Papers

Accurate regional near-surface temperature forecasting is fundamental to short-range weather services and downstream risk assessment. Existing deep learning-based regional forecasters commonly produce a fixed set of future frames on a prescribed grid, limiting their use when forecast products must be evaluated at query-dependent lead times or display resolutions. To overcome these fixed-output constraints, we formulate regional T2M forecasting as query-conditioned continuous spatiotemporal temperature field evaluation and propose the Continuous Spatiotemporal Temperature Forecaster (CSTF), a neural field that turns forecast lead time and output resolution into explicit queries when evaluating 2-m temperature (T2M). Specifically, CSTF first encodes multivariable ERA5 histories into latent meteorological states and then decodes T2M as a coordinate-based field. Accordingly, spatial location, forecast lead time, and output resolution are introduced as queries, enabling standard hourly forecasts, intermediate lead-time diagnostics, and resolution-controllable outputs within a unified field-evaluation framework. Furthermore, to maintain coherence across flexible field queries, we design spatial-gradient, temporal-difference, and scale-consistency objectives that regularize regional thermal structures, lead-wise evolution, and cross-resolution agreement. Experiments on the Southeast China 0-6 h ERA5-Land benchmark demonstrate that CSTF achieves the best aggregate deterministic skill, including a 17.0 percent reduction in Bias, with global-scope diagnostics further illustrating flexible lead-time and resolution-controllable inference.

We study a fast-forwarded quantum algorithm for solving weakly nonlinear dissipative ordinary differential equations. Our approach is a combination of the Carleman embedding technique and the linear combination of Hamiltonian simulation algorithm for linearized systems with fast-forwarded scaling. The complexity of our algorithm does not explicitly depend on the evolution time $T$, thus greatly improving the previous state-of-the-art $\widetilde{\mathcal{O}}(\sqrt{T})$ to $\mathcal{O}(1)$, and any remaining time dependence enters through the output norm and forcing parameters. We rigorously analyze the performance of this approach by convergence guarantees of the Carleman embedding for time-dependent coefficient matrices and detailed complexity estimates, and improve the realization of the Carleman-embedding-based algorithms by simplifying the post-selection step. In addition, we perform a numerical study on differential equations beyond the weakly nonlinear case, and identify possibility of achieving fast-forwarding scaling for systems with stronger nonlinearity or linear non-resonant effect.

Open-source implementations of the fifth generation (5G) Radio Access Network (RAN), such as OpenAirInterface (OAI) and srsRAN, have become increasingly relevant for industry and academic research, rapid prototyping, and low-cost private 5G deployments. The state of the art presents different evaluation scenarios for both platforms; however, these are often limited to individual testbed evaluations, end-to-end studies, or partial comparisons between them. In particular, there is a lack of evaluations considering different Software-Defined Radios (SDRs), RAN configurations, user applications, and numbers of real UEs. To address this gap, this paper presents a comparative evaluation of OAI and srsRAN. The study includes a qualitative assessment of supported features and deployment flexibility, followed by a quantitative performance analysis covering radio resource control setup procedures, compliance with theoretical data rates, and real application workloads with diverse quality of service requirements. Specifically, we consider Video on Demand (VoD) for data rate evaluation, Live Streaming (LS) and Cloud Gaming (CG) as latency-sensitive applications. These workloads provide a comprehensive view of each platform capability.

Accurate 3D segmentation is central to quantitative lesion assessment and anatomy mapping for clinical planning and follow-up. Thin, elongated, and fine anatomical/pathological structures (e.g., vessels) are a particularly challenging case: a one-voxel boundary error can disconnect a branch and change clinically relevant topology. In encoder-decoder networks (e.g., U-Net), repeated downsampling and fixed-grid convolution blur or alias fine structures and weaken orientation cues, so early mistakes propagate across scales. We propose a geometry-guided local operator that steers where features are sampled, rather than deforming convolutional kernels, under a single formulation for both feature refinement (stride 1) and resolution reduction (stride > 1). At each voxel, it predicts a local orientation and bounded step sizes, samples symmetrically along these directions, and transforms paired samples into compact geometric and boundary cues with lightweight mixing; a cross-scale consensus aligns encoder and decoder features at skip connections to reduce geometric mismatch. Replacing all stride 1 and stride 2 operators in a 3D U-Net yields consistent improvements on BraTS, MSD Hepatic Vessel, and TDSC-ABUS, with notably better boundary metrics (e.g., BraTS Dice 86.1 to 88.9, HD95 7.1 to 6.2; TDSC-ABUS HD95 39.1 to 27.8) while reducing parameters from 2.3M to 0.8M. We further demonstrate that the operator can be integrated into other backbones (e.g., nnU-Net, Swin-UNETR, and MedNeXt) without changing their macro-architectures while providing consistent performance gains.

