Papers

Large Language Model (LLM) agents increasingly explore many candidate states of relational data in parallel, each of which should remain isolated, reproducible, and auditable, preferably through the same SQL interface used for ordinary data work. Existing tools support this requirement only partially: source-code version control does not scale to large datasets, whereas relational databases manage large data efficiently but rarely expose native branching, comparison, and merging. We present Git4Data, a database-native version-control layer for agentic workflows. Git4Data treats a database as a repository and a table as a versioned object, exposing Git-style operations (snapshot/tag, branch, diff, and merge with explicit conflict-resolution policies) through SQL extensions. Implemented in MatrixOne, a cloud-native relational database, Git4Data leverages immutable object storage and MVCC to make the cost of these operations proportional to the size of the change rather than the size of the data. On the BranchBench agentic branching workloads, Git4Data outperforms DoltDB by up to an order of magnitude. Overall, we believe this work sheds light on how relational databases can better support AI agents through efficient versioning.

Federated learning (FL) lets institutions train a shared model without exchanging data, and Low-Rank Adaptation (LoRA) makes this practical at scale by communicating only compact low-rank updates. Biomedical imaging is a compelling setting for this combination: patient data are archived behind privacy regulations, and institutions differ widely in scanners, protocols, and compute. Such heterogeneity raises the question of how federated LoRA updates should be aggregated, increasingly pressing as multimodal vision-language models become central to medical image analysis. We benchmark federated Parameter-efficient fine-tuning (PEFT) of BiomedCLIP for chest radiograph classification across four public cohorts on three continents (USA, Vietnam, Spain). Federated LoRA adaptation improves shared-class AUC on all four cohorts over the unadapted BiomedCLIP backbone (mean 0.687 to 0.802), showing that the gains come from federated adaptation rather than from the pretrained model's zero-shot ability. Relative to isolated single-cohort training, federation improves the weaker cohorts while largely preserving the strongest and approaches a centralized reference (0.812) that pools all data. The singular value decomposition (SVD)-based product-space aggregation introduced by FlexLoRA is essential to this gain (naive factor averaging drops mean AUC by 0.097), whereas a drift-correcting optimizer (FedProx) shows no benefit over FedAvg in our single-seed runs, consistent with LoRA's low-rank updates already limiting client drift. Biomedical vision-language models can thus be adapted collaboratively across heterogeneous, geographically distributed institutions without centralizing data.

An AI agent can perform well on benchmarks and still be unsuitable for deployment. Existing AI-agent benchmarks measure whether an agent can complete realistic professional work, whereas enterprise deployment asks a different question: whether an agent can meet a required reliability level, under acceptable human oversight, and at tolerable cost. We introduce Reliable Enterprise Agent Deployment (READY), a framework for qualifying AI agents for deployment on enterprise workflows. READY preserves each workflow's own definition of successful execution while applying a common qualification procedure. Given an agent, a workflow, and a class of candidate oversight policies, READY measures the reliability and operating cost of the human-AI system, selects the minimum-cost policy that satisfies a specified reliability target, and statistically qualifies it on held-out cases. The resulting deployment profile characterizes the supported operating point: reliability, human-oversight burden, and cost. READY is implemented as an open testbed that decouples workflow specification, execution, evaluation, and qualification, and runs on existing agent-evaluation infrastructure. In an end-to-end clinical-audit case study spanning 16 agent systems and 750 cases, READY reveals differences hidden by autonomous performance: two systems separated by only 0.3 points in autonomous accuracy (72.8% vs. 72.5%) require 39.2% versus 29.6% human review, respectively, to qualify at the same 76% reliability target under the evaluated oversight policy. READY thus shifts enterprise agent evaluation from how well can the agent perform the work? to under what conditions, and at what cost, can it be reliably deployed? By making those conditions explicit and statistically testable, READY provides a basis for comparing agent systems, setting oversight requirements, and making evidence-based deployment decisions.

