arXiv:2606.14123v1 Announce Type: new Abstract: Deployed knowledge-tracing models are typically frozen after training, yet systematic per-item logit bias arises, from limited per-item expressivity in backbone architectures and from post-deployment shifts in item properties, degrading prediction quality. Global post-hoc calibrators such as Platt scaling, temperature scaling, and isotonic regression improve probability estimates but leave discriminative ability, as measured by AUC, unchanged. This AUC invariance is a structural consequence of monotone score-only transforms; recovering the stranded discrimination requires conditioning on item identity. We propose SLC (State-space Logit Correction), which converts binary observations to Gaussian pseudo-observations via Laplace/IRLS, applies empirical-Bayes shrinkage through a Kalman smoother, and fits an offset-Platt link. The state-space formulation also yields a detectability bound that characterizes the Bernoulli information floor, explaining why temporal tracking provides no benefit at current data densities. Across four datasets, five backbones, and three seeds, SLC improves AUC on all four datasets and NLL on three, with the advantage concentrating on sparse items. Cross-domain controls suggest that the same phenomenon can arise beyond education when the deployed backbone leaves entity-level bias.
Science Journals
arXiv:2606.14129v1 Announce Type: new Abstract: Reconstruction-based anomaly detection is attractive for industrial inspection, but scaling it from category-specific training to a one-for-all setting is challenging. A single model must reconstruct diverse normal appearances without copying abnormal details, which exposes two coupled failure modes: identical shortcut, where anomalies pass through the reconstruction path, and mis-reconstruction, where normal categories are confused with one another. We propose \textbf{BoRAD}, a label-free training framework that treats this as a representation-capacity allocation problem. BoRAD uses a shared learnable prototype bank to impose two complementary regularizers: spatial prototype alignment contracts local within-prototype variation to suppress anomaly copying, while prototype-relative global alignment preserves between-prototype structure and improves sensitivity to abnormal angular deviations. The prototype bank and prediction heads are used only during training; inference remains a standard teacher-student feature discrepancy pass, with no class labels, negative pairs, memory retrieval, or prototype lookup. BoRAD achieves competitive one-for-all anomaly detection performance, including 86.2\% mAD on MVTec AD, 80.7\% mAD on VisA and 73.1\% mAD on Real-IAD. Diagnostic analyses further show reduced anomaly leakage, improved normal-category separability, and stronger anomaly-normal score separation.
arXiv:2606.14138v1 Announce Type: new Abstract: Large heat flux is commonly associated with vigorous convection and turbulent mixing. Here, we show that this connection is not fundamental. Using a marginal energy-stability theory, we identify near-optimal convective heat transport with the saturation of an energy-stability constraint rather than with turbulence intensity. The theory selects mean temperature profiles whose fluxes closely approach the best available optimal transport states and rigorous upper bounds, predicting the asymptotic scaling $Nu\sim0.0245Ra^{1/2}$ at large Rayleigh number. These profiles exhibit a hierarchical structure consisting of conductive inner layers, logarithmic-like intermediate layers, and a stably stratified bulk, closely mirroring optimal transport calculations and suggesting that maximal convective heat transport emerges near marginal energy stability. More strikingly, the same profiles can be converted into exact conductive states through prescribed internal thermal forcing. Direct numerical simulations show that an initially turbulent flow then relaxes to a motionless state while maintaining a large wall heat flux. Energy-stability saturation therefore provides both a physical interpretation of transport limits and a route to high-flux heat transfer without turbulence.
arXiv:2606.14139v1 Announce Type: new Abstract: Full waveform inversion (FWI) recovers subsurface velocity from seismic recordings by solving a severely ill-posed, nonconvex PDE-constrained optimization. Classical regularizers stabilize the inversion but fail to reproduce realistic geological structures; recent diffusion-prior methods improve realism at the cost of a fragile trade-off between data fidelity and prior consistency. We propose Decoupled Latent Optimization (DLO), which relaxes the standard latent-optimization formulation into a quadratic-penalty objective over an auxiliary physical variable and a latent variable. The data-fidelity gradient acts in physical space, the diffusion sampler contributes only through a decoded prior sample, and the standard smoothed-velocity initialization of classical FWI is preserved. On the OpenFWI benchmark, DLO outperforms classical regularizers and existing diffusion-based methods under clean, noisy, and missing-trace acquisitions. The prior, trained on 70*70 OpenFWI models, transfers directly to the Marmousi and Overthrust benchmarks, where DLO recovers intricate fault structures and remains robust to initialization smoothing and measurement noise.