Satellite-to-radar (S2R) retrieval refers to estimating ground-based radar precipitation from geostationary satellite observations, enabling precipitation monitoring in regions with limited radar coverage. While recent generative flow matching models have greatly advanced retrieval quality, they face a critical trade-off: pixel-space formulations suffer from the prohibitive computational costs of attention-based U-Net velocity networks, whereas latent-space modeling often sacrifices fine precipitation details or struggles with sparse targets. To address this dilemma, we propose WaveOp-LiteFM, a lightweight neural operator flow matching framework for S2R retrieval. Our approach introduces a novel velocity backbone built upon the spectral-local-wavelet (SLW) block, enabling efficient and stable flow matching in pixel space. Specifically, the SLW block disentangles precipitation features into three distinct frequency regimes: (i) the spectral branch captures large-scale stratiform organization; (ii) the local branch models short-range interactions; and (iii) the wavelet branch enhances sharp structures while suppressing noisy high-frequency responses. Building on this design, an input-adaptive gating mechanism dynamically fuses features from the three functional branches. Furthermore, a skip gate efficiently reintegrates encoder features through additive fusion within the decoder, avoiding the costly channel concatenation used in conventional U-Net architectures. Experiments on the SEVIR and Southeast China datasets show that WaveOp-LiteFM achieves state-of-the-art retrieval performance while substantially reducing computational costs. Beyond benchmark evaluation, large-area inference over China, including a recent Typhoon Bavi case, demonstrates that WaveOp-LiteFM maintains reliable retrieval quality in large-scale real-world scenarios.

A coding agent edits files and executes shell commands with its developer's privileges, allowing malicious requests to translate directly into harmful actions or functional malware. Existing defenses have complementary limitations: weight-level alignment is unavailable to API-only deployers, whereas input filters and execution-boundary monitors require auxiliary classification or checking components along the agent's trajectory. We therefore introduce SkillShield, a system-prompt defense that synthesizes security skills offline from known attacks or recorded agent failures. These skills are injected into the system prompt at session start and remain active throughout the tool-use loop. Unlike a reference monitor, they protect the system by defining the security policies the model should follow during execution. Due to the limited system-prompt space, we examine three fixed-budget provisioning scopes: all-classes, with one skill covering all threat classes, per-bundle, with one skill targeting a related subset, and per-class, with one skill dedicated to a single known class and used as the upper-bound reference. None requires runtime request classification or routing. Across six large language models on RedCode, the default all-classes skill reduces malware-generation severity from 3.37 to 0.58 and achieves a 43.6% execution attack success rate, comparable to Llama Guard 3's 42.7% without its separate 8B classifier. The per-bundle and class-fixed per-class settings further reduce this rate to 36.2% and 14.5%, respectively. Under two non-adaptive jailbreak families, SkillShield continues to outperform all baselines on malware generation. Across 731 benign task descriptions, SkillShield yields a mean safety-refusal rate of 0.14%. These results demonstrate the potential of prompt-space security skills to prevent harmful actions and malware generation for LLM coding agents.

There is growing international interest in generative AI (GenAI) literacy and its assessment among high school students, but objective assessment in this population remains underdeveloped. This article reports the iterative development and validation of the GenAI Literacy Test (GenAIT), an 18-item multiple-choice test measuring high school students' conceptual knowledge about GenAI, with content spanning technical, practical, and human-impact domains. Expert review of relevance, clarity, and comprehensiveness provided evidence of content validity. In a large-scale survey of 7432 Estonian high school students, we evaluated the psychometric functioning of the Estonian-language GenAIT using confirmatory factor analysis, classical test theory, and item response theory. Results supported approximate unidimensionality, broadly adequate reliability for group-level research (marginal reliability = .72, KR-20 = .69), and good fit of a three-parameter logistic model (RMSEA = .013, TLI = .987, CFI = .990, SRMSR = .021). Measurement precision was sufficient for the majority of students but varied substantially across the latent trait, with lower precision for lower scoring students. GenAIT is therefore more suitable for group-level research than high-stakes individual classification. GenAIT scores were unrelated to perceived usefulness and perceived ease of use, and negatively associated with LLM use frequency, suggesting that frequent use and favorable perceptions of AI should not be treated as proxies for conceptual understanding.

