Papers

Latency-sensitive edge AI services must balance strict deadlines, output quality, and limited compute and energy budgets. However, static CPU provisioning wastes resources because inference cost varies substantially across inputs, model variants, and runtime conditions. We present PRISM, a prediction-guided runtime framework that jointly selects model variants and CPU allocations for containerized edge microservices. Using container-level energy monitoring and lightweight regression models, PRISM adapts each pipeline stage in place and minimizes predicted CPU-package energy under deadline, resource, and offline model-level Quality of Result (QoR) constraints. We evaluate PRISM on more than 52,000 requests in an Automatic License Plate Recognition (ALPR) pipeline with detection and recognition stages. For detection, PRISM reduces energy consumption by 36 % compared to the strongest static configuration while preserving a comparable success rate and using less than half of the average CPU allocation. For recognition, it reaches near-static-best performance with lower average CPU allocation. These results show that predictive in-place adaptation is a practical mechanism for making time-sensitive AI microservice pipelines more energy-efficient at the edge.

While Multi-Task Learning (MTL) is essential for adapting Large Language Models (LLMs) to diverse domains, prevailing LoRA-based methods rely on complex routing mechanisms that partition task-specific knowledge. In this work, we reveal that such routing-based designs are prone to a training-inference discrepancy, where stochastic routing decisions under distribution shifts compromise inference stability. Driven by a second-order Taylor analysis that exposes the instability induced by routing variance, we challenge the training-inference discrepancy and propose Consistency-Driven Low-Rank Adaptation (CD-LoRA). By eliminating routers entirely, CD-LoRA employs a consistency-driven alignment mechanism to enforce representation congruence across tasks in a shared low-rank space. This paradigm fosters robust, task-agnostic features without explicit partitioning overhead. Extensive experiments show that CD-LoRA consistently outperforms state-of-the-art multi-adapter baselines, offering a simpler, router-free, and more stable solution for multi-task PEFT. The code is available at the anonymous link https://github.com/zhaqian21/CD-LoRA.

We present a route-resolved comparative assessment of Rigetti's Cepheus-1-108Q and IBM Heron-r2 processors using the established do-nothing state-transfer protocol. Rather than proposing a new protocol, we use this deterministic, low-complexity task as a high-resolution spatial probe. For each evaluated initial qubit, we report two complementary quantities: the largest tested radius within which every evaluated shortest route satisfies the operational success rule, and the longest successful route identified within the evaluated route family. To address the question "How far can you do-nothing on a quantum computer?", we prepare a quantum state, route it outward using SWAP gates, apply the inverse preparation at the destination, and route it back for measurement. We compare the measured probability of returning $0$ with the empirical cutoff $2/3$. For this circuit ordering, that comparison does not certify above-classical unknown-state transfer. While this trivial state-transfer protocol serves as the most intuitive baseline, actively preserving a quantum state across a physical lattice proves to be a non-trivial task that exposes the information to cumulative relaxation, dephasing, and environmental cross-talk. In the highlighted IBM QPU case, we identify an isotropic radius of 10 and a successful path of swap distance 27, whereas the highlighted Rigetti Cepheus case exhibits an isotropic radius of 1 but selected above-threshold routes reaching swap distance 8. These results reveal a sharp distinction between uniform spatial reliability and best-route performance. The presented quantities are empirical and conditional on the evaluated route families, finite-shot decision rule, calibration state, and execution time; they are not architecture-wide constants.

Human-in-the-loop real-world reinforcement learning enables rapid acquisition of effective robotic manipulation policies for individual tasks, often within tens of minutes. Yet it remains unclear how to extend this paradigm to continual learning, where a single policy must acquire new skills without losing previously learned behaviors. Existing real-world continual learning methods do not explicitly constrain prior behaviors, leading to severe catastrophic forgetting. We introduce Continual Interactive Distillation for Embodied Reinforcement Learning (CIDER), a continual reinforcement learning framework that freezes the accumulated historical policy as a teacher before learning each new task and interleaves task learning with distillation-based retention. We further introduce gradient routing to separate the gradients used for acquiring new tasks from those used for preserving prior behaviors. We evaluate our method with a single shared actor on six real-world household and industrial manipulation tasks. Interactive Distillation maintains high measured success on previously learned tasks across our six-task real-robot sequence while acquiring each new task in 10 to 20 minutes, whereas every baseline forgets at least one previous task. Additional ablations reveal the key design choices that govern the tradeoff between stability and plasticity in real-world continual reinforcement learning.