LLM-based multi-agent systems have shown strong performance on complex tasks, yet continual improvement from interaction experience remains challenging. Existing self-reflection methods build experience memories, but memories are mostly hard to invoke, refine, or scale, while agent skills offer a more actionable unit: structured procedural knowledge that specifies when to act, how to act, and which resources or tools to use. We introduce MASkills, a continual learning framework that optimizes multi-agent LLM systems through agent skills. MASkills presents a new agent-optimization pipeline that integrates skill-conditioned credit assignment, hierarchical credit aggregation, and momentum-smoothed optimization, enabling agent skill libraries to evolve through refinement, induction, consolidation, and pruning. Experiments on HotpotQA, LoCoMo, and GAIA demonstrate the effectiveness of MASkills across multiple agentic tasks. Our code is available at https://github.com/DaRL-GenAI/MASkills

LLM-based multi-agent systems (MAS) are increasingly considered for high-stakes decision-making, yet outcome-based fairness audits can miss where risks arise within the decision trajectory. We present SCOPED-Hiring, a process-aware fairness diagnosis pipeline for LLM-based hiring MAS. SCOPED-Hiring constructs controlled resume variants, runs role-based hiring committees, logs over 311K structured decision trajectories, and converts trajectory fields into quantitative fairness signals organized by six diagnostic lenses: final outcome, counterfactual, process, pathway, dynamic, and design effects. SCOPED-Hiring reveals that balanced final hire rates can mask hidden trajectory unfairness in multi-agent decision trajectories: career gaps trigger suspicion, proxy cues shape qualification judgments, and identity cues lead to unequal investigation. Targeted repair guided by these diagnoses reduces total layered burden by 72.3% while shifting the hire rate by only 1.86 pp, showing that process diagnosis can guide effective repair. Project Page: https://scoped-hiring-project-page.vercel.app/

We present TropicalCycloneNext (TC-Next), a multimodal deep learning model that forecasts tropical cyclone track and intensity at $6-24$h leads by leveraging a foundation model's forecast of atmospheric kinematic and thermodynamic fields, along with GridSat infrared satellite imagery. Trained only on GraphCast forecasts over the Western Pacific (WP), yet reliant only on generic atmospheric variables, TC-Next on GraphCast lowers track error by $15-44\%$ and intensity error by a factor of $3-6$ relative to the rule-based tracker TempestExtremes; applied without retraining to Pangu-Weather and IFS HRES forecasts, it stays ahead on both. Applied zero-shot to the generic weather fields of WeatherNext Cyclones on the 2025 WP season, it attains lower average intensity error at every lead time, and lower or comparable average track error, than that model's specialized direct tracker in a deterministic comparison. Ablations show that the satellite branch lowers track error at every lead time and intensity error at longer leads. TC-Next thus gives a common, independently verifiable basis for comparing the AI forecasts that inform cyclone-warning and climate-adaptation decisions as best available science.

Visual modality enhances the capabilities of multimodal large language models (MLLMs) but also introduces a safety concern: a benign textual query may convey harmful intent when grounded in a visual image. We term this cross-modal safety drift and our pilot studies show that the safety response rate for such requests is substantially lower than that for requests containing explicitly unsafe text. This paper aims to systematically study this issue. First, we conduct an empirical analysis to identify representative unsafe response patterns. Building on these, we interpret model representations and attentions, revealing that visually risky cues receive limited attention and weakly trigger refusal. Motivated by the observation that safety signals from unsafe text processing can be transferred, we propose safety-awareness representation transfer (SRT), a lightweight direction-refinement method that mitigates cross-modal safety drift with a frozen MLLM backbone. Experiments across multiple benchmarks and models show that SRT effectively improves safety in diverse cross-modal settings while preserving utility. Code is available at https://github.com/cucu220123/safety-awareness.

Planning is a central capability that enables agents to decompose complex long-horizon tasks into manageable steps. Test-time search and training-based methods improve planning but incur high inference costs or require expensive training data. Self-evolving memory instead accumulates reusable experience from agent interaction outcomes into an external memory bank, so planning capability keeps improving at inference time without parameter updates. However, existing self-evolving memory methods share an inherent credit assignment problem: they rely on final task outcomes as feedback, but such outcomes conflate plan quality with execution errors and environmental factors, so the accumulated planning experience is often biased and noisy. To address this problem, we propose Credit-Aware Hierarchical Memory Evolution (CHIME), a self-evolving memory framework that maintains a separate planning bank and execution bank and follows an attribute-before-memorize principle: CHIME first attributes each task outcome to the plan, the execution, both, or neither, and then updates only the corresponding memory bank. Extensive experiments on four long-horizon agent benchmarks show that CHIME consistently outperforms state-of-the-art training-based and self-evolving memory baselines. Further analyses reveal several interesting findings. For example, CHIME accumulates effective memory with far fewer items. In addition, the learned memory values faithfully reflect downstream utility: high-quality planning memories are more valuable than execution memories. Finally, the accumulated memory effectively transfers across backbone models. Code will be released at https://github.com/ATH-MaaS/Marco-DeepResearch.