arXiv:2606.14141v1 Announce Type: new Abstract: Sound events are entities with semantic identities, locations, and trajectories, but current audio-language models usually reason about clips as global event content. Conversely, sound event localization models track source directions over time but offer limited semantic coverage for language reasoning. To address this gap, we introduce ST-AudioQA, a spatio-temporal audio QA dataset and benchmark built from first-order ambisonic (FOA) renderings of static and moving sound sources. Each scene provides source identity, activity, direction, distance, and motion metadata, enabling dense trajectory supervision and questions about what is sounding, where it is, how it moves, and how sources relate. We further propose ST-Audio Encoder, a time-resolved FOA audio encoder that learns event semantics together with source trajectories, and ST-AudioLM, which connects the audio tokens from the encoder to an LLM for spatio-temporal audio QA. Experiments show that this representation improves the semantic-localization tradeoff and yields stronger reasoning performance than static spatial and localization-oriented baselines.
arXiv:2606.14580v1 Announce Type: new Abstract: Identifying persuasive rhetorical cues is critical across domains, from detecting information manipulation and improving AI safety to advancing public health communication. We propose Persuasion Index (PI), a taxonomy of 15 dimensions grounded in persuasion theories from psychology and communication, and one transparent implementation using 55 sub-features built from lexicons and rule-based detectors. The taxonomy is modular: individual detectors can be replaced while preserving the theoretical structure. By evaluating PI on four public datasets varying in domain, style, and outcome measures, we show that PI provides a shared feature space for interpreting rhetorical patterns associated with persuasion-related outcomes. Linear models show that PI features carry meaningful predictive signal while remaining computationally lightweight. Dimension-level analyses reveal recurring associations between PI dimensions and persuasion outcomes across datasets, while also highlighting topic- and stance-specific variation. We release PI as an open-source package and web interface for principled and auditable analysis of human and AI-mediated communication.
arXiv:2606.14156v1 Announce Type: new Abstract: Learning systems based on IF-THEN rule representations readily offer interpretability, making them a crucial focus in contemporary AI research. A key objective for such rule sets is to achieve both high discriminative power and interpretability. While existing state-of-the-art algorithms implicitly prioritize predictive accuracy, they often fall short on one or more quality metrics that ensure interpretability, such as coverage and parsimony of rule sets. Motivated by this, this paper propose the development of CDPR, which aims to create highly accurate and interpretable rule sets for classification problems. To the best of our knowledge, this represents the first attempt to establish such an approach. In this study, we introduce two algorithms rooted in submodular maximization, which not only provide provable guarantees on coverage but also yield rule sets that are both discriminative and parsimonious. We empirically demonstrate that rule sets learned through our approaches achieve higher accuracy and interpretability and has more than a 2.5-fold improvement in average coverage rates when compared to the next best algorithm.
arXiv:2606.14159v1 Announce Type: new Abstract: Protein-ligand binding affinity (PLA) prediction is critical in drug discovery. Despite the notable advancements in machine learning-based approaches, existing methods struggle to jointly characterize local geometric organization and globally coordinated cross-molecular interactions, limiting their ability to model complex binding mechanisms. Here, we propose RicciBind, a geometric representation framework that integrates curvature-guided hierarchical structure learning with optimal transport (OT)-based cross-domain alignment to model molecular interactions. Specifically, RicciBind leverages Ricci curvature to capture local interaction tightness within molecular structures, enhancing structural awareness and organizing atomic interactions into curvature-aware hierarchical representations. An OT-based cluster matching mechanism then aligns protein and ligand clusters across heterogeneous domains under geometric constraints, enabling globally consistent correspondences and revealing higher-order interaction patterns beyond local neighborhoods. By coupling curvature-guided structure encoding with OT-driven cross-domain alignment, RicciBind effectively models complex interaction semantics and substantially improves both the accuracy and interpretability of binding affinity prediction. Extensive experiments demonstrate that RicciBind achieved superior predictive performance and generalization across PLA benchmarks and virtual screening tasks. Ablation studies further confirmed the essential role of Ricci curvature in enhancing molecular interaction representations.