Web agents promise to automate complex digital workflows, but their training remains limited by synthetic environments that look plausible while hiding broken links, inconsistent states, or infeasible tasks. We address the gap between scalable environment generation and trustworthy agent learning by constructing synthetic web environments that are executable, auditable, and grounded in backend state. Our framework represents each generated website as a structured scaffold of pages, navigation links, database records, state-change markers, and task constraints, then verifies and repairs structural, semantic, consistency, and feasibility defects before policy training. During interaction, ordinary UI transitions are executed deterministically, while persistent backend updates are invoked only through validated state-change markers, enabling dense rewards compiled from verified task-progress predicates. Across 500 synthetic environments spanning six domains, our method reduces task-blocking defects and improves feasible-task rate from 48.6% to 94.8%, while producing stronger PPO policies and improving transfer to WebArena, WebShop, and MiniWoB++ without LLM calls at evaluation time. These results show that verified synthetic environments can serve as a scalable and reliable training substrate for compact web agents, shifting synthetic webagent learning from surface-level plausibility toward executable, state-grounded supervision.

Reliable planning requires converting natural-language instructions into executable symbolic specifications, yet large language models remain brittle without costly PDDL annotations and may exploit solver success in semantically unfaithful ways. We study how to learn faithful natural-language-to-PDDL formalization using only solver feedback, without human-written demonstrations. We propose a solvergrounded multi-role reinforcement learning framework where a single language model acts as an Actor, Judge, and Editor for generation, verification, and repair. The Actor proposes PDDL specifications, the Judge provides a solver-calibrated quality signal, and the Editor performs bounded diagnostic-conditioned refinement. On PlanBench, our method improves average success from 35.5% for LLM+P to 70.8%, achieves 66.3% faithful success, and reduces semantic drift to 6.4%. These results show that organizing solver feedback into generation, verification, and repair roles enables more scalable and faithful annotation-free symbolic planning

Locally deployed Large Language Models (LLMs) via inference engines such as Ollama run without the moderation and abuse detection present in API-served models. Therefore, the safety of LLMs depends on the defense mechanisms used, and their effectiveness depends on the assumptions on which they were designed. This paper does an audit of defense mechanisms under jailbreak attacks on locally deployed models. Some defenses provide formal guarantees (SmoothLLM, Erase-and-Check, Sequential Monitors), while others rely on empirical detection results (Semantic Smoothing, Self-Denoised Smoothing, Perplexity Filtering). Instead of merely observing that defenses fail, we trace each failure back to the specific assumption: for every defense, we extract the condition it relies on, derive the empirical pattern a violation should produce, and test that prediction on six open-weight models (14B to 35B parameters) with a corpus of 100 jailbreak prompts taken from more than 40 public sources, totalling 13,800 evaluation records.

Human tactile perception relies on complex multisensory cues. Yet the relationship between tactile signals and perceptual representations remains poorly understood, limiting the integration of touch in digital environments and human-like robotic perception. To address this gap, we developed a computational framework comprising three interconnected deep learning models that map multisensory touch data to material perception, without relying on hand-crafted features. The models represent progressively different routes from tactile signals to material class: from low-level interaction signals to perceptual attribute distributions (Model 1), from predicted attribute distributions to material classification (Model 2), and directly from tactile signals to material categories, bypassing intermediate representations (Model 3). By combining deep learning with Integrated Gradients, the framework achieved high accuracy while offering interpretability, revealing which sensory modalities most strongly drive its decisions. Our results show that deep learning can approach near-perfect material classification when unconstrained by intermediate perceptual stages, but matching human-like performance is harder once those stages are modeled explicitly. Notably, thermal cues emerged as particularly informative across all models, providing robust signals for material differentiation. The results offer a computational account of how tactile signals lead to material perception and show how interpretable deep learning can both approach human-level performance and reveal cues that robotic and haptic systems need to incorporate.