Scientific benchmarks are commonly built by domain experts who write tasks and cross-check one another's work, or who adapt existing material from textbooks, published papers, and online resources. These routes can produce strong evaluations, but they require substantial per-item labor. Language models can reduce this repeated work by proposing candidates quickly. The remaining problem is acceptance. We target scientific questions whose answers require computations with specialist software rather than unaided reasoning alone. A candidate is invalid if its script fails or returns a different answer, or trivial if a model answers it without the software. We present ToolGate, which treats every generated item as a proposal and keeps it only if three gates pass. First, an executable solution script must reproduce the proposed answer when run with the scientific software. Second, randomized no-tool screening rejects candidates that models can already solve from the prompt alone. Third, a tool-using agent must solve each survivor within a fixed time limit. We instantiate ToolGate in FEniCSx with 500 generation attempts. The local-verification gate retains 478 candidates. For final reporting, we rescreen this pool after generation: two randomized no-tool screens exclude 222 from the reported pool, and direct GPT-5.5 API calls at medium reasoning (the API default) exclude another 121. Of the remaining 135, a GPT-5.5 Codex CLI agent with access to FEniCSx solves 130; exact deduplication leaves 128 unique protocol survivors. ToolGate turns repeated answer checking and difficulty screening into an auditable process while leaving domain design and final review to experts.

Mineral exploration requires integrating heterogeneous geochemical, geophysical, and geological evidence, yet existing prospectivity systems often provide only opaque scores or heatmaps. We present MineTRACE, a web-based system for evidence-grounded exploration of eight commodities: Cu, Au, Ni, W, Sn, Co, Ta, and Mn. Users can explore prospectivity maps, query locations or regions, inspect supporting evidence, and interact through natural language. A transparent expert tree, informed by geological knowledge and known deposits, combines multi-source evidence into interpretable prospectivity scores. For a new location, the conversational assistant retrieves the score and supporting evidence from the analysis pipeline and presents them in natural language. The scorer achieves spatial AUC values of up to 0.917 across different test scenarios, while end-to-end evaluation assesses query accuracy and response grounding. MineTRACE makes public geoscience data easier to access, interpret, and verify, supporting more efficient and transparent mineral exploration.

Multimodal Large Language Models (MLLMs) have achieved strong performance on structured visual understanding tasks such as chart and document question answering. However, existing benchmarks typically evaluate these domains in isolation, leaving underexplored a key capability: whether models can use textual context to determine how chart evidence should be selected, interpreted, and aggregated. We introduce DocHop, a benchmark for integrated chart–context reasoning in document-style images. In DocHop, the document narrative specifies multi-step compositional constraints, while charts provide the corresponding data values. Questions are grounded on a semantic reference label defined in the narrative, requiring models to resolve target entities from context before aggregating evidence across multiple charts. To enable systematic evaluation, we construct DocHop via a stochastic logic-first generation pipeline with controllable reasoning depth and visual density, covering 2,074 examples across six task categories. Experiments on a wide range of proprietary and open-source MLLMs show a substantial gap to human performance: annotators achieve over 90% accuracy, while the best model reaches only 62.83%. Reasoning-enhanced models consistently show improved results, but performance degrades as reasoning complexity increases. Overall, DocHop provides a controlled testbed for challenging multi-hop document reasoning.

Reliable web-agent monitoring is difficult when model-internal uncertainty signals such as token logits are unavailable. In this work, we study prefix-level risk prediction for web agents using observable trajectory signals: given an evolving prefix, estimate whether the current execution remains on track or is tending toward failure. We derive two observable trajectory representations: Macro features summarize cross-step agent–environment behavior and feedback, while Micro features measure the consistency of intention, action, and anticipated state change through repeated black-box queries. Instead of inheriting the final result label, we label the first critical error that remains uncorrected in the observed continuation and is associated with final failure as a key-step boundary, preserving valid early prefixes of failed trajectories as on track. Across WebArena-Lite and Online Mind2Web web agent benchmarks with five open- and closed-source backbones, observable trajectory signals are competitive with internal-signal baselines. The resulting predictors also support early intervention under fixed false-cut budgets and transfer across held-out website categories. These findings show that observable trajectory signals support valuable risk prediction abilities.