arXiv:2606.14162v1 Announce Type: new Abstract: Large-scale video diffusion models often fail to preserve 3D structure over time, causing geometric drift and implausible motion under viewpoint changes. Existing methods usually enforce geometric consistency by using explicit geometry reconstructions, such as depth maps, point clouds, or reconstructed 3D structures, to define conditions, supervision, or reward signals, making the generator sensitive to errors from upstream geometry pipelines. We propose VideoWeave, a latent-space post-training framework that uses implicit geometry-model features to constrain the generative distribution, providing a more flexible and non-rigid form of guidance that mitigates the impact of reconstruction errors from geometry models. Specifically, VideoWeave adapts these features into geometry latents and jointly models them with video latents in a shared denoising space, allowing geometry to shape the generative distribution during training. To support this process, we build GeoVid-80K, an 80K-video dataset with paired appearance and geometry representations. Experiments on text-to-video and image-to-video generation show that VideoWeave improves geometric coherence while preserving strong visual quality. VideoWeave project page at https://videoweave.github.io/
arXiv:2606.14164v1 Announce Type: new Abstract: Fuzz drivers are essential components of greybox fuzzing, as they encapsulate target interfaces, define test spaces, and largely determine fuzzing effectiveness. Existing fuzz drivers typically rely on crash-based oracles for security testing, overlooking library functionality and limiting bug detection capability. In this paper, we present the first study on metamorphic-based fuzz oracle enhancement (MFOE), which augments existing fuzz drivers with metamorphic-based oracles derived from metamorphic relations (MRs). Since constructing and integrating such oracles requires substantial domain knowledge, automating MFOE is challenging. To address this challenge, we propose MetaFOE, an LLM-based framework that automatically generates and integrates metamorphic-based oracles. We evaluate MetaFOE on OSS-Fuzz drivers using three modern LLMs and five prompt strategies. MetaFOE generates 3,475 MRs, of which 77.3% are applicable, and implements 12,351 meta drivers, with 6,228 being valid. After three hours of fuzzing, the valid meta drivers improve edge coverage by an average of 18.7% and trigger 1,528 unique crashes. Our results demonstrate both the effectiveness of metamorphic-based oracle enhancement and the feasibility of using LLMs to automate MFOE, providing valuable insights for advancing greybox fuzzing.
arXiv:2606.13807v1 Announce Type: new Abstract: In this study, we present a novel time-stepping scheme for multiscale differential equations describing oscillatory systems with well-separated scales, where the scale separation is controlled by a small parameter $\epsilon$. The time-stepping method is related to a multi-modal WKB approximation and relies on a transformation of variables derived in this work. The analysis reveals that, in the transformed formulation, the leading-order oscillations are either eliminated or appear only at higher asymptotic order. The method is applied to ordinary differential equations, including the well-known van der Pol oscillator. We investigate the accuracy of the proposed method numerically for different parameter regimes, in particular for decreasing values of $\epsilon$, and study how the parameters of the numerical scheme must be adapted as $\epsilon$ is reduced. In the presented numerical tests, the computational cost remains bounded as $\epsilon$ is decreased.
arXiv:2606.14172v1 Announce Type: new Abstract: Multimodal Attributed Graphs (MAGs) model real-world entities by coupling graph topology with heterogeneous attributes such as text and images. They support graph-centric tasks requiring structural and class-discriminative representations, and modality-centric tasks requiring fine-grained cross-modal correspondence. However, existing MAG methods often rely on fixed graph contexts or uniformly fused representations, causing task-agnostic propagation and over-compressed fusion that hinder diverse task requirements and modality-specific evidence preservation. To address this, we propose CoMAG, a unified MAG backbone that learns task-adaptive reliable contexts and modality-preserving alignment within them. CoMAG first conducts Reliable Context Learning by estimating edge reliability from multimodal semantic consistency, complementing raw topology with semantic neighbors, and selecting context components through a task-aware gate. It then performs Modality-preserving Hop-token Alignment by maintaining modality-specific multi-hop trajectories, matching modality-hop tokens across modalities, and decoupling shared and private representations. Thus, CoMAG produces graph and modality representations from one forward pass while retaining modality-specific cues. We further analyze stable propagation, over-smoothing mitigation, and modality-collapse control. Experiments on nine OpenMAG datasets compare CoMAG with feature-only, graph-only, multimodal, and unified MAG baselines across graph-level prediction, modality matching, and graph-conditioned generation. Results show that CoMAG achieves the best reported performance, demonstrating that task-adaptive reliable contexts and modality-preserving alignment improve structural prediction, cross-modal matching, and graph-conditioned generation while retaining sparse edge-linear complexity.