Scalable Vector Graphics are a fundamental medium for resolution-independent visual content, yet the deep learning community lacks a continuous, dense, and invertible latent space for vector representations, the kind of foundational building block that Variational Autoencoders and their descendants have long provided for raster images. We introduce SLS (SVG Latent Space), a Transformer-based autoencoder that learns compact dense representations of individual SVG paths, the atomic visual elements from which any SVG image can be composed. By modeling SVG commands, coordinate data, and visual properties within a unified BPE-based token vocabulary, SLS learns fixed-size latent representations that jointly capture structure and appearance, and can be decoded back into valid, style-consistent SVG paths with high fidelity. The resulting embedding space is robust, invertible, and structured: embeddings lie on a unit hypersphere, enabling efficient similarity search, composition, and downstream conditioning through simple vector-space operations. Finally, we demonstrate that SLS generalizes across diverse tasks reducing their FLOPs by over 150 times compared to token-based approaches, and establishing a general-purpose latent foundation for vector graphics research.

Electronic circular dichroism (ECD) is a key experimental probe for assigning the absolute configuration of chiral molecules, but interpreting measured spectra often requires costly time-dependent density functional theory (TDDFT) calculations. The few existing machine learning methods for ECD prediction are faster, but their accuracy still falls well short of that required for experimental applications. Here we introduce PhysECD, a parity-aware E(3)-equivariant framework that builds the spectrum from its underlying quantum-chemical quantities, ensuring correct chiroptical symmetry by construction. To improve generalization and support experimental applications, we construct SCECD, a conformer-resolved dataset of 578,104 computed spectra spanning diverse natural products with one to five stereocentres. PhysECD predicts ECD spectra from conformer geometries in milliseconds, achieving state-of-the-art accuracy on the CMCDS benchmark and SCECD. Validation against 77 experimental spectra further demonstrates that PhysECD can serve as a fast alternative to TDDFT for assigning the absolute configuration of complex chiral molecules.

Runtime thermal management of high-performance chips depends on fast and accurate full-chip thermal maps. Conventional simulators typically estimate power traces from performance metrics first, which adds overhead. This work proposes TherMapNet, an attention-guided thermal simulator that predicts full-chip thermal maps directly from performance metrics. A Transformer encoder captures temporal evolution by treating the time series of each metric as a token, improving modeling of dynamic workloads. A CNN then extracts fine-grained spatial features. For the CNN, a dual-branch channel-spatial attention convolution module (DACM) and a triplet loss are used to improve spatial learning and reconstruction accuracy. TherMapNet is applied to a multi-core CPU (AMD Ryzen 7 4800U) and a many-core GPU (NVIDIA GeForce RTX 4060). Experiments show that it outperforms prior thermal simulators, with RMSE below 0.26 C and inference under 2.4 ms on an NVIDIA GeForce RTX 3090 GPU. These results indicate that TherMapNet can support high-quality runtime thermal management of modern multi-core chips.

Use of artificial intelligence in motor control applications that can be deployed within the microcontrollers need to comply with real-time demands of motor control systems. A supplementary controller approach that can be integrated within existing microcontrollers is presented. The proposed approach is implemented on a weeteq motor control integrated circuit and tested in the lab. A complete unsupervised motor control deployment solution was developed, and the real-time system response correction demonstrated under dynamic loads measured with and without the supplementary controller. The solution provides 100% coverage of real-time data at control loop sample rate and model inference period between 100usec and 1msec. The importance of latency for reduction of dynamic regulation margin during transient response is demonstrated with up to 68% reduction of the system response error. A novel key performance indicator based on principal components transform is introduced that provides a quantitative figure of merit for improvement of the transient response, in terms of the dynamic regulation margin, stability considerations and iterative improvements of consecutive regression model outputs. The significant events related to dynamic changes in the equipment and external operating conditions are detected at milliseconds resolution and recorded as highly compressed vectors representing deviation of the system response from linear steady state conditions. The resolution in time and accuracy of this vector data will enable a new level of system level optimisation that has not been possible using the time series data from IoT sensors in current equipment health monitoring solutions.

