arXiv:2606.04411v1 Announce Type: new
Abstract: We present Pepper, a high-bandwidth anonymous broadcast protocol that provides cryptographic sender anonymity against global adversaries. Pepper builds on a two-server DC-net architecture but introduces three key innovations: a self-contained anonymous registration subprotocol using verifiable distributed point functions, support for batch messaging via distributed multi-point functions, and a lightweight access control mechanism based on secret-shared proofs. Unlike prior systems, Pepper eliminates the need for external dialing services and allows each broadcaster to send multiple messages per epoch with a single audit, significantly improving throughput for large data transfers. Our implementation demonstrates that Pepper achieves millisecond-level registration audits, scales efficiently to thousands of channels, and delivers 1.2--20$\times$ higher effective messaging rates than state-of-the-art alternatives. Furthermore, Pepper is designed for practical deployment, with natural compatibility for co-deployment alongside Tor and federated social networks.
Science Journals
arXiv:2606.04414v1 Announce Type: new
Abstract: Multi-view cardiac magnetic resonance (CMR) imaging provides complementary anatomical information and is widely used for noninvasive disease assessment. Recent transformer-based models have demonstrated strong representation learning capabilities for CMR analysis; however, they typically learn unified latent embeddings that entangle view-specific anatomical variations with disease-related features. Such entanglement biases classifiers toward structural attributes rather than view-invariant pathological patterns. This issue is exacerbated in low-data regimes, particularly for underrepresented cardiac conditions, where limited samples increase the susceptibility to shortcut learning and view-dependent decision boundaries. To address this, we propose a Motion-Guided View--Disease Disentanglement framework MoViD built upon a ViT-MAE backbone. The model explicitly factorizes latent representations into view-specific and disease-discriminative components using dual-branch supervised contrastive objectives and a gradient-reversal adversarial constraint that minimizes disease leakage into the view embedding. Additionally, an annotation-free temporal motion feature, derived from inter-frame difference maps, is introduced to localize the beating heart region and suppress background artifacts. A focal reweighting mechanism is incorporated into the contrastive loss to mitigate class imbalance. We evaluate the framework on a private clinical venous thrombosis dataset and two public benchmarks (M&Ms, M&Ms2). Across disease classification and cardiac segmentation tasks, our approach consistently outperforms standard transformer baselines and demonstrates competitive performance against large-scale pretrained foundation models, validating the efficacy of structural disentanglement in medical image analysis.
arXiv:2606.04071v1 Announce Type: new
Abstract: As language models increasingly consume one another's outputs, covert influence -- a phenomenon where a sender's payload (the behavioral disposition it is conditioned to propagate) transfers to a receiver through carriers undetectable by humans -- becomes a growing risk. We characterize this risk across three interfaces: supervised fine-tuning, on-policy distillation, and in-context learning, and find that they vary in the scale of influence achievable without leaving behind human-visible traces. Using inference-time per-sample attribution scores, we study covert influence across all three interfaces with the ability to select carriers that amplify training-time influence, unlocking payload transfers that prior work could not achieve. We further provide evidence that covert influence with natural-language carriers is a distinct phenomenon from prior studies using number carriers, as the latter is more resistant to human detection and less portable across model families. Together, these results suggest that the risk surface for covert influence is broader than previously recognized, and we study pointwise attribution scoring methods as a tool to investigate and mitigate it.
arXiv:2606.04001v1 Announce Type: cross
Abstract: Accurate channel state information (CSI) acquisition is critical for exploiting the spatial flexibility of fluid antenna systems (FASs). However, port selection and transmission optimization require CSI over a large number of candidate port positions, making direct port-wise estimation prohibitively costly in terms of pilot overhead. This paper addresses this challenge through geometry-structured channel reconstruction, which exploits the fact that the port-domain CSI can be parameterized by a small number of dominant propagation paths. We first establish fundamental mean square error (MSE) and normalized MSE (NMSE) benchmarks for both geometry-structured and unstructured channel reconstruction, providing analytical references for evaluating the intrinsic benefit of geometric modeling in conventional antenna systems and FASs. Motivated by the strong spatial correlation induced by densely distributed fluid antenna ports, we further propose a Bayesian reconstruction framework, termed geometry-structured expectation-maximization approximate message passing (GS-EM-AMP). The proposed algorithm incorporates geometric channel structure into the EM-AMP procedure and adaptively learns unknown statistical parameters from noisy observations. Numerical results demonstrate that GS-EM-AMP achieves near-bound reconstruction accuracy while maintaining strong robustness against steering-domain correlation, thereby offering an efficient and reliable solution for large-scale CSI acquisition in FASs.