Monolithic world models predict the entire next state at every step, spending capacity re-predicting the static majority of a scene and injecting error into it. We ask whether explicitly modeling change (a per-object change gate plus a residual delta head that perturbs only the objects the gate flags) is a more effective and interpretable bias for physical prediction and control. On a MuJoCo tabletop pushing benchmark scaling from 3 to 8 objects, the sparse/residual model predicts next-state poses 2.5 to 4.6 times more accurately than a dense multilayer perceptron at 8.6 to 11.1 times fewer parameters, sustains change-detection F1 of 0.80 to 0.87 where the dense baseline is degenerate, transfers across object counts with zero retraining (99.4 percent F1 retention), and reaches about 90 percent of its full-data accuracy with a quarter of the data. In autoregressive rollout it compounds far less error, hugging the no-motion floor while the dense model drifts. Finally, inside a sampling-based planner, prediction-only models fail (though a true-simulator oracle solves the task with the identical planner, confirming the planner is sound), but once featurized and trained for the states a planner visits, the sparse model begins to plan (0.23 plus or minus 0.06 success over three seeds) while the dense monolith stays at zero at every seed. Modeling what changes, rather than re-predicting the whole world, is a simple, effective bias for object-centric physical AI; code, data generators, and all checkpoints will be released upon publication.

Long-context inference retains a growing key–value (KV) cache during decoding, which consumes substantial GPU memory and can reduce generation throughput. This bottleneck remains in hybrid language models because their residual global-attention layers can dominate context-dependent cache demand. We study how to allocate this state under an aggregate KV-residency budget. We introduce HeadWiseKV, a training-free framework that compresses the residual global KV caches of hybrid language models while preserving their native local, recurrent, and linear paths. It assigns each physical KV head a static, multilevel history window, making cache demand predictable before serving. We formulate this allocation as a restricted operational rate–distortion problem and propose SeqCalib as the core policy-generation algorithm in HeadWiseKV. SeqCalib processes layers in execution order and conditions each decision on the lower-layer policy used at deployment, thereby accounting for interactions across depth. A grouped-cache runtime materializes the selected policy as actual per-head KV residency rather than a mask over a full cache. We evaluate downstream quality across four hybrid long-context models and study physical residency and serving behavior on Qwen3.6-27B. HeadWiseKV retains near-Full-KV RULER and LoCoMo quality across the evaluated models. In the fixed-model systems study, it reduces sampled peak device memory by 8.59% at a 112K context length and extends the largest verified successful context from 114K to 161K.

Large language model question answering over power-grid models must respect a fixed context budget. We introduce seed-anchored graph rendering, a deterministic method that prioritizes query-local graph evidence without adding method-specific tuned or learned parameters beyond the shared hop bound and context budget. The method provides a checkable condition under which predefined seed-local answer-bearing render units are preserved in a greedy bounded-context prefix. We evaluate the approach on Common Information Model (CIM) network models exchanged through the Common Grid Model Exchange Standard (CGMES). On two budget-binding CGMES encodings, naive descriptions-first rendering retains local evidence for every single-hop item but only 0.12 and 0.00 of multi-hop items, whereas seed-anchored rendering retains all such evidence. On a preregistered fresh 100-item bank from the SmallGrid topology family, accuracy rises from 0.450 to 0.970 under a fixed 8,000-character context budget. Under a common retrieval and rendering pipeline, the standards-native seed-anchored graph matches or exceeds extracted graph representations produced by LightRAG, Microsoft GraphRAG, and HippoRAG, while avoiding LLM graph-construction tokens. The results are specific to the evaluated CIM/CGMES models, reader, and context budget; they concern budget-bounded retrieval rather than general question answering.

Accurate segmentation of polyps and skin lesions is pivotal for clinical diagnosis, yet existing methods struggle with low contrast, ambiguous boundaries, and cross-domain distribution discrepancies. Discriminative networks and most diffusion-based segmentation approaches predict standalone binary masks, leaving the visual priors of large-scale pretrained generative models largely unexploited. We propose InstEditSeg, a unified generative framework that reformulates medical segmentation as an instruction-driven image editing problem. Instead of emitting a mask, the model renders a color-coded overlay on the original image, conditioned on a textual instruction, so that the edited output aligns with the natural image distribution learned by latent diffusion models and mitigates the domain gap between natural and medical imagery. To recover fine anatomical structures, we introduce DINOv3 as an auxiliary visual encoder and a DINO Feature Guidance Block that builds a multi-scale feature pyramid. The pyramid is fused into the diffusion U-Net by channel concatenation and zero-initialized convolution so that hierarchical discriminative priors can be injected without perturbing the pretrained weights. A dual-branch classifier-free guidance strategy requiring only two forward passes per denoising step reduces inference cost. On polyp and skin lesion benchmarks the framework achieves accuracy competitive with strong discriminative baselines, and it further demonstrates concrete advantages of the generative formulation: notably better cross-domain generalization on unseen data, more complete multi-lesion segmentation, instruction-conditioned task control, and sampling flexibility. We also analyze the strengths and limitations of the paradigm, including its color sensitivity and unsupported attribute-conditioned selection. Code is available at: https://github.com/wincharm001/InstEditSeg.