Pixel-space diffusion models directly model image distributions but remain difficult to optimize. Recent methods alleviate this challenge through target reparameterization, while still relying on a fixed clean-image target throughout denoising. Through empirical analysis, we identify a scale-time mismatch: image structures become predictable from coarse to fine as noise decreases, whereas existing models are forced to predict the full image even under high noise, resulting in low-SNR gradients that hinder optimization. To resolve this mismatch, we propose Observation Operator Diffusion, a unified framework that aligns both the supervision trajectory and feature refinement with the intrinsic recovery order of image structures. Specifically, we replace fixed full-image supervision along the standard flow path with a time-indexed observation trajectory that evolves from coarse structures to the full image during denoising. This trajectory is instantiated with a family of Gaussian-Lanczos operators at varying observation scales, yielding a path-consistent training objective. We further introduce GL-CoDA, a decoder that injects scale-specific Gaussian-Lanczos observations across decoding stages for coarse-to-fine feature refinement. Extensive experiments show that the proposed approach converges substantially faster while consistently improving generation quality, achieving an FID of 1.52 on ImageNet-256.

Over the past months, the way developers guide agentic AI coding tools has moved up several levels of abstraction, from phrasing prompts to engineering context to configuring the harness around the underlying model. In June 2026, practitioners began to describe a new level called loop engineering: Instead of prompting an agent interactively, developers design systems that prompt agents for them. The term spread rapidly, accompanied by bold claims and vocal skepticism, but its adoption in software projects has not yet been measured. We present (1) an exploratory review of the emerging gray literature on loop engineering, (2) a traceability analysis and an exploratory mining study of 36,710 software repositories, and (3) the design of a controlled experiment to study how loop goals, static analysis checks, and decision policies affect task success, code quality, resource usage, and token consumption of agent loops. Our literature review found broad agreement on what a well-engineered loop contains: agent loops started on a schedule or by repository events and bounded by machine-checkable stop conditions, persistent state files, verifier subagents, token budgets, and defined points of escalation to humans. These sources provide no systematic evidence of the claimed benefits. In the mining study, we found autonomous agent loops in 217 open-source GitHub repositories using heuristics and behavioral signals. Nearly all of these repositories committed the configuration that starts their loops, but almost none of them committed the state files that the gray literature suggests. Our proposed experiment is a first step toward closing the evidence gap identified in our literature review. We call for further empirical studies of the benefits and costs of loop engineering for software engineering tasks.

Multimodal large reasoning models often rely on long Chain-of-Thought (CoT) traces in which a substantial fraction of tokens, such as repeated visual descriptions, self-reflection, and other visually-disengaged filler, inflate inference cost without contributing to the answer. Existing CoT compression methods optimize output length but never measure whether a reasoning token is actually grounded in the image. We propose VIG (Visual Information Gain), an information-theoretic GRPO reward that scores each reasoning token by how much the image reduces its predictive uncertainty. VIG is computed online from two forward passes of the same policy, one with and one without the image, so no reference chains, external annotations, or auxiliary reward models are needed. Across six main multimodal reasoning benchmarks and three Qwen3-VL-Thinking model sizes (2B/4B/8B), plus an additional R1-Onevision-Bench evaluation on 8B, VIG consistently improves the accuracy–efficiency trade-off, supporting our central claim: efficient multimodal reasoning emerges from raising visual information density, where every reasoning token earns its place by anchoring to the image, rather than from imposing a length budget. Our source code is available at https://github.com/chaser682/vig.

We describe our approach to the Big Cross-Modal Attenuation Correction (BIC-MAC) challenge, which asks for a pseudo-CT in Hounsfield Units to be synthesized from Non-Attenuation-Corrected PET (NAC-PET), DIXON MRI and a topogram, and scores both the pseudo-CT and the Attenuation-Corrected PET (AC-PET) reconstructed from it. Three ideas carried our improvements over the organizers' 3D U-Net baseline. The loss matters more than the architecture: we compute the $L_1$ error in the Carney attenuation-coefficient ($\mu$) space that the CT metric itself uses, weighted by anatomical region. Only once that loss was in place did the unregistered DIXON MRI work as extra input channels. A fixed convex combination of two independently trained models then beat both of its members on three of the four metrics and ranks first overall on the public validation leaderboard.