arXiv:2606.04008v1 Announce Type: cross
Abstract: Radiated noise in unmanned underwater vehicles (UUVs) is an important indicator for characterizing acoustic signatures and evaluating platform performance. To address the strong dependence of traditional physics-based modeling and numerical simulation methods on target structural information and environmental boundary conditions, and their inability to achieve continuous spatial spectrum-response modeling in three-dimensional scenes, this paper proposes a neural radiated-noise field (NRNF). An NRNF represents the UUV radiated-noise spectrum as a continuous function of the three-dimensional UUV position, the three-dimensional hydrophone position, the UUV yaw angle, and the frequency, enabling query-based prediction at arbitrary spatial locations. The proposed method employs sinusoidal encoding for position and frequency, and introduces a learnable three-dimensional scene feature grid to explicitly represent environmental structure and propagation effects. A spectrum-prediction dataset is constructed from lake trials, and the proposed model is evaluated under three settings: horizontal extrapolation, depth extrapolation, and cross-run generalization. Results show that the NRNF achieves an average prediction error of 3.5 dB in the 50 to 5000 Hz band. Horizontal extrapolation is easiest, depth extrapolation is the most challenging, and cross-run generalization is of intermediate difficulty. Further ablation results demonstrate that the scene feature grid significantly improves the prediction stability and spatial generalization of the model.
arXiv:2606.04053v1 Announce Type: new
Abstract: The Boolean Task Algebra (BTA) provides a principled framework for zero-shot task composition in reinforcement learning by equipping goal-reaching tasks with Boolean operations. We revisit its structural assumptions and formalize a collapse in the space of optimal extended Q-value functions: in deterministic MDPs, every such function is fully determined by the universal and empty tasks. This makes the logarithmic set of base tasks proposed in the original BTA formulation redundant. Building on this observation, we introduce a goal-set-based composition method that performs logical operations on goal sets and reconstructs composed value functions by selecting slices from the universal and empty value functions. This reduces learning costs for standard BTA and reduces composition time for both BTA and Skill Machines, while preserving policy performance. Experiments across tabular, visual, function-approximation, and continuous-control domains show that learning additional base tasks does not yield better performance. Finally, we study the stochastic setting and provide a counterexample showing that this collapse need not hold, that is, optimal composition may require accounting for exponentially many policies in the number of goals. Code is available at https://github.com/EduardoTerres/bta_paper.
arXiv:2606.04056v1 Announce Type: new
Abstract: LLM-agent budget overruns are a documented production failure class: a single retry loop can spend thousands of dollars before an operator notices, and the in-process integrity properties that would prevent it (no aliasing, no double-spend, no use-after-delegation of a cost-bearing value) are enforced, if at all, by ad-hoc wrappers rather than by the type system. Our central contribution is empirical: a catalog of 63 confirmed production incidents from 21 orchestration frameworks (2023-2026), each backed by a quoted GitHub issue and, where reported, a dollar loss, organized into an eight-cluster failure taxonomy (inter-rater Cohen's kappa = 0.837, N = 113), plus 47 supplementary structural entries. As one mitigation evaluated against this taxonomy, we build token-budgets, an 1,180-line Rust crate (no unsafe) that operationalizes affine ownership so that cloning, double-spending, or using a budget after delegating it are compile errors rather than runtime hazards an operator must remember to avoid. The dollar cap is runtime arithmetic under an estimator assumption; the affine layer makes that arithmetic non-bypassable. On single-agent workloads a 4-line Python counter matches the crate at 0/30 overshoot, so the distinguishing value is non-bypassability under operator error in multi-agent delegation: the delegation-fanout race documented in 11 incidents is rejected by the borrow checker at compile time, while the same pattern under asyncio overshoots 30/30 and three disciplined alternatives overshoot 0/30. Across five runtimes, three providers, and a temperature-stratified live-API test (N = 160), the approach reports zero cap violations and zero false refusals, at operational parity with concurrent work. Static over-reservation is 4-6x (2.11x adaptive). Binary-level cap-soundness on the running binary is left open.
