarios. Experimental results show that EvoRAG improves reasoning accuracy by $7.34\%$ over state-of-the-art KG-RAG frameworks. The source code has been made available at https://github.com/iDC-NEU/EvoRAG.

Web3 prediction markets, exemplified by Polymarket, have gained prominence for leveraging collective intelligence to forecast a wide range of social, political, and sports events. However, among the thousands of prediction market events, consensus disputes still arise due to imperfections in market mechanisms. On Polymarket alone, the trading volume involving disputed events has reached $972,370,804.71, underscoring the critical need for objective and efficient dispute resolution. In this study, we introduce large language models (LLMs) to: (1) evaluate whether web-enabled LLMs can reproduce the decision quality of UMA's on-chain voting process once a dispute has been raised, and (2) predict, based on event rules, which market events are likely to face future disputes before they occur. Our findings show that LLMs are unable to reliably predict which events will become disputed in advance; however, once a dispute is initiated, web-enabled LLMs achieve 89.58% agreement with UMA's final resolutions and demonstrate strong stability.

Self-organized criticality (SOC) is widely proposed as a fundamental mechanism for collective behavior, yet its role in objective-driven, heterogeneous adaptive systems underpinning real complex systems remains less understood. We introduce EvoSK, a minimal evolutionary model in which agents perform memory dependent reinforcement learning on a rugged Sherrington-Kirkpatrick landscape while the population evolves through extremal replacement of the least fit agents. We demonstrate that this coupled dynamics drives the system to a critical state residing on the transition boundary between ergodic and non-ergodic phases. At this boundary, the system exhibits scale-free evolutionary avalanches with a mean-field exponent $τ\approx -1.5$, while simultaneously achieving collective rewards that surpass those of any manually finetuned, non-evolutionary regime. Our results provide a mechanistic link between the statistical physics of ergodicity breaking and the functional optimality of complex adaptive systems, suggesting that the edge of ergodicity breaking acts as a robust attractor for systems adapting on rugged, high-dimensional landscapes.

sPHENIX is a state-of-the-art experiment at the Relativistic Heavy Ion Collider (RHIC), dedicated to the study of heavy-flavor and jet physics. Its precision tracking system, combined with streaming readout, enables heavy-flavor measurements with high-statistics and essentially unbiased data samples. During the 2024 run, sPHENIX was fully commissioned and recorded a sample of 100 billion unbiased $p$+$p$ collisions, together with a minimum-bias Au+Au dataset. The 2025 run further expanded the sPHENIX dataset with high-statistics $p$+$p$, O+O and Au+Au collisions. This extensive $p$+$p$ sample opens the door to heavy-flavor measurements with orders of magnitude more statistics than previously available at RHIC. Notably, there has been no prior measurement of the $Λ_c^+ / D^0$ baseline in $p$+$p$ collisions at RHIC energies. The large sPHENIX dataset now enables the first exploration of key open questions, such as the hadronization mechanism of baryons and the strange-to-light flavor meson ratio.