Bangla large language model (LLM) safety is difficult to evaluate with English-centric or standard-script benchmarks because Bangla users routinely write across scripts, spellings, code-mixed forms, and regional registers. This paper presents BanglaVeilGuard, a compact Bangla-first safety benchmark and lightweight prompt guard for six language forms: standard Bangla, Romanized Bangla, Banglish, code-mixed Bangla–English, noisy Bangla, and dialectal Bangla. The benchmark contains 2,366 quality-filtered prompts and a held-out 354-prompt evaluation split spanning unsafe, safe, and safe-sensitive requests. BanglaVeilGuard uses non-destructive multi-view normalization with a prompt-risk classifier and thresholded pre-generation gate, allowing it to screen prompts for heterogeneous target models without changing their weights. Across target-model families, guarded runs reduce attack success under deterministic response scoring from 93.8–100.0% to 6.3% for Claude Opus 4.8, BanglaLLama, and TituLLM; TigerLLM-1B with BanglaVeilGuard achieves 78.2% accuracy with 8.8% ASR. The prompt guard also attains 88.5% unsafe recall, substantially above the evaluated prompt-only guard baselines. The main remaining cost is over-refusal on dialectal and noisy benign prompts, revealing a concrete safety-helpfulness frontier for Bangla LLM deployment.

Aerial payload transport using multirotor unmanned aerial vehicles is challenging because payload geometry, contact interaction, grasp stability, flight control, and narrow-passage traversal are strongly coupled during pickup and transport. Object-specific grippers often cannot adapt their footprint or grasp geometry when the payload shape or passage width changes. This paper presents a vision-guided morphing quadcopter for multi-geometry payload transport through narrow passages. The proposed platform uses four hybrid arm-leg structures that function as both landing supports and grasping members. A centrally placed actuator drives a tendon-based morphing mechanism, enabling all four arms to synchronously retract or expand for object grasping, footprint reduction, and post-transport release. Onboard vision estimates the payload geometry and passage width, while endpoint force feedback is used to confirm grasp contact during payload engagement. A phase-wise mission planner, PID-based flight stabilization, and morphology-adaptive grasp controller are implemented in a MuJoCo simulation environment. The framework is evaluated using box, cylindrical, and spherical payloads, representing flat-faced, rolling-curved, and fully curved contact conditions. Across the three cases, the simulated system completes the pickup-transport-release sequence with a maximum RMS position error of 0.31 m, a final drop-zone error below 0.18 m, a compact grasp footprint of 0.09-0.21 m2, and a footprint reduction of 75.0-89.7 percent. The results demonstrate that a single-actuator morphing quadcopter can adapt its grasp footprint for the transport of payloads with different geometries while reducing its overall footprint for narrow-passage traversal.

Embodied Referring Expression Grounding is the task of enabling an agent to navigate in real environments and to localize a remote object based on natural language instructions. In this scenario, the agent needs to select one view for navigation at each step and identify a specific object among all candidate objects at the destination. However, most of the previous approaches fail to distinguish between views and objects, instead processing them using the vanilla vision encoder, which results in ambiguous representations of both views and objects. To address the above issues, we propose ViSMoE, which equips sparse Mixture-of-Experts with a visual-aware routing policy for the embodied agent. This framework processes different types of visual information specifically, resulting in discriminative visual representations for both views and objects. Experimental results on REVERIE and SOON datasets demonstrate that ViSMoE outperforms the previous state-of-the-art methods, showing the superiority of our proposed method.

Carousel interfaces have been widely used in video and music streaming services, yet it remains unclear how to properly evaluate recommender systems in these two-dimensional layouts. N2DCG has been proposed to address this gap by adapting NDCG to carousel-based recommendation, but it relies on unverified assumptions borrowed from the single-list web-search setting that do not transfer well to two-dimensional carousel layouts. We identify two substantial limitations of N2DCG: its ideal ranking, used for normalization, violates carousel constraints, and its discount function does not reflect user browsing behavior observed in empirical data. To address both limitations, we propose a reformulation of N2DCG that normalizes appropriately by respecting constraints and uses an empirically grounded discount function. We validate the proposed metric, showing that it better reflects users' empirical behavior on real-world eye-tracking data and better predicts the comparison results of carousel layouts simulated based on empirical examination patterns.