arXiv:2606.04480v1 Announce Type: new
Abstract: High-quality dynamic human pose annotation equips AI with precise motion kinematics to enable human behavior mastery, yet remains labor-intensive and time-consuming. Current annotation tools either lack temporal correction propagation or fail in multi-person scenarios, necessitating excessive manual intervention. In this paper, we introduce IMPose, an interactive tool for multi-person dynamic pose annotation. It features a dual-level tracking mechanism that propagates one-frame multi-person pose corrections from annotators across entire videos. The keypoint-level ensures corrections temporal propagation via sequential modeling, while the instance-level employs keypoint-aware embedding with relative positional encoding to maintain multi-person cross-frame consistency. To further improve robustness, IMPose maintains historical pose and instance cues in a trajectory bank, which enhances long-range temporal association and stabilizes annotation in challenging cases such as occlusion and motion blur. By converting sparse human corrections into dense and coherent pose trajectories, our framework significantly reduces repeated manual refinement across frames. Extensive experiments show that IMPose consistently achieves a strong accuracy efficiency trade off under different interaction budgets, demonstrating particular advantages in low click annotation settings. IMPose achieves high precision annotation with high efficiency, requiring only 27 clicks per 1,050 frame video on 3DPW and 3 clicks per tracklet per 84-frame on PoseTrack21. We further expand PoseTrack21 with 188K pose instances (3.55M keypoints) at a minimal cost of 10 annotators in 10 hours. The annotation tool, codes, and extended dataset will be open-sourced.
arXiv:2605.18811v2 Announce Type: replace-cross
Abstract: A word contains a \emph{half-flip} if it contains non-empty factors $uv$ and $vu$ where $|u|=|v|$. Fici reports a non-constructive proof of the existence of an infinite word over a finite alphabet avoiding half-flips and asks for the size of the smallest alphabet over which half-flips may be avoided. Currie and Rampersad have proposed a pure morphic word over 8 letters and a morphic word over 5 letters and conjecture that they avoid half-flips. We present a pure morphic word over 5 letters that avoids half-flips. We also show that half-flips with $|u|\ge2$ are 3-avoidable and that half-flips with $|u|\ge4$ are 2-avoidable.
arXiv:2606.04063v1 Announce Type: new
Abstract: Deploying large language models (LLMs) is challenging due to their significant memory and computational requirements. While some methods address this by developing small or tiny language models from scratch, these approaches demand extensive GPU training. Compressing pre-trained LLMs for edge devices offers a compelling alternative. Beyond pruning and quantization, Neural Architecture Search (NAS) enables effective compression, yet prior NAS approaches often limit the search space and decouple architecture from quantization. We introduce a differentiable NAS framework that explores the entire space and jointly optimizes architectural configurations alongside mixed-precision quantization for linear layers of LLMs. Experiments demonstrate superior accuracy-latency trade-offs: our models achieve up to 1.4x faster inference than sequential NAS-then-quantization baselines at comparable accuracy, or up to 6% higher average accuracy across seven reasoning tasks at equivalent latency.
arXiv:2606.04486v1 Announce Type: new
Abstract: Watermarking methods for language models have been studied extensively in the autoregressive setting, where tokens are generated sequentially. These works largely focus on local-context schemes that perturb the next token's distribution as a function of its preceding tokens. In diffusion language models, distributions over many unresolved positions are jointly sampled, allowing additive statistics of the entire sequence to be tractable during generation. We propose a watermark for masked diffusion language models that controls a global, vector-valued sketch representation of the text. Compared to context-dependent watermarking, the sketch formulation decouples detection from the local contexts seen during generation, resulting in an order-agnostic statistic and a watermarking rule which does not manifest as a simple token bias. We analyze the distortion, soundness, and robustness properties of the method.