Guideline-consistent semantic segmentation requires more than category recognition, as real-world labeling policies demand fine-grained, task-specific decisions. Recent multi-agent refinement systems improve compliance with such textual guidelines by detecting and correcting errors. However, they are stateless: feedback from the critiquing agent is discarded, causing the same guideline-specific mistakes to be repeatedly rediscovered and corrected across the dataset at the cost of additional refinement. We introduce InsightSeg, an episodic memory mechanism that converts successful correction episodes into reusable, visually grounded insights. A meta-analyzer distills each qualifying episode into directive natural-language insights and anchors them to the local image regions that caused the error using patch-level visual concept vectors. On subsequent images, these concepts are matched against dense patch embeddings to retrieve relevant insights, which condition the segmenting agent before making its first prediction. This shifts the system from correcting recurring errors to preventing them, improving segmentation quality before any refinement occurs. Across Waymo and Cityscapes, InsightSeg improves both first-pass and final guideline-consistent segmentation performance while requiring fewer refinement steps, demonstrating that multi-agent refinement can become more accurate and efficient by drawing on past correction experience.

Agent evaluations face two distinct evidentiary questions: whether a reported claim is recomputable from retained evidence (sufficiency), and whether the retained records cover the committed experiment set (coverage). Generic logs and hash-linked transcripts answer neither reliably. We introduce ClaimReceipt, a claim-relative receipt specification and selective verifier that binds typed transaction evidence to a signed experiment manifest and returns PASS, INVALID, or INCONCLUSIVE per claim. We freeze the specification before implementation (SHA-256 18d109...b81). On 1,392 historical buyer–seller records, a CR-2 verifier reproduces all five manually labeled audit verdicts, exactly replays 600 deterministic and 792 post-generation records, makes every one of 13 declared field groups non-redundant under tested ablations, and returns the expected result on 11/11 semantic faults with 0/8 false positives. We then run a separate prospective CR-3 epoch: 30 assignments are committed before inference, terminal receipts are signed and chained, and private evidence is encrypted for an auditor. Complete evidence yields coverage and accounting PASS; withholding one terminal receipt returns INCONCLUSIVE_COVERAGE, while withholding all private openings preserves coverage and protocol verification but makes economic claims inconclusive, exactly matching a preregistered prediction. Receipt instrumentation adds 0.021% of model-inference time and 9.9 KB per transaction. A specification-legibility probe indicates that our own frozen specification is not yet unambiguous to an independent reader. Claim verification therefore requires both claim-sufficient evidence and a committed universe against which omissions become visible.

As LLM coding agents increasingly perform end-to-end engineering work, we lack empirical characterization of how they behave on systems-level requirements: schema design, async orchestration, configuration correctness, and retrieval-filtering trade-offs. We present a case study of one such agent implementing a multi-component data system against a detailed pre-existing specification. Storage technologies, schema, entity-resolution algorithm, and retrieval-filtering strategy were fixed in advance; the agent autonomy was in the implementation, in diagnosing and fixing defects it introduced, and in interaction-design choices left open. Over a single session, we catalog five such defects, categorized by constraint violated and detection method. We further evaluate, on the public HotpotQA benchmark, the one retrieval trade-off specified in that architecture: restricting candidates to a graph-identified entity set before ranking versus unfiltered search. We substitute the benchmark gold evidence labels for entity identification, since we lacked LLM access to run that stage, and report standard recall rather than the benchmark own accuracy metrics. Across retrieval budgets from 1 to 10 and 100 questions against a pooled corpus of 2994 paragraphs, filtered recall reaches its ceiling by a budget of 3, expected once candidates are restricted to the gold paragraphs themselves, while unfiltered search recovers all required evidence only 69 percent of the time even at a budget of 10, a gap that holds at every budget tested, with sign test p less than 0.0001. We close with a discussion of where the agent autonomy succeeded versus required correction, including one instance where a claimed performance fix was never re-measured on the regression that motivated it.