Deep learning has achieved strong performance in encrypted traffic classification (ETC), yet its computational cost limits deployment on resource-constrained network devices such as routers and middleboxes. Existing compression methods mainly operate on weights, channels, hidden representations, or predictions, but do not explicitly determine which protocol fields and structural contexts should remain. We propose Pruned Traffic Trees (PTT), a three-level protocol-structured model family that treats native protocol structures as compression units. PTT-Full learns protocol-structured representations and field salience from complete Protocol Tree Graphs (PTGs), with flow-level self-supervised learning and protocol-presence-aware sparse execution. The learned salience and TopK+$k$ closure construct Distilled PTGs (PTG-Ds) for PTT-Distilled, while PTT-Lite inherits this topology and reduces width through structure-aligned transfer and flow-level logits distillation. Under flow-disjoint and Strong Information Information (SII)-masked settings, PTT-Full achieves Macro-F1 scores of 0.9519 and 0.9416 on CSTNET-TLS1.3 and CipherSpectrum, while PTT-Lite retains 0.9325 and 0.9136 with 80.3% and 61.3% fewer parameters, 98.85% and 98.78% lower effective GFLOPs, and 8.75$\times$ and 8.46$\times$ CPU inference speedups. These results demonstrate that treating protocol structure itself as the compression object enables effective performance-efficiency trade-offs for lightweight ETC.

Reinforcement learning with verifiable rewards has become the dominant recipe for improving large language model reasoning, yet it presumes large human-curated task collections. Zero-data self-play removes this dependency, but existing methods vet learnability only by probing candidates and rejecting post hoc, never learning where along an environment's difficulty axis to place a task, and credit the solver with sparse terminal rewards alone. We recast zero-data self-play as a pursuit-evasion game: in LURE, an LLM evader positions tasks along each environment's difficulty axis to stay one step ahead of a planner-executor pursuer that hunts it down through verifiable interaction. The evader is trained on a capture-frontier reward that peaks when the solver captures it on exactly half of its rollouts, turning barely catchable into a learned positioning strategy rather than a hand-tuned rejection band. The pursuer earns capture-anchored dense process credit, in which monotone verifier progress is group-normalized jointly with the terminal capture under a round-anchored KL that keeps the co-evolution stable. Across three verifiable reasoning environments and three backbone families, LURE outperforms advanced baselines under unified/specialist settings, while the unified model attains stronger aggregate OOD zero-shot accuracy than all trained baselines across nine held-out benchmarks from three task families.

Text-to-image (T2I) diffusion models offer powerful visual generation, but their controllability creates a critical safety challenge: adversarial prompts can steer the denoising trajectory toward policy-violating content such as explicit nudity or graphic violence. Existing safeguards mostly act before generation through prompt filtering or after generation through image classification, leaving the diffusion process itself unguarded and often yielding only refusal rather than safe visual repair. We introduce GuardPaint, a speculative decoding framework for safe T2I generation that intervenes inside the diffusion trajectory without modifying the base model. A lightweight auditor monitors intermediate images, localizes unsafe regions, and triggers surgical inpainting repair only where needed. Candidate repairs are generated by a policy-aligned inpainter and selected through a guarded tournament that accepts edits only when they improve policy compliance while preserving prompt fidelity and perceptual quality. Across five jailbreak families SneakPrompt, MMA, PGJ, DACA, and RABell and UNet/flow-matching models including SD~1.5, SDXL, SD~3.5, and FLUX.1-dev. GuardPaint reduces attack success and harmful generations with minimal degradation to image quality, prompt fidelity, and benign behavior. Content warning: This paper contains examples involving nudity and violence that some readers may find disturbing, distressing, or offensive.