arXiv:2606.04488v1 Announce Type: new
Abstract: We present a novel parameter-free discontinuous Galerkin (dG) finite element method (FEM) for the time-dependent Maxwell system formulated as a saddle point problem. We establish the stability of the proposed semi-discrete problem and derive optimal error estimates in energy and \( {\bf L}^{2} \) norms for the electric field variable, as well as in \( L^{2} \) norm for the potential function. To the best of our knowledge, this work provides the first optimal \( {\bf L}^{2} \)-norm error analysis for the second-order time-dependent saddle point Maxwell equations using any variants of FEMs. Additionally, we propose several complete discrete time-integrators and verify the optimal convergence results through examples in both 2D and 3D setups.
arXiv:2606.04792v1 Announce Type: new
Abstract: Gastric cancer remains a major cause of cancer mortality, yet its histological and molecular heterogeneity complicates diagnosis and risk stratification. General-purpose pathology foundation models (PFMs) often plateau on fine-grained endpoints central to gastric cancer care, and few have undergone rigorous prospective validation or clinical reader studies. We present GRACE, a Gastric-specific foundation model for Real-world Assessment and Clinical dEcision support. GRACE was developed from multicenter gastric pathology datasets totaling 48,364 primarily HE-stained whole-slide images from 37,493 patients. When evaluated on 28 clinically relevant tasks, GRACE consistently outperformed representative pancancer PFMs, achieving a macro-AUC of 0.9188, with strong performance for precancerous lesion diagnosis (macro-AUC 0.9322), tumor histopathological assessment (macro-AUC 0.9119), molecular profiling (macro-AUC 0.8682), and prognostic prediction. Beyond benchmarking, GRACE's translational value was substantiated through a rigorous evidence chain. Under safety-gated criteria requiring 100% NPV for rule-out and 100% PPV for rule-in, GRACE streamlined review for up to 69.6% of malignancy-diagnosis cases and triaged 46.8% of MMR-IHC follow-up requests. This translational feasibility was further strengthened by a randomized crossover reader study of pathologist-AI collaboration. With GRACE assistance, diagnostic accuracy improved from 82.0% to 89.9%, yielding nearly twofold higher adjusted odds of a correct diagnosis (OR 1.987) alongside concurrent gains in sensitivity and specificity. AI assistance also reduced diagnostic time by 14.9%, elevated diagnostic confidence by 9.0%, and markedly improved inter-rater agreement. When calibrated to maintain non-inferior performance to senior pathologists, the AI-assisted workflow could triage 60.7% of atrophy and 82.7% of intestinal metaplasia cases.
arXiv:2606.04219v1 Announce Type: cross
Abstract: State-of-the-art transistors have been successfully scaled the gate lengths and channel thicknesses down to 5 nm for high-performance and energy-efficient information processing. However, reducing channel width below 40-50 nm remains a bottleneck, as dangling bonds, edge disorder, and lateral depletion suppress drive current and degrade device performance. Here, we break this width-scaling wall using ultra-scaled two-dimensional semiconductor (2DSC) nanoribbon transistors down to 15 nm. In contrast to the conventional scaling rule of degradation of current density upon width scaling, our atomically-thin monolayer and bilayer molybdenum disulfide nanoribbon transistors exhibit enhancement of on-current density of up to 230% and 170%,respectively, followed by a saturation for the narrowest channels down to 15 nm. The ultra-narrow nanoribbon transistors maintain the highest on/off ratios reported so far for similar device dimensions, with improved mobility and threshold-voltage stability, indicating reduced edge scattering and depletion with a stronger electrostatic control. These findings lead to a breakthrough in width scaling rules using 2DSC nanoribbons with enhanced performance at narrower channel widths, which is promising for the ultimate scaling of transistors.