Large language models (LLMs) are increasingly used as assistants for statistical and data science work, yet existing evaluations largely assume the analysis target is already specified. In practice, users arrive with informal goals and heterogeneous data, leaving the model to decide what statistical task is implied and which data are relevant. We first formalize this upstream step as Statistical Problem Formulation and decompose it into two subtasks: (1) Statistical Problem Classification and (2) Variable Identification & Role Assignment. We then introduce StatFormBench, a benchmark built from five cross-domain statistics textbooks and a data science case library, covering diverse problem types, data representations, and scenario styles. It contains 1,013 samples spanning 20 coarse-grained and 85 fine-grained statistical problem categories. Across 14 open- and closed-source LLMs, the best zero-shot models reach only 72.0 fine-grained classification accuracy and 63.2 variable set overlap. No model performs consistently best across the two subtasks, while enhanced prompting strategies yield only limited or inconsistent gains. We release the benchmark data on Hugging Face at https://huggingface.co/datasets/THU-CongLab/StatFormBench and the evaluation code on GitHub at https://github.com/THU-CongLab/StatFormBench.

Large language models (LLMs) are increasingly embedded in organizational work, yet their errors often pass human review. Prior research locates such failures in users' capability to review LLM output or their engagement in doing so. We develop an alternative, retrieval-based account of human oversight and posit that error detection is more effective when oversight-relevant information is accessible to users at the moment of review. Across two randomized lab-in-the-field experiments with 640 customer-facing employees, we show that self-generated explanations improve error detection and strengthen recall of verification-relevant reasoning, while cues that reactivate such reasoning help sustain detection under repeated LLM use. Theoretically, we identify information retrievability as a distinct precondition for effective oversight and specify generative encoding and cue-supported reactivation as mechanisms that build and sustain it. Practically, lightweight onboarding self-explanations and daily retrieval cues can make human oversight more resilient as LLM use becomes routine.

Ultra-low-bit language models can reduce storage and memory bandwidth, but a nominal "1.58-bit" label does not fully describe the stored representation, retained capability, or runtime behavior. We study an end-to-end post-training conversion of Qwen, an instruction-tuned 4B-parameter model, using KOTMS rotation, E2M-ATQ ternarization, and GPTQ-style error compensation from TWLA. The experiment is weight-only: activations remain at 16-bit precision, so ILA-AMP is omitted. We evaluate effective bit accounting, task capability retention, perplexity, calibration sensitivity, checkpoint composition, and deployment behavior. The final conversion uses 1.641 effective bits per weight for quantized linear weights, with 81.62% of model parameters targeted. Across ten scored capability comparisons, accuracy falls from 64.5% to 54.7%. Degradation is uneven: BoolQ retains 84.6% chance-corrected teacher performance, while ARC-Challenge retains 43.8%. Perplexity rises from 13.639 to 18.748 on WikiText-2, 24.700 to 31.992 on PTB, and 19.831 to 28.966 on C4. A subsequent packing run preserves the ternary planes and scales, reducing reported model size from 8.29 GiB to 3.96 GiB with essentially unchanged perplexity. A separate third-party packing attempt was lossy and is excluded from the primary artifact claim. The packed artifact has not been benchmarked end-to-end for task accuracy or generation throughput. A preliminary Triton GEMV microbenchmark is 4.6x slower than FP16 cuBLAS on one tested shape. We therefore do not claim that compression alone yields faster inference.

Agent harnesses increasingly want to run small language models (SLMs) on the microtasks around a frontier large language model (LLM) planner: auto-approving shell commands, writing memory, selecting tools, ranking past turns. We ask whether off-the-shelf SLMs meet practitioner-defined thresholds and, when they fail, why, and whether quantization changes the answer. We build a benchmark of 4 such microtasks with fixed prompts and automatic metrics, each with a pre-specified threshold $\tau$ anchored to a cheap non-LLM baseline and a CI-aware eligibility rule (a configuration passes only if its confidence bound clears $\tau$). Sweeping Qwen3 0.6/1.7/4/8B at their best (FP16, greedy, one frozen prompt, no tuning), we find an eligibility gap: 0 of 16 (4 tasks $\times$ 4 models) configurations pass (verified by checking the raw outputs and parser behavior). A logprob decision-threshold diagnostic (T1/T3/T4; T2 via a context-length/cascade probe) separates the failures into capability deficits and failures that can be addressed by changing the decoding threshold (4 regimes). Quantization to 4-bit (RTN/GPTQ/AWQ) does damage that depends on model size and moves no configuration into eligibility (certified on the reconstructable hard-label tasks T1/T3, diagnostic/windowed robustness on T2/T4), so the gap tracks model size more than precision; it replicates on Llama-3.x (12/12 ineligible) and is robust to the anchor choice (a $\tau$-sweep) and to prompt wording (0/112 eligible across the original plus 3 neutral paraphrases per cell). The practical implication: place SLMs behind a baseline that meets the CI-backed threshold, and use the SLM only where the baseline fails to meet the threshold; e.g. a 4B re-ranker over a BM25 shortlist beats BM25 ($+0.047$ [0.020, 0.073], without itself certifying eligibility).