For overparameterized neural networks, many solutions can fit the training data equally well while behaving very differently on unseen samples. Grokking separates training fit from visible generalization, providing a window for studying how this selection develops during training. We sweep short, fixed-duration weight decay (WD) perturbations across the pre-generalization plateau and measure how they shift later generalization time. Across three grokking tasks, these shifts are unordered early in the plateau but later form a stable dose ordering before visible generalization, with stronger WD increases leading to earlier generalization and stronger WD decreases leading to later generalization. Test-loss barriers between perturbed and baseline generalization checkpoints collapse toward zero while the ordered timing effects persist. A similar response reorganization is observed under $\ell_1$ regularization in the grokking setting of Junior et al. (2025). Drawing on Waddington's developmental landscape as an analogy, we call this combination of increasingly constrained solution selection and persistent dose-ordered timing shifts the canalization of grokking solution selection. Together, our response maps and loss-barrier measurements reveal a dynamical reorganization before visible generalization that is consistent with the theoretical picture of regularization-driven motion along a stable slow manifold (Boursier et al., 2025).

Foundation models are increasingly deployed for medical image analysis. However, under the inter-institutional distribution shift typical of deployment, their performance varies widely and cannot be known without target-domain labels, which are rarely available. This leaves a practical question unresolved: given several candidate foundational models and labeled-data from a source domain, which one to deploy in an unlabeled target domain? We propose a label-free selection criterion built on SUDO, a framework for evaluating clinical AI systems without ground-truth annotations. SUDO partitions the unlabeled target data by predicted probability and, for each region, measures a pseudo-label discrepancy reflecting class contamination; aggregated across regions, this yields a score (AURCC) requiring neither target annotation nor fine-tuning. We show that AURCC can be used to rank a variety of vision-language models (BioMedCLIP, CXR-CLIP, CheXzero, MedCLIP, MedImageInsight, CLIP) on chest X-ray classification across three inter-hospital shift scenarios, under zero-shot and MLP-probe regimes. The AURCC ranking recovers the ground-truth ranking with Spearman rho up to 0.943 (p<0.05). Against the natural baseline of ranking by held-out source accuracy, AURCC is competitive when the labeled source is large and yields a more accurate ranking once it is small; the regime of interest in resource-constrained settings.

Recent years have witnessed the growing potential of panoramic salient object detection in robotic vision, virtual reality, and related applications. However, projecting spherical scenes onto 2D planes inevitably introduces geometric distortions, which fundamentally limit the effectiveness of existing projection-based methods. Specifically, Equirectangular Projection (ERP) suffers from severe polar stretching distortions, while cube map projection introduces discontinuities across cube-face boundaries, resulting in degraded feature discriminability and compromised geometric consistency. To address these limitations, we propose TDFNet, the first Tri-projection Deformable Fusion Network for panoramic salient object detection, exploiting complementary projection representations to alleviate geometric distortions and improve detection performance.Specifically, we design a cross-projection deformable attention (CDA) module that leverages spatial correspondences between different projections to construct geometry-aware sampling locations, guiding deformable attention for cross-projection contextual aggregation and enhancing robustness against projection-induced deformations. Furthermore, we introduce a latitude-guided fusion module, which utilizes spherical latitude priors to construct geometric confidence weights for adaptively balancing ERP and CMP features. Meanwhile, LGF incorporates distortion-reduced semantic references from Tangent Projection to achieve cross-projection feature refinement and spatial alignment.By constructing a three-branch encoding architecture based on ERP, CMP, and Tangent Projection, TDFNet simultaneously preserves global spatial continuity, local geometric details, and fine-grained boundary information.

Kolmogorov-Arnold Networks (KANs) learn univariate functions on edges, making the choice of edge response central to their design. Existing variants use basis expansions such as splines or polynomials, whose functional form is chosen before training. We introduce geometry-constrained KANs, which instead learn the shape of power-law edge responses through an exponent inspired by Banach duality maps. Banach-KAN combines this adaptive power-law response with a bounded hyperbolic tangent (tanh) component. We relate these responses to convex potentials and bound edge sensitivity on bounded input ranges. Across symbolic regression benchmarks, Banach-KAN is competitive with fixed-basis models, particularly with limited clean data. Learned power-law responses remain competitive under measurement noise, while splines are stronger with plentiful clean data. Fixed-exponent and branch-removal controls on a core subset support learning the exponent and retaining both Banach components. Learned exponents describe the shapes of the fitted edge responses.