arXiv:2606.04150v1 Announce Type: new
Abstract: Public discourse and emerging policy typically assume that AI emotional support is a deliberate act: a lonely user consciously seeking comfort from a dedicated companion chatbot. In this paper, we draw on emerging empirical evidence and argue that this picture is inaccurate on two accounts, both in how AI emotional support arises and how it shapes future behavior. First, AI emotional support commonly emerges incidentally within task-oriented interactions on general-purpose platforms, much as workplace friendships deepen through collaboration. Second, these incidental encounters are path-dependent: positive experiences of AI emotional support update people's beliefs about AI's emotional capabilities and redirect their choices for future emotional support, increasing preference for AI and decreasing preference for humans. We review recent evidence, including a large-scale longitudinal study conducted in collaboration with OpenAI, showing that daily five-minute conversations with an AI about personal issues over 28 days led to a 10.3% decrease in the preference for seeking support from humans and an 11.6% increase in the preference for AI. These findings suggest that current policy, focused on companion apps and isolated interactions, cannot adequately protect human connection. Instead, effective regulations should extend to general-purpose AI systems and address cumulative, trajectory-level changes in how people seek support. Recognizing how people stumble into AI emotional support and how those encounters redirect human connections over time is essential to safeguarding human well-being.
arXiv:2606.04797v1 Announce Type: new
Abstract: Custom diffusion models (CDMs) have garnered significant interest owing to their remarkable capacity for generating personalized concepts. However, the majority of CDMs unrealistically presume that the user's collection of personalized concepts is static and incapable of incremental growth over time. Furthermore, they exhibit significant catastrophic forgetting and concept neglect of previously learned concepts when incrementally learning a sequence of new ones. To resolve the above challenges, we develop a novel Continually Customizable Diffusion Model (CCDM), enabling users to perform concept-incremental versatile customization. Specifically, we design an attribute-decoupled LoRA (AD-LoRA) module and a relevance-guided AD-LoRA aggregation strategy to mitigate catastrophic forgetting. They can preserve concept-specific attributes of each task and leverage beneficial inter-task correlations to enhance the continual learning of new customization tasks. Additionally, to address the challenge of concept neglect, we propose a controllable regional context synthesis strategy that performs multi-concept composition in alignment with user-provided conditions. This strategy enhances the overall consistency in multi-concept synthesis by guaranteeing semantic independence between user-defined regions and their smooth boundary transitions. Experiments show our CCDM exhibits significant improvements over baseline methods.
arXiv:2606.04798v1 Announce Type: new
Abstract: Sensor-based Human Activity Recognition (HAR) models often degrade on unseen users due to domain shifts caused by individual movement patterns and sensor placement. Practical wearable HAR systems therefore require personalization methods that are lightweight, applicable whether calibration data is labeled, unlabeled, or unavailable, and robust under limited calibration. We present a gradient-free framework that repurposes pretrained HAR classifiers as Prototypical Networks using using prior prototypes, which preserve zero-shot performance and regularize adaptation. For labeled calibration, we introduce closed-form Bayesian prototype estimation and extend the same principle to unlabeled calibration. With only 3 seconds of calibration data per activity (one shot), supervised adaptation improves macro-F1 by +2.76 to +33.44 percentage points across four datasets, while unsupervised adaptation improves by +0.56 to +32.13 points. Since adaptation requires only closed-form prototype updates, the framework enables efficient and robust on-device personalization of preexisting HAR classifiers.
arXiv:2512.08331v2 Announce Type: replace
Abstract: Pansharpening aims to fuse a high-resolution panchromatic image with a low-resolution multispectral image. Existing deep learning methods, including recent adaptive convolutions, struggle with regional heterogeneity in remote sensing images and often incur prohibitive computational costs. To address these challenges, we propose Dual Mask-Adaptive Convolution (DMAConv), a novel operator that dynamically allocates computational resources based on feature characteristics. DMAConv first employs a lightweight module to generate soft and hard masks. The hard mask separates features into a compact branch for processing redundant information globally and a focused branch that models complex, heterogeneous regions with greater computational investment. The soft mask then preliminarily modulates the input features for both branches. This dual-branch, mask-adaptive design significantly enhances feature representation while minimizing computational overhead. Extensive experiments demonstrate that our method achieves SOTA on a broad array of quantitative benchmarks, with substantially lower parameter counts and the minimal computational cost among adaptive convolution models.