Depression assessment from multimodal clinical interviews requires integrating dispersed evidence from multiple symptoms into a coherent PHQ-8 profile. This process is hierarchical: relevant evidence is often sparse and context-dependent within local question-answer exchanges, multiple exchanges jointly support symptom-level judgments, and the final assessment depends on the coherence of the complete symptom profile. Existing LLM systems either process interviews holistically or distribute work across generic agent roles; neither design necessarily provides an explicit orchestration mechanism that coordinates evidence access, item-score authority, bounded feedback, and state recording across these levels. To address this gap, we introduce HiMA-MDD, a hierarchical multi-agent harness that aligns this assessment hierarchy with three agent layers. After non-agentic preprocessing constructs context-preserving multimodal QA units, Layer 1 identifies candidate QA-to-item relations and supports bounded item-grounded evidence routing. Layer 2 assigns symptom groups to operational factor specialists, with one specialist responsible for each provisional item score. Layer 3 audits the complete provisional profile, requests at most one round of targeted revision, and reconstructs the verified PHQ-8 profile. This layered design naturally yields a Hierarchical Evidence Trace, preserves all intermediate evidence, judgments, and revisions for auditability. The final item scores then deterministically produce the total score and screening decision. Using Qwen2.5-72B-Instruct as the harness backbone, our experiments on E-DAIC demonstrate that HiMA-MDD outperforms the compared state-of-the-art methods.

LLM agents are moving from single-prompt use to long task streams in which reusable memory becomes a core capability for terminal, software-engineering, and web tasks. Such memory is useful only when stored experience remains reliable across hundreds of interactions, but two failure modes break that assumption in practice. The first is unreliable admission: failed trajectories,accidental successes, and misleading observations enter memory because they appear relevant, then mislead later decisions. The second is memory drift: long-running banks accumulate duplicate, stale, and conflicting records that retrieval alone cannot repair. MemGuard's key distinction is to treat verifier output not as a one-shot filter, but as persistent lifecycle metadata. It converts multi-criteria score-token verification into reward, confidence, label, and uncertainty descriptors that are attached to every candidate before activation and reused during retrieval, conflict resolution, summarization, and archival. We evaluate MemGuard on Terminal-Bench 2.0, SWE-Bench Verified, WebArena, and Mind2Web across four backbones, comparing against four memory baselines plus a verifier-only control under matched runtime budgets. Averaged over five seeds, MemGuard achieves the best success metric and lowest average steps in all 16 backbone-benchmark settings, improving over ReasoningBank, the strongest prior baseline among the memory methods we evaluate, with a largest gain of 7.9 success-rate points on WebArena, 5.6 step-success-rate points on Mind2Web, and 2.4-3.5 points on terminal and software-engineering benchmarks. Code is available at https://github.com/whyyyyy123/MemGuard.

In Industrial Internet of Things systems, learning-enabled Digital Twins (DTs) support remote monitoring by using data reported by distributed devices to maintain digital representations of physical processes. When uplink resources are limited, a base station cannot collect new observations from every device at every communication slot and must decide which devices should transmit. This decision becomes challenging when the physical process changes after the DT models have already been trained. In such cases, recent observations alone may not keep the DT accurate, because the learned model may no longer match the underlying process. This paper studies how to schedule observation requests so that the DT maintained at the base station remains close to the true physical process. We define the Ensemble Disagreement Indicator (EDI) as an uncertainty measure computed at the base station from the spread among estimates produced by independently trained DT predictors. Building on EDI, we propose R-VoU, a reliability-aware value-of-update scheduler that prioritizes the observations expected to most improve the DT by reducing a cost based on uncertainty and predicted DT error. The DT is further adjusted online using the difference between received observations and current predictions. Experiments on process manufacturing data show that, under limited communication budgets, R-VoU achieves the lowest combined cost and DT estimation error among the compared schedulers that use only information available before each scheduling decision.

The current progress of Clinical Vision Large Language Models (C-VLLMs) has substantially improved digital diagnostics, still these frameworks often endure lesion noises, modality misalignment, hallucination, and missed contextual grounding in complex clinical cases. Moreover, prevailing agent systems usually depend on static and non-adaptable pipelines and lack the versatility necessary for complex medical reasoning. To resolve these difficulties, we present BioMed-Agent-RL, a unified medical agent that incorporates adaptive orchestration, policy, and reward-based reinforcement learning (RL) models for biomedical applications. To ensure reliability, it invokes clinical context-aware preference optimization (CPO), direct preference optimization (DPO), and group relative policy optimization (GRPO) with dynamic entropy regulation. This pipeline utilizes a multimodal meta-learning approach that operates as a field-specific expert and human judgment synthesizer. The agent adaptively utilizes a set of model-level expertise, such as clinical grounding and reasoner, lesion segmenter, and field-specific synthesizer, across various clinical modalities (e.g., X-ray) by utilizing an iterative and adaptive RL approach. The agent learns to seriously synthesize misleading, conflicting vision cues and trust in inherent reasoning, while specialist advice is faulty. An intensive ablation study is conducted across multiple benchmarks, and the agent significantly outperforms existing state of the art models, such as GPT-5, attaining up to ~73% accuracy (gain of ~5%) over contemporary baselines. As a result, the framework suggests a new standard for building factual, reliable, robust, and expert-like intelligent agent systems for independent clinical reasoning.