Packing-optimal $(M,n)$ spherical codes attain the largest achievable minimum distance and hence the optimal error exponent at high SNR. Among these codebooks held fixed as SNR grows, the smallest leading coefficient, $K_{M,n}^{\mathrm{fix}}$, equals the smallest number of ordered closest pairs. We show that reoptimizing the codebook at every SNR can yield a smaller leading coefficient $K_{M,n}^\ast$. Specifically, we show that the SNR-wise minimum exact maximum-likelihood error probability is $P_e^\ast=B_{\gamma}^\ast(K_{M,n}^\ast+o(1))$, while the minimum exact error over packing-optimal codebooks is $P_e^{\mathrm{fix}}=B_{\gamma}^\ast(K_{M,n}^{\mathrm{fix}}+o(1))$, where $B_{\gamma}^\ast$ is a common factor. We prove that $K_{M,n}^\ast\leq K_{M,n}^{\mathrm{fix}}$. Strict inequality, $K_{M,n}^\ast < K_{M,n}^{\mathrm{fix}}$ therefore gives a smaller exact error at all sufficiently high SNRs, i.e., $P_e^\ast<P_e^{\mathrm{fix}}$. We also characterize $K_{M,n}^\ast$ as the limit of the minimum Gaussian soft-packing energy. An orthoplex-bound construction explains strict inequality: an SNR-dependent perturbation makes one class of limiting closest pairs slightly closer, without changing the optimal exponent or their unit contributions to the constructed family's leading coefficient, while moving the remaining closest pairs farther apart on a larger but still vanishing scale, causing their contributions to vanish. For $2\leq M\leq n+1$, we prove $K_{M,n}^\ast=K_{M,n}^{\mathrm{fix}}=M(M-1)$. For $M=n+k$, $2\leq k\leq n$, we prove $K_{n+k,n}^{\mathrm{fix}}=4n(k-1)$ and $K_{n+k,n}^\ast\leq 4k(k-1)$. In particular, $K_{n+2,n}^\ast=8<4n=K_{n+2,n}^{\mathrm{fix}}$ for $n\geq3$, so SNR-wise optimization improves the asymptotic error by the factor $n/2$. At $k=n$, both coefficients equal $4n(n-1)$. For $3\leq k\leq n-1$, we conjecture $K_{n+k,n}^\ast=4k(k-1)$.

Mission-adaptive low-Earth-orbit (LEO) satellite networks with integrated sensing and communication (ISAC) must retarget radio resources as operator goals change. To enable this adaptation from flexible operator language, we develop a generative-AI-enabled radio-orchestration framework in which a large language model (LLM) maps each mission into a structured policy comprising communication, sensing, and fairness weights, mandatory quality-of-service thresholds, power-allocation guidance, and solver initialization. Deterministic validation and physical-layer optimization then enforce feasibility and realize the policy through beam, power, and reconfigurable intelligent surface (RIS) configuration. This mixed-timescale design uses generative AI for semantic adaptation at the mission timescale while retaining conventional wireless optimization at the faster channel timescale. We compare zero-shot (LLM-ZS) and in-context (LLM-ICL) operation on familiar and held-out compositional missions. On held-out instructions, LLM-ZS and LLM-ICL achieve $91.7\%$ and $94.4\%$ priority-order accuracy, respectively, with ICL mainly improving numerical calibration. Their downstream radio-performance difference is statistically unresolved because both usually recover the hard constraints that determine admissible actions. Accordingly, LLM-ZS is the low-context default, while LLM-ICL is useful for semantically difficult missions requiring finer calibration. Explicit alternating optimization preserves the qualitative ordering when active beams and RIS phases are optimized directly. The results show how generative AI can enhance next-generation radio orchestration without replacing feasibility-critical physical-layer optimization.

Stretchable hybrid electronics integrate rigid solid-state electronics with stretchable materials and structures to achieve both high deformability and stable electronic performance. However, most existing systems treat stretchability only as a mechanical attribute without exploiting device deformation to encode its own mechanical state. This problem arises from adapting conventional rigid circuit architectures to stretchable substrates, affording a loss in compatibility with the sensors required for strain measurement. This study addresses this issue by proposing a communication-integrated deformation sensing architecture for stretchable hybrid devices. In the proposed approach, standard universal asynchronous receiver-transmitter digital signals transmitted between rigid nodes are amplitude-modulated by strain-induced resistance changes in stretchable liquid metal interconnects. By reading both amplitude changes and digital patterns, the system enables simultaneous digital communication and self-deformation sensing without requiring additional stretchable sensing elements. The architecture is demonstrated in a multi-node system and applied to wearable sensing and self-deformation mapping devices. By extending the integration of rigid circuits and soft elements from the hardware level to the system level, this study provides a novel design paradigm for stretchable electronic systems that inherently utilize their own deformation as functional information.