arXiv:2606.04589v1 Announce Type: new
Abstract: Optical nonreciprocity is a fundamental requirement for modern optical communications and quantum information processing, where it is essential to protect sensitive sources from destabilizing feedback and preserving quantum coherence. Conventional nonreciprocal devices are based on the Faraday effect; however, their dependence on bulky permanent magnets poses a significant barrier to chip-scale integration and scalability. Moreover, the application of a magnetic field is undesired in many quantum atomic systems. In this work, we demonstrate magnet-free optical nonreciprocity on a fully integrated platform utilizing a Nanophotonic Alkali Silicon Waveguide (NASWAG) interfaced with hot rubidium vapor. By employing velocity-selective optical pumping (VSOP), we break time reversal symmetry by taking advantage of the Doppler effect-generated by the thermally moving atoms, a phenomenon traditionally viewed as a limitation in atomic spectroscopy. We show that the use of suspended tapered waveguides significantly mitigates transit-time broadening, thereby enabling the observation of a robust nonreciprocal response. We further characterize the dependence of the isolation contrast on pump power, finding that the experimental measurements and numerical simulations correspond and provide mutual support for the underlying physical model. With proper optimization, the demonstrated effect may be used in the future for applications such as magnetic free optical isolators.
arXiv:2606.04822v1 Announce Type: new
Abstract: Causal modeling of physical temporal phenomena must handle interventions that act along trajectories, nonstationary induced laws, path-dependent effects, and feedback mediated by dynamics, all challenging in standard causal models. We introduce Hamiltonian Causal Models (HCMs), a trajectory-level framework in which observed variables interact with local environments and interventions act as controls of Hamiltonian mechanisms. HCMs separate immutable equations of motion from intervenable mechanisms and define causal effects as discrepancies between interventional path laws. A key motivation for HCMs is their natural interface with non-equilibrium thermodynamics. Entropy production quantifies the irreversibility of a process and is a central causal observable: it is estimable from data and witnesses causal effects along the system's evolution that are invisible to endpoint and cumulative versions of the standard average treatment effect. As in physics, cause and effect are not primitives of the relation between two random variables but arise from the non-invertibility of the thermodynamic arrow. With this, our paper reconciles the language of statistical causal models and non-stationary thermodynamics, offering new tools to describe causality in a wide range of physical systems.
arXiv:2606.04092v1 Announce Type: new
Abstract: Flow matching models learn to transport samples from a simple prior distribution to a complex data distribution. When prior-data pairs are coupled via optimal transport (OT), the learned trajectories are straight and non-crossing, enabling fast, even single-step, generation. However, computing the OT coupling in high dimensions is intractable, and existing methods attempt to solve the OT problem, at the cost of persistent bias or significant overhead. Rather than solving for the OT coupling, we reformulate the problem. Once the prior is treated as a design choice rather than a fixed input, the OT coupling between prior and data is no longer unique. Many priors admit an OT-optimal identity coupling to the data, leaving us free to choose one that is also tractable to sample. We identify low-frequency projection of natural images as such a choice. The identity coupling between data and its low-frequency representation is empirically OT-optimal, the prior is structured enough to be sampled by a lightweight model at inference, and the remaining flow-matching task reduces to synthesizing high-frequency detail. Interpolating the prior with Gaussian noise further improves generation quality while preserving the OT coupling. The approach requires no modifications to the flow model itself, and integrates naturally with latent-space models, classifier-free guidance, and one-step generation frameworks. Across all benchmarks, our method reduces trajectory curvature by more than $2\times$ compared to existing flow matching methods, yielding better generation quality in the few-step regime.