arXiv:2606.14188v1 Announce Type: new Abstract: We present CORD-SLS, a real-time control method for safe deformable object manipulation, with a focus on ropes and cloth. At its core is a GPU-parallel differentiable simulator with contact smoothing which enables efficient gradient-based planning through intermittent contact. To robustly satisfy constraints under model and sensing uncertainty, we develop a real-time, GPU-parallel output-feedback robust model predictive control (MPC) algorithm that plans with this simulator. We further show that the simulator accelerates model-based RL for training neural manipulation policies. To improve real-world robustness, we use conformal prediction to calibrate visual-feedback and perception-error bounds for MPC, producing reachable tubes that enable high-probability safe control. We evaluate CORD-SLS on high-dimensional, contact-rich rope and cloth manipulation tasks in simulation and hardware, including obstacle avoidance, routing, folding, and smoothing. Across settings, CORD-SLS achieves millisecond-speed planning, exceeding baselines in safety, speed, and task success.
arXiv:2606.14189v1 Announce Type: new Abstract: The influence maximization problem aims to select a set of seed nodes that maximize the influence, i.e., the average number of influenced nodes, at the end of a spreading process. It has been widely studied with applications in viral marketing, public health campaigns, and social influence. In networks with pronounced community structure, existing approaches often yield an uneven distribution of influenced nodes across communities, which is unfair. Although the fair influence maximization (FIM) problem has been studied for pairwise networks, it remains largely unexplored for hyper graphs, which more accurately represent real-world systems involving group interactions. We introduce FIMH, a heuristic seed-selection algorithm for FIM on hyper graphs, under the Susceptible-Infected Contact Process (SICP) spreading model. FIMH iteratively estimates the contribution of each candidate node to influence and fairness and selects the node that best trades off these two objectives as an additional seed using a parameter-free utopia-distance criterion. Experiments on seven real-world hypergraphs demonstrate that FIMH achieves an influence comparable to that of state-of-the-art IM methods while significantly reducing influence disparity. Analysis of the topological properties of the selected seed nodes and their contributions to influence and fairness further supports the effectiveness of FIMH.
arXiv:2606.14191v1 Announce Type: new Abstract: We propose and analyze a generalized compartment model for hacker dynamics in cybersecurity systems. This model is an extension of a recently introduced framework, replacing the bilinear interaction term with a broad class of nonlinear incidence functions. This provides greater modeling flexibility and allows for the description of a wider range of cyber-propagation and information-spreading processes. First, we investigate the qualitative dynamics of the model. We establish the positivity and boundedness of solutions, derive the basic reproduction number, and characterize the existence of hacker-free and hacker-present equilibria. We obtain local and global asymptotic stability results, yielding a complete description of the global dynamics in terms of the basic reproduction number. Next, we develop a second-order, positivity-preserving, nonstandard finite difference (NSFD) scheme for numerical simulations. Unlike many existing NSFD approaches, which are typically first-order accurate, the proposed method achieves second-order convergence while preserving positivity for arbitrary step sizes. Furthermore, the proposed method reproduces the asymptotic stability properties of the continuous model, making it suitable for long-time simulations. Numerical experiments validate the theoretical results and demonstrate the superior accuracy and qualitative performance of the proposed scheme compared to several first-order methods.
arXiv:2606.14192v1 Announce Type: new Abstract: Auto-bidding is a core component of real-time advertising systems, where decisions must optimize long-term performance under budget and cost constraints, while online exploration is prohibitively risky. Offline reinforcement learning and, more recently, Transformer-based sequence modeling have shown promise for learning bidding policies from logged data, but their unimodal and purely parametric formulations often collapse multiple effective bidding strategies into suboptimal averaged actions and perform unreliably under sparse or long-tail traffic. To mitigate these limitations, we propose DRIVE (Distributional and Retrieval-Augmented Bidding with Value Evaluation), a unified Transformer-based framework that decouples candidate action generation from decision making for offline auto-bidding. DRIVE combines distributional action modeling, retrieval-augmented candidate generation from high-quality historical decisions, and value-based evaluation to select the most promising bid at inference time. Extensive experiments on AuctionNet and additional offline reinforcement learning benchmarks demonstrate that DRIVE consistently improves bidding performance and generalizes well across multiple Transformer-based methods.