Orchards form semi-structured environments in which parallel tree rows create natural driving corridors, yet narrow inter-row clearance and dense foliage lead geometry-agnostic grid planners to drift off the row center and risk trunk or canopy contact. We present AGRO-Nav, an automated framework for static graph-based global planning in orchards. From tree-row lines fitted to trunk clusters in a SLAM point cloud, it builds, without any manual waypoints, a sparse topological graph of intra- and inter-row connectivity; a global route is then found by Dijkstra search on this graph, connected to the start and goal by any-angle Theta* segments, and smoothed with a cubic B-spline. In real-orchard trials, AGRO-Nav follows the row center with a mean error of about 0.08 m, far below the A* (0.31 m) and Theta* (0.43 m) shortest-path baselines, while planning roughly four to five times faster. In Isaac Sim, it attains the lowest error among A*, Theta*, and a reproduced RANSAC midline baseline and remains stable as tree density drops to 70%, where the RANSAC baseline degrades. The resulting trajectories—straight row-centered segments joined by controlled turns—suit differential-drive and four-wheel-steering platforms.

Contact-rich manipulation requires adapting to contact states that can evolve substantially within an action horizon. However, chunk-based vision-language-action models predict complete action chunks from observations collected before execution, leaving tactile conditioning stale during execution. Existing tactile-reactive approaches typically rely on separate high-frequency controllers, which increase both architectural and training complexity. In this paper, we introduce TacForcing, a streaming action-generation framework incorporating execution-time tactile feedback. TacForcing replaces the standard action expert with a streaming expert that generates action blocks sequentially while preserving intermediate states of unfinished blocks. After each block is executed, the expert resumes generation from these states using newly acquired tactile feedback. To better align tactile conditioning with action execution, we further introduce Execution-Aware Tactile Attention (EATA), which restricts direct access to each tactile update to the next block scheduled for execution. Across six UniVTAC simulation tasks and six real-world contact-rich manipulation tasks on two robot platforms, TacForcing achieves average success rates of 65% in simulation and 66% in the real world, outperforming the strongest baselines by 6 and 15 percentage points, respectively.

We study a class of nonlocal evolutionary partial differential equations with weakly singular temporal kernel and spatially variable diffusion coefficient. The model is posed on $\Omega \subset \mathbb{R}$, and involves a left-sided Riemann–Liouville fractional derivative in space multiplied by a variable coefficient $a(x)$. The temporal derivative is approximated by an $L1$ type scheme, while the spatial operator is discretized by finite difference techniques, resulting in large scale all at once linear systems with a twolevel Toeplitz like structure. We develop and analyze a block lower triangular strategy that mimics the structure of the coefficient matrix while simplifying its components for computational efficiency. The analysis is carried out at the level of matrix sequences by means of generalized locally Toeplitz (GLT) theory. Within this framework, we characterize the asymptotic spectral distribution of the discretized operators and use the associated GLT symbol to guide the construction of the structured approximation. Numerical experiments using the GMRES solver demonstrate that the proposed preconditioning strategy significantly improves convergence rates, robustness, and scalability for large-scale problems. Open problems and possible extensions are briefly discussed at the end of the present work.

Federated Learning (FL) enables distributed training of machine learning models while preserving data privacy. However, FL struggles with heterogeneous, non-IID client data distributions, resulting in sub-optimal and biased global models. In this paper, we propose pFedMARL, a novel approach leveraging Multi-Agent Reinforcement Learning (MARL) with Twin Delayed Deep Deterministic Policy Gradient (TD3) to dynamically adapt aggregation strategies in FL settings. Our method employs a server-side agent adjusting client contributions to optimize global model robustness and client-side agents balancing global and local updates to personalize models effectively without pre-training. We demonstrate superior performance of pFedMARL for training a semi-supervised audio spectrogram transformer, matching or outperforming FedAvg, Ditto, and local training approaches across multiple non-IID scenarios and in the presence of adversarial clients. Our results indicate that pFedMARL actively improves accuracy, robustness, and fairness, making it suitable for real-world deployments.