Adapting large pretrained vision models under limited data and frozen-backbone constraints remains a central challenge in transfer learning. While lightweight adapters and parameter-efficient fine-tuning methods are widely adopted, most rely on generic multilayer perceptrons or low-rank linear updates, offering limited control over the spectral and geometric structure of feature transformations. We investigate whether structured nonlinear feature lifting can improve representational alignment in frozen regimes. We introduce Quantum-Inspired Nonlinear Adapters (QINA), compact modules that perform learnable trigonometric feature lifting followed by bounded nonlinear aggregation. The design induces structured oscillatory basis functions with an explicit norm-dependent Lipschitz bound, enabling spectral reshaping of pretrained representations without increasing the receptive field or significantly expanding parameter count. Importantly, the method operates entirely within standard deep learning frameworks and does not require quantum hardware. Through systematic experiments across natural and medical imaging datasets, classification and segmentation tasks, multiple adapters and placements, and varying training budgets, we show that performance in frozen regimes is primarily representation-limited. Nonlinear lifting improves adaptation, and the proposed structured trigonometric formulation consistently outperforms identity baselines, fixed Fourier feature mappings, and parameter-matched baseline adapters. Within the evaluated frozen-backbone settings, structured spectral parameterization provides a more effective inductive bias than generic nonlinear adapters. This work highlights the importance of geometry- and spectrum-aware adaptation mechanisms for large pretrained vision models.

BraTS local synthesis replaces masked regions in T1-weighted brain MRI with plausible tumor-free tissue while preserving observed anatomy. We present CATCH, conditional 3D diffusion in an invertible Haar-wavelet domain. Its denoiser receives noisy target coefficients, voided-image coefficients, and a signed mask; tumor-excluded wavelet reconstruction and a hole-focused loss guide training, and hard compositing preserves observed voxels. We compare fixed masks, tumor-component augmentation, and a weighted mixture of tumor-derived, irregular-blob, and ellipsoidal masks. Of 25 development cases, five prespecified cases select each arm's checkpoint and all 25 of their trajectory aggregations; a separate 75-case internal set compares the frozen pipelines and selects a weighted mixture for organizer evaluation. Five-trajectory averaging yielded internal SSIM/PSNR/MSE (mean$\pm$SD) of $0.80\pm0.13$, $19.18\pm1.80$dB, and $0.010\pm0.005$. As the sole officially evaluated pipeline, weighted mixture yielded $0.772\pm0.119$, $20.89\pm3.27$dB, and $0.0098\pm0.0054$ on the 219-case BraTS 2026 validation set. Against compute-matched random augmentation internally, it improved SSIM by 0.019 (95% bootstrap CI: 0.013-0.025), PSNR by 0.95dB, and MSE by 0.003; all three paired comparisons remained significant after Holm correction. Results favor the complete weighted-mixture policy within CATCH; absent official fixed- and random-pipeline scores and a directly comparable external baseline limit broader conclusions.

Sofroniew et al. (2026) report that emotion concepts in Claude Sonnet 4.5 are represented as vectors whose geometry mirrors human affect psychology. We replicate the representational core of that study on the base pretrained model google/gemma-2-27b, inheriting every disclosed parameter, resolving unspecified steps by disclosed rules, and changing only the subject model. From 205,200 newly generated Claude Sonnet 4.5 stories matching the original corpus design, we extract 171 emotion vectors and recover the core results. The leading principal components form an affective circumplex (PC1 carries 26.7% of variance against the original study's ~27%, PC2 13.4% against ~14%), emotions cluster into similar intuitive families, and the geometry holds across a broad late-middle band. The valence axis aligns with human norms (r = 0.72, against 0.81) and is stable across scales and depth. Arousal aligns at r = 0.67 (against 0.66) but only at the full 171-emotion scale and late depth, so we do not classify it as replicated. Extending the original analysis, a 46-layer sweep locates a sharp seam at L22-26, where the geometry consolidates and the vocabulary readout becomes legible. An embedding-layer baseline finds much of the geometry already present in the static token embeddings, with arousal as the exception. On top-activating held-out text, the geometry predicts token-level co-activation at r = 0.907. At least 52% of vectors peak on structurally non-conceptual tokens, a measured floor for max-activation confounds. Even when a document contains the vector's emotion word, the peak lands on that word only 6.1% of the time. Because the subject is a base model and the stimuli are Claude-generated fiction, the recovered structure is a property of the pretrained representation of Claude-rendered emotion. The original's causal and assistant-facing analyses are out of scope. Code and data are released.

