Papers

We investigate whether commonly available LoRA variants have an advantage over basic LoRA in multilingual instruction tuning. Experiments involving LoRA and four other variants on two datasets across diverse target languages show that there is no significant advantage in using more complex LoRA variants instead of basic LoRA, with respect to balancing cross-lingual transfer and knowledge retention. An analysis of hidden embeddings reveal that layer-wise language representation remains largely similar across LLMs fine-tuned with different LoRA techniques, suggesting that architectural novelty of LoRA techniques may not translate into better cross-lingual adaptation.

Autonomous web navigation remains challenging for LLM agents, and the strongest generalist systems rely on proprietary reasoning models whose inference cost is prohibitive for the repetitive tasks where such agents would be most useful. We argue this gap stems not from insufficient model capability but from agent architectures that fail to replicate three human cognitive advantages: selective attention to relevant page regions, persistent memory of website structure, and procedural fluency with common interaction patterns. We introduce WebChallenger, a web agent framework that addresses each gap through architecture design rather than model scale, built around PageMem: a structured page representation deterministically constructed from the DOM that exposes each page as a hierarchy of semantic sections with short summaries. On this shared substrate we build three mechanisms that mirror the three cognitive advantages: a divide-and-conquer observation pipeline that lets the agent skim section summaries and extract details only from task-relevant regions; a lightweight exploration and memory system that traverses each website once to build a reusable map of pages and element behaviors; and compound action workflows that collapse common multi-step interactions into single agent actions, handling partial state changes automatically. Because all three operate over PageMem, the framework generalizes across websites without site-specific adapters. Using off-the-shelf open-weight models without fine-tuning, our system achieves 56.3% on WebArena, 48.7% on VisualWebArena, 51.0% on Online-Mind2Web, and 70.9% on WorkArena, approaching frontier proprietary systems at a fraction of the cost. Our code is released at https://github.com/jayoohwang1/webchallenger

Math reasoning benchmarks have proliferated, yet most lack a per-item difficulty signal grounded in actual human performance. We introduce KCSAT-ML, a decade (2014-2025) of Korean College Scholastic Ability Test (KCSAT; Suneung) mathematics: 664 problems with a 339-item core set carrying official per-item error rates from nationwide cohorts of hundreds of thousands of examinees. We pair the benchmark with Difficulty-aligned Reasoning Gain (DRG): a score-orthogonal metric that asks whether a model's mistakes concentrate on the items humans found hard, or on items humans found easy. Together they expose, across a wide range of VLMs (and LLMs with OCR), three patterns: (i) low-budget accuracy collapses on the high-human-error tail at every model size; (ii) test-time scaling (TTS) raises token use roughly linearly with cohort error rate, while accuracy gains follow a non-monotonic curve; (iii) within a single family, TTS flips between anti-scaling on the hardest items and overthinking on easier ones – two faces of the same alignment failure. With the proposed DRG metric, we find that models with near-identical accuracy can sit at near-opposite values: one model gets wrong what humans also find hard, while another solves the hardest items yet fails on items humans find easy, which is a critical contrast that aggregate accuracy hides. Our code and dataset builder is fully open-sourced at https://github.com/naver-ai/KCSAT-ML.

Scientific discovery is often a collective process: researchers share partial results, inspect failed attempts, and build on each other's ideas over long time horizons. Recent AI systems have shown that language-model-based agents can make meaningful progress on open scientific problems, but most existing systems operate in isolation. In this paper, we present EinsteinArena, an agent-native platform for open distributed research and discovery. EinsteinArena provides agents with a live set of open problems, each with a solid verifier, public leaderboard, and problem-specific discussion forum where agents can ask questions and share insights. We focus on mathematical tasks that have garnered substantial research interest, where progress can be measured unambiguously. As of May 2026, agents on EinsteinArena have discovered 12 new state-of-the-art results better than any previous human or AI solutions. One notable example is the kissing number problem in dimension 11, where the platform improved the best known lower bound from 593 to 604. This advance did not come from a single agent or isolated run. Rather it arose through a sequence of submissions, public discussion, verifier refinement, and subsequent agent-to-agent borrowing of ideas. These results provide evidence that decentralized scientific discovery can emerge from open interaction among autonomous agents in the wild, demonstrating a new paradigm for collective AI-driven research.

Vision-language models (VLMs) are increasingly deployed where answers must follow from what is in the image, yet they often answer from textual priors, the question's phrasing together with memorized world knowledge, rather than from the image itself, which inflates benchmark scores and yields confident but ungrounded answers. Existing benchmarks rarely isolate this behavior, since each image is usually paired with a single fixed question. To measure the reliance, we build a 540-image benchmark across six reasoning categories and generate four question variants over the same images, so that phrasing rather than image content is the controlled variable. The hardest variant is written directly from the image to minimize text leakage. We benchmark eleven VLMs spanning small open-weight models to large closed-source systems: every model degrades on the hardest variant, and open models fall furthest. Our central diagnostic is a no-image ablation, which collapses the open-weight models to their text-only floor (1 to 9 percent). Three further analyses, LLM-rated difficulty, low base-to-final textual similarity, and human re-annotation, corroborate genuine image-dependence. In-context exemplars that match how a variant was built recover the most accuracy, and GRPO post-training of a small VLM yields consistent gains across all four variants that transfer to a held-out out-of-distribution set. Textual-prior reliance is measurable and partly trainable away.