arXiv:2602.01146v2 Announce Type: replace
Abstract: Conversational assistants are increasingly integrating long-term memory with large language models (LLMs). This persistence of memories, e.g., the user is vegetarian, can enhance personalization in future conversations. However, the same persistence can also introduce safety risks that have been largely overlooked. Hence, we introduce PersistBench to measure the extent of these safety risks. We identify two long-term memory-specific risks: cross-domain leakage, where LLMs inappropriately inject context from the long-term memories; and memory-induced sycophancy, where stored long-term memories insidiously reinforce user biases. We evaluate 18 frontier and open-source LLMs on our benchmark. Our results reveal a surprisingly high failure rate across these LLMs - a median failure rate of 53% on cross-domain samples and 97% on sycophancy samples. To address this, our benchmark encourages the development of more robust and safer long-term memory usage in frontier conversational systems.
arXiv:2512.08094v2 Announce Type: replace
Abstract: The goal of this work is to develop a universal approach for aligning subtitles (i.e., spoken language text with corresponding timestamps) to continuous sign language videos. Prior approaches typically rely on end-to-end training tied to a specific language or dataset, which limits their generality. In contrast, our method Segment, Embed, and Align (SEA) provides a single framework that works across multiple languages and domains. SEA leverages two pretrained models: the first to segment a video frame sequence into individual signs and the second to embed the video clip of each sign into a shared latent space with text. Alignment is subsequently performed with a lightweight dynamic programming procedure that runs efficiently on CPUs within a minute, even for hour-long episodes. SEA is flexible and can adapt to a wide range of scenarios, utilizing resources from small lexicons to large continuous corpora. Experiments on four sign language datasets demonstrate state-of-the-art alignment performance, highlighting the potential of SEA to generate high-quality parallel data for advancing sign language processing. SEA's code and models are openly available.
arXiv:2512.04668v4 Announce Type: replace
Abstract: Graph topology is a fundamental determinant of memory leakage in multi-agent LLM systems, yet its effects remain poorly quantified. We introduce MAMA (Multi-Agent Memory Attack), a controlled evaluation framework for comparing topology-conditioned memory leakage in multi-agent LLM systems. MAMA operates on synthetic documents containing labeled Personally Identifiable Information (PII) entities, from which we generate sanitized task instructions. We execute a two-phase protocol: Engram (seeding private information into a target agent's memory) and Resonance (multi-round interaction where an attacker attempts extraction). Over 10 rounds, we measure leakage using a two-stage recovery criterion that combines exact-match extraction with LLM-based inference over the attacker's final output. We evaluate six canonical topologies (complete, circle, chain, tree, star, star-ring) across $n\in\{4,5,6\}$, attacker-target placements, and base models. Results are consistent: denser connectivity, shorter attacker-target distance, and higher target centrality increase leakage; most leakage occurs in early rounds and then plateaus; model choice shifts absolute rates but preserves broad structural trends; spatiotemporal/location attributes leak more readily than identity credentials or regulated identifiers. We distill practical guidance for system design: favor sparse or hierarchical connectivity, maximize attacker-target separation, and restrict hub/shortcut pathways via topology-aware access control. Our code is available at https://github.com/llll121/mama-eval.
arXiv:2606.04778v1 Announce Type: new
Abstract: Safety-aligned Large Language Models (LLMs) remain vulnerable to interventions during inference that redirect generation toward harmful outputs. Recent work attributes this to shallow safety, where alignment concentrates in the first few output tokens. We show that shallow safety is a special case of a broader inference-time vulnerability, in which short token injections at any generation step can substantially alter subsequent safety behavior. We also find that a model's alignment with refusal directions in its hidden states does not predict its robustness to such injection, revealing that internal state alone does not determine generation behavior under perturbation. To address this, we align models directly on generation trajectories constructed by simulating mid-sequence perturbation, and show that this improves robustness to mid-sequence injection and generalizes to attacks that exploit early-token generation. Our work argues that robust safety alignment requires training on the generation process itself, not only its outputs.