High-rate neural telemetry requires an implant-towearable link whose circuit partition, waveform mapping, packet processing, and failure semantics are mutually consistent. This paper specifies a chip-oriented, one-way impulse-radio ultrawideband (IR-UWB) uplink comprising an implanted packet engine and free-running transmitter, a tissue-proxy channel, and a wearable mixed-signal receiver and digital baseband.

This exploratory pilot study evaluates the scope and perceived accuracy of personal information output from ongoing conversational interactions in generative AI systems using GPT-5.2 Instant and GPT-5.2 Thinking, categorized into three output types: Fact, Inference, and Confidence. Based on the evaluation results obtained from 15 Japanese participants, differences in model design have limited impact on personal information output tendencies. Compared with the Inference type, the Fact type shows a more conservative output pattern. Regarding attribute categories, the findings indicate that Core Personal attributes associated with identification are treated relatively conservatively, whereas Behavioral and Linguistic attributes show higher accuracy across both Fact and Inference outputs. Furthermore, Holistic Profile, Psychological and Cognitive, and Residual attributes are more readily inferred, even when not supported by explicit factual outputs. Notably, the lack of null outputs for these attributes in the Inference type suggests that such inferred profiles may be constructed from indirectly available contextual information. The findings may contribute to future discussions regarding privacy awareness and personal information inference in generative AI systems.

High classification accuracy alone is insufficient for clinical image analysis, where calibrated confidence and reliable uncertainty estimates are essential. This study proposes a reliability-aware Hybrid-K ensemble selection framework for multiclass cervical cytology classification using the SIPaKMeD dataset. Nine deep learning architectures were evaluated using a fixed stratified five-fold partition and three training seeds. After post-hoc temperature scaling, models were assessed using macro-F1, accuracy, AUROC, expected calibration error (ECE), worst-class ECE (WC-ECE), area under the risk-coverage curve (AURC), Brier score, and negative log-likelihood (NLL). Models were ranked using an equal-weight composite score, and Hybrid-K ensembles were formed from the top-ranked models using soft voting. Robustness was examined using 5,000 Dirichlet-sampled metric-weight vectors, leave-one-metric-out analysis, and corrected paired testing across 15 fold-by-seed evaluations. The final Hybrid-2 ensemble, comprising Swin-Tiny and TinyViT-5M, reduced AURC by 43%, NLL by 17%, and WC-ECE by 36% relative to the best individual model. It was selected in 96.8% of random weighting scenarios, remained unchanged across all leave-one-metric-out analyses, and improved the full composite score. However, per-metric gains were not statistically significant after Holm-Bonferroni correction (all adjusted p >= 0.168). Because post-hoc calibration did not use a fully independent calibration set, calibration-dependent results should be interpreted as exploratory internal estimates. Overall, the framework identified a compact ensemble robust to alternative metric weightings and improved reliability point estimates under internal validation on a single dataset.

Large language model (LLM)-based multi-agent systems (MAS) predominantly rely on centralized orchestration and lack formal verification mechanisms for agent reliability, participation, and system-level behavioral alignment. These shortcomings leave open environments severely vulnerable to uncooperative or malicious agents. This work proposes DART, a Directed Acyclic Graph (DAG)-based reputation and incentive regulation framework for trustworthy multi-agent collaboration, combining centralized operational orchestration with blockchain-enabled decentralized governance and accountability. DART unifies DAG workflow orchestration, capability and reputation-aware task allocation, dynamic behavior updates, multi-factor incentives, and smart contract accountability paired with IPFS storage. Under this paradigm, agent selection dynamically balances task alignment, historical reputation, and workload, while post-execution behavioral evidence continuously calibrates agent trust and the probability of future participation. Evaluated across four axes, DART achieves 93.6% Pass@1 on GSM8K and builds a full-stack application in 142 s using two agents, outperforming centralized baselines. Across five independent 150-round longitudinal trials, Full DART achieves a mean task success rate of 93.33 +/- 2.26%, output quality of 0.9357 +/- 0.0117, retry rate of 0.2307 +/- 0.0816, and allocation delay of 1.1153 +/- 0.0408 s, consistently outperforming its ablated configurations DART isolates persistent and intermittent malicious agents, obtaining a 99.3% output containment rate and restoring system success to 99.8%. These results demonstrate the potential of coupling reputation, incentives, DAG-based coordination, and verifiable blockchain-enabled governance to support adaptive and accountable multi-agent collaboration.