Tool-Integrated Reasoning (TIR) is a fundamental capability for LLM agents to solve complex tasks by interacting with external tools iteratively. Reinforcement Learning (RL) has become the dominant paradigm for enabling this capability. However, existing approaches typically assign uniform trajectory-level advantages and treat all correct tool calls equally, ignoring the varying difficulty and learning value across trajectories and reasoning steps. This can lead to imprecise learning signals that do not adequately distinguish between trivial and challenging tool-use patterns. To address this limitation, we propose HiDiffTIR, a Hierarchical Difficulty-aware policy optimization framework for multi-turn TIR. HiDiffTIR performs difficulty-aware credit assignment at both trajectory and turn levels, enabling the policy to focus on more informative trajectories and harder reasoning steps. Notably, this fine-grained optimization is achieved without additional supervision, relying solely on group-level statistics derived from standard RL rollouts. Extensive experiments on three tool-using benchmarks demonstrate that HiDiffTIR consistently improves multi-turn TIR performance and tool invocation accuracy over strong RL baselines, highlighting the necessity of difficulty-aware credit assignment for effective policy optimization in tool-integrated LLM agents.

Chain-of-Thought (CoT) reasoning has significantly enhanced the multi-step problem-solving capabilities of large language models (LLMs) by introducing explicit intermediate reasoning. However, advanced Large Reasoning Models (LRMs) often exhibit overthinking behaviors, including excessively long reasoning steps, redundant steps, and high computational overhead. Existing token-length reward strategies aim to promote concise outputs, but often result in pseudo-conciseness, where token count is reduced, yet redundant reasoning persists, leading to longer and less structurally efficient chains. To address these limitations, we propose ChainPrune, a novel reasoning path semantic structural optimization method to efficiently and controllably synthesize self-generated high-quality training data. We initially consolidate self-generated reasoning paths into a tree-based structure, followed by a multi-criteria dominant path selection process for preference data construction that formulates shallow reasoning trajectories while preserving essential reasoning steps. To further enhance the quality of reasoning, we incorporate a DPO-based preference learning method combined with supervised loss, effectively mitigating false reward suppression. This innovative integration significantly enhances both the efficiency and effectiveness of our reasoning framework. Comprehensive experimental results demonstrate significant reductions in step length and computational overhead, while maintaining or even enhancing accuracy.

The rapid evolution of energy structures has positioned microgrids as pivotal components of next-generation power systems, offering enhanced resilience and renewable energy integration. However, the inherent low inertia, complex dynamics, and poor model conditions of microgrids necessitate advanced data-driven frequency control strategies. Although reinforcement learning (RL) has demonstrated certain potential and advantages, existing RL methods often struggle with generalization across diverse microgrid configurations and lack adaptability to unseen environments, particularly when explicit system parameters are unavailable. To address these challenges, in this paper, we introduce a novel prompt decision transformer (Prompt-DT) architecture for microgrid frequency control. Unlike traditional approaches that rely on hard-to-obtain environmental characteristic parameters, the proposed method leverages few-shot expert historical trajectories as prompts to guide autonomous perception and adaptive decision-making. In addition, we propose a context-aware training and execution mechanism utilizing self-supervised contrastive learning to enhance environment recognition and prompt utilization efficiency. In addition, a physics-informed prompt design technique that filters prompts based on cumulative reward and frequency volatility is proposed, ensuring high-quality physical guidance during online execution. Finally, to ensure generalization in unseen environments with limited data, we develop a lightweight finetuning approach that achieves performance comparable to full-parameter finetuning with minimal adjustments.