arXiv:2606.14193v1 Announce Type: new Abstract: Filtered approximate nearest neighbor (FANN) search must satisfy both vector similarity and structured predicates, yet evaluations remain brittle because real hybrid workloads are rarely shareable and existing benchmarks rely on ad-hoc synthetic or semi-real constructions. We argue that realism hinges on execution-driven query difficulty: failures in early filtering trigger over-fetching of additional candidates, shaping latency, throughput, and recall. Building on this insight, we propose $\alpha$-Hardness, a query-level hardness metric that models the conditional execution chain via the over-fetch factor and extends naturally to strategy-conditioned settings. Across diverse datasets and hybrid strategies, $\alpha$-Hardness exhibits strong monotonic alignment with empirical performance, while common proxies such as selectivity or attribute-vector correlation are frequently unstable or strategy-inconsistent. We further introduce HCBGen, a hardness-controlled benchmark generator that uses $\alpha$-Hardness as an explicit control signal to synthesize workloads under coarse bias modes or to match a target hardness profile. Our experiments show that widely used benchmarks occupy a narrow, relatively easy portion of the hardness spectrum, masking robustness gaps that emerge under harder queries. Finally, we demonstrate that matching hardness distributions enables privacy-preserving proxy workloads that closely reproduce performance trends, bridging research benchmarks and real evaluation.
arXiv:2606.14195v1 Announce Type: new Abstract: Kalman filters based on the Embedded Latent Transfer Operators (ELTO) emerge as novel statistical tools for sequential state estimation. However, a critical limitation stems from their use of simplified noise models, which fail to dynamically adapt to non-stationary processes. To address this limitation, we introduce an ELTO-based Bayesian filtering approach with a new structured parameterization for the filter's noise model. This parameterization enables structured noise adaptation, which couples the data-driven learning of an optimal time-invariant noise model with dynamic parameter adaptation that responds to changes in dynamics within non-stationary processes. Empirical results show that our structured noise adaptation improves the filter's dynamic state estimation performance in noisy, time-varying environments.
arXiv:2606.14198v1 Announce Type: new Abstract: Incremental nonlinear dynamic inversion (INDI) is a prominent flight-control strategy valued for its robust disturbance rejection; however, its formal stability verification has traditionally been limited to linearized dynamical models. This paper presents a formal nonlinear stability certificate for a saturated INDI pitch-rate controller for a hybrid vertical take-off and landing (VTOL) aircraft by representing the INDI controller via an equivalent recurrent equilibrium network (REN). By casting the saturated INDI architecture as a REN, the closed-loop dynamics are exactly mapped to an augmented state-feedback system. This structural equivalence enables the use of sum of squares (SOS) programming to synthesize a locally valid Lyapunov function without relying on conservative bounding approximations. The resulting certificate yields an inner estimate of the region of attraction (RoA) that explicitly accounts for actuator saturation, formally verifying the controller's stability in operating regimes where standard linear margins lose their validity.
arXiv:2606.14199v1 Announce Type: new Abstract: Large language models are increasingly deployed as human simulators for interactive evaluation and social simulation. Yet helpfulness-driven post-training pulls them toward a homogeneous, overly agreeable assistant register, creating a behavioral Sim2Real gap. We present OdysSim, the largest open systematic investigation of behavioral foundation models, i.e., models trained to simulate human behavior at scale. We propose SOUL, a taxonomy of five capability axes (CONV, SS, COG, ROLE, EVAL) that unifies 62 datasets and 23 benchmark tasks under one framework. Specifically, we curate the OdysSim corpus (21.4M interactions, 10B tokens, retrofitted with back-generated social contexts), construct the SOUL-Index benchmark, and develop an end-to-end training recipe combining midtraining, task-specific RL, and expert distillation. The resulting open 8B OSim model ranks first or tied-first on 8 of 23 tasks, outperforming any individual frontier model by this count, with the strongest gains on conversational and social tasks. Its outputs are also more human-like in length, formatting, and word choice, and it transfers zero-shot to out-of-distribution user simulation on $\tau$-bench, nearly matching real users on reaction alignment (93.2 vs. 93.5). We further show that LLM-as-judge RL induces reward-hacking patterns, and that our detectors can mitigate them during post-training. Together, our findings suggest that behavioral foundation models require rethinking the LLM training paradigm. We release all artifacts to support future research.
arXiv:2606.14108v1 Announce Type: new Abstract: We introduce Adelic operation-preserved embeddings (AOE), a training-free representation that captures both a number's real value and its modular (p-adic) signatures. This construction preserves additive and multiplicative structure by design, turning numerical input into embeddings that "speak in the language of mathematics." Unlike prior approaches that rely on task-specific retraining, AOE is plug-and-play and drops seamlessly into existing architectures. On algebraic combinatorics benchmarks, it delivers consistent gains including the first-ever perfect accuracy on the Weaving Pattern task-while suggesting a principled path forward for overcoming the long-standing "number problem" in AI.