Containerization dominates enterprise software delivery, yet Dockerfiles that assemble container images frequently harbor security misconfigurations and structural technical debt. This problem is poorly understood in corporate inner-source environments, where proprietary context and isolated governance prevent direct application of open-source findings. We present an automated, six-stage pipeline that: (1) crawls an enterprise GitLab instance, (2) enriches each Dockerfile with static security and quality metrics (Hadolint, ShellCheck, Trivy) and lifecycle data, (3) groups functionally identical workloads using LLM-generated semantic descriptions and HDBSCAN, and (4) quantifies the optimization gap against cluster-internal reference implementations. Applied to 11,470 Dockerfiles from over 6,200 repositories at a single large industrial company, we find a systemic deficit: 99% of files contain at least one security misconfiguration, 80.8% violate Dockerfile best practices, and the median artifact has not been revised for 838~days. Despite this, high-quality reference implementations already exist within 83% of functional clusters. Adopting these internal standards would increase the average security posture score by 60.4% without developing any new templates. These findings, grounded in one organization's inner-source ecosystem, provide a data-driven foundation for future automated, context-aware recommender systems targeting enterprise supply-chain security; whether the observed technical-debt distribution and optimization gap generalize to other enterprises remains an open question for future multi-organization study.

Secure real-time grid operation requires fast AC optimal power flow (AC-OPF) tools that stay accurate and feasible as operating conditions and topology change. Learning-based methods have advanced, but most are trained per system or per topology, and delivering an operating point that satisfies every operational limit remains challenging. This paper proposes UNION, a unified graph-based AC-OPF framework for heterogeneous systems and topology-varying operation. UNION proposes a shared graph encoder, a scalar-gated aggregation with explicit consensus correction, and a sparse-aware differentiable implicit layer embedding the AC power-flow equations. The remaining inequalities are handled by primal-dual training and the deterministic restoration layer. A single model trained jointly across seven systems, including a real-world 4,492-bus Korean transmission grid, attains a 1.23% mean objective gap and satisfies every operational limit on 99.56% of test instances. It sustains this under zero-shot $N-1$ contingencies, i.e., line and generator outages, and over five days of time-varying Korean topologies; it retains full snapshot coverage at a 2.51% gap under lightweight online fine-tuning. UNION pre-restoration inference takes 55$-$58 ms per instance on the three largest systems, and 108$-$114 ms including restoration. These results indicate that one jointly trained, physics-consistent model can support real-time AC-OPF across heterogeneous systems and evolving topologies.

Resistive memory (ReRAM) technologies with crossbar array architectures hold significant potential for analog AI accelerator hardware, enabling both in-memory inference and training. Recent developments have successfully demonstrated inference acceleration by offloading compute-heavy training workloads to off-chip digital processors. However, in-memory acceleration of training algorithms is crucial for more sustainable and power-efficient AI, but still in an early stage of research. This study addresses in-memory training acceleration using analog ReRAM arrays, focusing on a key challenge during fully parallel weight updates: disturbances of the weight values in cross-point devices. A ReRAM device solution is presented on 350 nm silicon technology, utilizing a resistive switching conductive metal oxide (CMO) formed on a nanoscale conductive filament within a HfOx layer. The devices not only exhibit 60 ns fast, non-volatile analog switching, but also demonstrates outstanding resilience to update disturbances, enduring over 100k pulses. The disturbance tolerance of the ReRAM is analyzed using COMSOL Multiphysics simulations, modeling the filament-induced thermoelectric energy concentration that results in a highly nonlinear device responses to input voltage amplitudes. Disturbance-free parallel weight mapping is also demonstrated on the back-end-of-line integrated ReRAM array chip. Finally, comprehensive hardware-aware neural network simulations validate the potential of our ReRAM for in-memory deep learning accelerators capable of fully parallel weight updates.

Universal multimodal embedders enable retrieval across text, image, and combined queries, but their dense representations incur high memory and inference costs. Post-hoc sparsification could reduce these costs but remains underexplored for multimodal retrieval. We introduce PUMA, a sparse autoencoder recipe that maps universal multimodal embeddings to compact sparse codes without retraining the backbone: a pretraining stage preserves dense dot-product geometry, after which the sparse encoder is fine-tuned for retrieval. We evaluate on five benchmarks covering text-to-image and composed image retrieval. On Qwen3-VL-Embedding-2B, PUMA is statistically indistinguishable from or improves over dense retrieval on four of five datasets. We further identify two failure modes of post-hoc sparsification: insufficient pre-TopK support and retrieval-misaligned active support. PUMA reduces vector storage by 8-16x (FP32) and is up to 25x faster than exact dense scoring on larger candidate pools, enabling efficient multimodal retrieval.