arXiv:2606.14205v1 Announce Type: new Abstract: Sum-rank codes provide a generalized framework for Hamming and rank-metric codes, with codewords represented as tuples of matrices and weight given by the sum of the block ranks. In this paper, we introduce and study a block-tensor-rank invariant for sum-rank metric codes. To each code, we associate its \emph{block tensor rank}: the smallest number of block-simple tensors, namely rank-one matrices supported inside single blocks, whose linear span contains the code. In general, determining the block tensor rank of a sum-rank code is challenging. Our main structural result shows that the block tensor rank decomposes additively across the blocks of the code, thereby reducing its computation to a tensor-rank problem on each block projection. Consequently, we derive two complementary lower bounds on the block tensor rank, referred to as the \emph{projection-wise bound} and the \emph{coordinate-code bound}. Moreover, by combining the coordinate-code bound with the classical Singleton and Griesmer bounds for codes in the Hamming metric, we obtain explicit lower bounds, called the \emph{Singleton coordinate-code bound} and the \emph{Griesmer coordinate-code bound}, respectively. We further construct families of sum-rank codes whose block tensor ranks attain the Singleton or Griesmer coordinate-code bounds. These constructions are based on Hamming-metric codes achieving the corresponding classical bounds. Finally, we show that, in certain cases, the block tensor ranks of two known families of sum-rank codes in the literature do not attain the Singleton coordinate-code bound.
arXiv:2606.14207v1 Announce Type: new Abstract: Resolving the spectral Raman signature of molecules grafted on a metallic support is often a difficult task, in which quantum chemistry methods allow for precious additional rationalization and signal attributions, especially to probe the formation of a bond with the support. In the specific case of gold-organic architectures based on a Au-C bond, only a limited amount of experimental and theoretical reference data are available in the literature, and Raman simulations based on quantum mechanics quickly become unaffordable with the size of the system. In this work, we evaluate the precision of a cost-efficient DFTB method to simulate Raman spectra of gold-organic systems at different scales, from gold complexes to functionalized gold surfaces. After a validation of the method through a careful comparison of DFTB Raman spectra of organometallic gold(I) and (III) complexes to DFT and experimental reference data, we discuss the case of molecules grafted on gold aggregates. For these simulations, the choice of the model (cluster or periodic surface) appears to be critical, and significant differences arise (positions and intensities of the peaks) when considering a full metallic slab, as allowed by the low computational cost of the method.
arXiv:2606.14209v1 Announce Type: new Abstract: In this paper, we evaluate the extent of undisclosed LLM-generated content in texts from the parliaments of the United Kingdom and Sweden. In many areas, such as in journalism or in academic writing, there are often requirements to clearly disclose whether AI tools, such as LLMs, have been used. In the case of parliamentary texts, the guidelines on disclosure of AI use are more vague. However, in order to maintain transparency and retain public trust, it is generally recommended that parliamentarians should state whether or not they have used AI when writing texts, such as parliamentary motions. Here, we train an interpretable (glass-box) text classifier using pre-LLM parliamentary texts and LLM-generated versions of such texts. We then apply the classifier to a test set containing recent parliamentary texts, finding a steady increase in undisclosed LLM use, in both parliaments, from 2022 onwards.
arXiv:2606.14210v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly deployed in privacy-sensitive domains, where users must balance the risk of data exposure through external APIs against the high computational cost of local deployment. Split learning has therefore emerged as a promising paradigm for LLM fine-tuning and inference under limited local resources. However, it introduces new privacy risks. Prior work primarily studies leakage of private input prompts, typically via inversion attacks on intermediate representations, while the potential for sensitive information leakage through generative response outputs remains largely unexplored. In this work, we unveil novel vulnerabilities of Split-LLM by presenting Patched Model Inversion with Dual-Sided Initialization (PIDI), a two-stage attack that simultaneously targets both private input prompts and output responses in Split-LLM settings. It combines dual-sided initialization with a patched inversion strategy to tackle long sequences, substantially outperforming prior inversion methods. To counter threats from both sides, we further propose the Adapter-based DualGuard with Mutual Information Defense (ADMI), which integrates an adapter-based local warmup strategy and mutual information regularization to provide a strong empirical privacy protection with minimal impact on task performance. Extensive experiments across diverse tasks and models demonstrate that ADMI effectively defends against PIDI and other state-of-the-art inversion attacks. Our code is publicly available at https://github.com/FLAIR-THU/VFLAIR-LLM.