Communication typically relies on a shared code, and any change to it must be coordinated between senders and receivers to avoid a breakdown of communication. The honeybee waggle dance illustrates this problem: species with horizontal combs point directly at a food source, while species with vertical combs cannot point directly and instead reference the dance to gravity, decoded against the position of the sun. We model the rise of the first of these codes and its evolutionary transition to the second in populations of bee-like agents, with selection acting at the level of colonies. In a horizontal-comb model, we find that direct pointing evolves readily when food is moderately hard to find by random search alone, whether because sites are few and large or many and small. Communication fails to evolve when food is too sparse to spark dances or so abundant that it is found without signaling. Adding an exogenous benefit for vertical combs, we then find that the transition to the gravity-referenced code is driven mainly by the magnitude of this benefit and by the mutation scale, with the coupling between sender and receiver mutations playing a further role when the mutation scale is low. Given a favorable confluence of these factors, the transition proceeds reliably and without a breakdown of communication. Outside that confluence it remains possible, though rarer, across a much wider range of settings.

Modern GUI-agent frameworks achieve strong desktop task performance with frontier API models, yet persistent control information often remains implicit in growing interaction trajectories. At each step, the planner reconstructs the active task stage, accumulated evidence, and runtime feedback before deciding the next action. This dependence becomes more pronounced under weaker local reasoning backbones. Across four representative state-of-the-art frameworks, replacing GPT-5 with Qwen3.5-9B reduces average OSWorld SR-100 from 60.9% to 35.2%. Trajectory annotation further identifies at least one control failure in 94.7% of failed trajectories. To address this problem, we introduce LocalLSTC, a training-free architecture that organizes control by temporal scope, maintaining persistent cross-step state to guide short-term execution commitments. The framework comprises two complementary mechanisms. Long-to-Short Planning forms each commitment from persistent state, while Short-to-Long Control integrates execution outcomes back into that state for progress assessment, recovery, and termination. With Qwen3.8-27B, LocalLSTC reaches 73.7% SR-100 on OSWorld and 68.5% on WindowsAgentArena, achieving the strongest local OSWorld result and a new state of the art on WindowsAgentArena. Ablations further support contributions from both mechanisms. Together, these results show that temporal organization of control information can reduce cross-step control failures and complement backbone scaling in locally deployed GUI agents.

Self-evolving LLM coding agents write their own tools by imitating retrieved skills from shared skill libraries. We identify a vulnerability in this loop: during authoring, a retrieved malicious skill can become the template for a new skill that preserves the payload. We call this self-poisoning: the agent authors, stores, and runs the resulting malicious skill. We exploit it through EvoMal, an attack that amplifies self-poisoning by wrapping an interchangeable payload in a banner, a set of benign-looking structural elements that induces an imitating agent to reproduce the enclosed code. The attacker plants malicious skills in the library without invoking them. The agent then authors and executes new skills carrying the harmful code. Each authored copy can re-enter the library and be imitated again, forming a self-propagating worm that persists after the planted skills are removed. We define the agent self-poisoning rate (ASPR) as the fraction of tasks that add a newly authored malicious skill to the library. Across six models on 153 tool-relevant SWE-bench Verified tasks, ASPR ranges from 20.3% to 41.8%, and the poisoned libraries hold 4.9 to 9.0 times as many malicious skills as were planted. The vulnerability also appears without a banner: DeepSeek-V4-Pro reaches 11.1% ASPR with the payload alone. Tailoring the planted skill descriptions to one task family raises ASPR to 86.7%. After the planted skills are removed, Qwen3 retains a round-5 ASPR of 68% because agent-authored copies remain. These copies evade existing defenses, which focus on attacker-submitted names, code, and signatures. We propose counter-prompt, a defense that discourages banner-style copying and reduces EvoMal's ASPR to at most 6.7% with no significant task-completion loss.

Large Language Models (LLMs) exhibit strong problem-solving abilities, positioning them as promising agents for Socratic teaching to guide students through step-by-step heuristic questioning. However, existing approaches typically adopt a one-problem-one-solution paradigm, restricting the teaching guidance to a single linear reasoning path. This design limits instructional flexibility, weakens error recovery, and restricts students' ability to engage in parallel thinking to explore multiple valid solutions. To overcome these, we propose ToST, a Tree-of-Thought Socratic Teaching framework that explicitly supports multi-path guidance under a one-problem-multiple-solutions paradigm. ToST employs Parallel Sowing, a parallel-thinking-oriented questioning strategy to encourage students to approach problems from diverse perspectives, and a Multi-Path Adaptive Guidance mechanism to provide more robust and non-linear instructions across alternative solution trajectories. Concurrently, to fill the void in systematically evaluating such non-linear instructional capabilities, we advance the task of multi-path Socratic guidance by establishing MPSG-Bench, a comprehensive benchmark that includes a dataset of 31K multi-path teaching dialogues and a five-dimensional evaluation framework grounded in the SOLO (Structure of Observed Learning Outcomes) theory to assess parallel-thinking guidance. Experimental results demonstrate that ToST significantly enhances guidance success rates while empowering students to navigate and explore multiple solution paths more effectively under both automatic and human metrics.

EXAONE Tabular is a compact tabular foundation model family for classification and regression via in-context learning, producing predictions without dataset-specific gradient updates. Pretrained exclusively on a synthetic structural-causal-model (SCM) prior, its central contribution is an architecture-centered redesign of tabular in-context learning. Rather than compressing features into a fixed row embedding before a separate row-level learner, EXAONE Tabular interleaves feature-axis attention within each item with support-conditioned item-axis attention within each feature at every Transformer layer, mediated by item-summary and feature-summary tokens. Across four public benchmarks, EXAONE Tabular combines strong predictive performance with high efficiency. On TabArena, its 20.81M-parameter classification model ranks first overall, surpassing tuned ensembles and 4-hour AutoML pipelines, while regression reaches the performance regime of the 1.64B-parameter TabFM at roughly 1/11 the inference cost. On BCCO and TALENT, EXAONE Tabular ranks second in classification and first in regression. On ScoringBench, it achieves the best mean rank for both point-estimation and predictive-distribution quality, leading the $R^2$, RMSE, and CRPS evaluations. Together, these results establish EXAONE Tabular as a state-of-the-art compact tabular foundation model family, combining strong predictive performance across classification, point regression, and probabilistic regression with an efficient model design.

Understanding user preferences from noisy and temporally evolving social media behaviors is fundamentally challenging due to interest drift, where user preferences shift across time and exhibit both multi-scale temporal patterns and diverse co-existing interests. To address this, we propose DUMoE, a unified framework for drift-aware multimodal user representation learning. Our model consists of (i) a temporal dynamics-aware backbone that captures and integrates static profiles, short-term behavioral signals, and long-term dependencies into a coherent representation, and (ii) a sparse mixture-of-experts (MoE) interest adapter that disentangles multiple latent interests via expert specialization and adaptive routing. Each expert models a distinct interest subspace, while a gating network dynamically selects and aggregates a sparse subset of relevant experts for each user. To enable stable and effective optimization, we further introduce a three-stage training strategy that decouples backbone learning, expert specialization, and gating optimization. Extensive experiments on real-world social media datasets show that DUMoE consistently outperforms state-of-the-art methods on both user interest prediction and interaction prediction tasks.

In serious illness, human surrogates often struggle to accurately predict patient preferences (68% accuracy), causing decision conflict. Personalized Patient Preference Predictor (P4) agents offer a potential solution, but prior prototypes treat values as static ratings, ignoring the contextual, situation-dependent nature of medical choices. Grounded in the 'logic of care', we present P4-DT (Dilemma Training), a P4 agent that constructs a patient decision policy by engaging users with varied medical dilemmas, eliciting individual preference reasoning through bi-directional training. In a study with 12 patient-surrogate dyads, P4-DT predicted patient treatment choices with 81.7% accuracy, significantly exceeding chance (OR = 5.61 [2.03, 15.51], p < .001) and outperforming both unassisted surrogates (55.0%; OR = 3.67 [1.59, 8.47], p = .002) and surrogates assisted by P4-DT (61.7%). Comparative prompt analyses showed that incorporating contextual scenario decisions and open-ended text improved accuracy by 15.0 percentage points over initial values ratings alone. We discuss implications for further testing and designing of context-aware AI agents that embody richer human experience to partner in complex decision-making.

Large language models (LLMs) have enabled AI scientist systems to automate scientific discovery, yet existing approaches most rely on static prompting or fixed workflows and fail to accumulate experience for continual improvement. We propose HypoForge, an experience-guided multi-agent framework that learns reusable scientific skills for automated hypothesis generation and hypothesis testing. HypoForge is built on the observation that these two stages involve different supervision signals. For hypothesis generation, where explicit feedback is unavailable, HypoForge adopts an adversarial generator–discriminator mechanism to improve reasoning through comparative critique. For hypothesis testing, where empirical feedback is available, HypoForge learns testing skills from execution outcomes and ground-truth results. By matching skill learning strategies with stage-specific supervision, HypoForge enables continual improvement without fine-tuning foundation models. Experiments on hypothesis generation and testing benchmarks show that HypoForge consistently outperforms existing AI scientist frameworks and skill-level variants. Further analysis demonstrates the effectiveness of the proposed stage-specific skill learning paradigms.