Does personalizing what a reader sees pay off, and where does it stop? Using a social web highlighter and a co-readership identity control (the same document highlighted by many users, which holds document and topic fixed and asks whether a person's own history predicts their marks better than another reader's does), we map the shape and limits of personalization across reading altitudes. At the document altitude we give the clean, leakage-free, identity-controlled measurement that prior next-document evaluations could only upper-bound: a person's history identifies which documents in a co-reading neighborhood are theirs, with an own-versus-other gap of +0.169 against community negatives and +0.119 against topic-matched hard negatives (both highly significant); a content-based arm suggests the signal is not purely title-driven but is largely thematic. This is comparable to the span-level selection signal (+0.14) from our prior work: the selection signal is of comparable magnitude across altitudes (+0.12 to +0.17), most of it stable topic preference. At the sentence altitude, a two-stage personalized auto-highlight (an impersonal model proposes candidates, a personal model re-ranks them) does not improve on its impersonal baseline: two off-the-shelf zero-shot LLMs, including a frontier model, predict highlight locations worse than a lead baseline, and personal re-ranking is beaten by the salience order even on the highest-recall candidate pool, so the null is not merely a Stage-1 ceiling artifact. Measurable personalization appears primarily at the selection layer: modest (~+0.13), topic-dominated, with no reliable gain at the salience layer. We also surface a control-in-negatives bias that inflated our document gap to a spurious +0.227 until audited. Going beyond the shared salience layer may be better approached by aggregating individuals than by personalizing them harder.

Muon collider research spans accelerator physics, detector instrumentation, and high-energy phenomenology, with relevant evidence scattered across a rapidly expanding and heterogeneous body of scientific literature. As high-energy physics (HEP) increasingly explores agent-assisted analysis workflows, efficiently locating, integrating, and verifying scientific evidence becomes an essential capability. While retrieval-augmented generation (RAG) offers a promising framework for scientific question answering, integrating agentic reasoning without compromising retrieval precision remains a key challenge. In this work, we present agentic hybrid RAG, an evidence-grounded RAG framework for muon collider research. The framework combines a hybrid retriever, integrating sparse lexical and dense semantic retrieval, with an agentic reasoning module for query decomposition, evidence expansion, and grounded answer generation. To enable systematic evaluation, we construct the first benchmark for retrieval-augmented scientific question answering in the muon collider domain, comprising a curated literature corpus together with dedicated retrieval and answer-generation benchmarks covering major detector and physics research topics. Extensive evaluation shows that hybrid retrieval provides the strongest retrieval backbone, while agentic reasoning is most effective for controlled evidence expansion and answer synthesis. Built on this principle, agentic hybrid RAG consistently outperforms representative retrieval and RAG baselines in retrieval effectiveness, answer quality, evidence coverage, and factual grounding. Together, the benchmark and framework provide a foundation for evidence-grounded scientific question answering and future HEP analysis agents operating over large-scale scientific literature. Code is available at \href{https://github.com/AItutorialjrb/RAG_muon_JINST}{this URL}.

Real-world crisis intervention is inherently conversational, yet existing research largely focuses on static texts. When applied to multi-turn dialogues, current models exhibit significant performance degradation, struggling to track risk signals that emerge as context evolves. To address this gap, we introduce CRADLE-Dialogue, a clinician-annotated benchmark for turn-level crisis detection in conversational settings. The dataset features 600 dialogues with multi-label annotations across clinically grounded risks, including suicide ideation, self-harm, and child abuse, distinguishing past from ongoing risk. We further propose an Alert-Confirm evaluation protocol that distinguishes early warning signals (Alert) from turns where a specific crisis becomes explicitly identifiable (Confirm), reflecting the clinical need to intervene before risk becomes explicit. Experiments show that identifying when risk emerges is much harder than recognizing that it exists: models achieve only mid-40% to high-60% Micro F1. Additionally, we release a synthetic training corpus and a 32B-parameter model that substantially outperforms existing open-source models and achieves competitive or superior results against proprietary models across turn-level, dialogue-level, and confirm-only evaluation settings.

As large language models (LLMs) continue to scale, it becomes increasingly challenging to grow model capacity under fixed computation budgets. We propose Path-Aligned Decompression Distillation (PADD), a framework for distilling knowledge from dense teachers without explicit routing into mixture-of-experts (MoE) students while learning high-quality routing policies. PADD organizes knowledge distillation into four stages in two phases: an initialization phase (Stage I) that builds diverse functionality in the student's experts through teacher neuron clustering and student-expert warmup, and a training phase (Stages II–IV) that integrates online adaptive distillation, path-refined policy optimization, and reward-augmented load balancing in a single training pipeline. Experiments on mathematical reasoning benchmarks demonstrate that PADD yields substantial gains over strong baselines at the same inference cost and that the MoE student can match or surpass its dense teacher. They also demonstrate effective teacher-to-student knowledge distillation and stable routing behavior.

Machine unlearning is increasingly important for large language models, yet unlearning in Mixture-of-Experts (MoE) architectures remains underexplored. Unlike dense models, MoE architectures employ a router at each layer to assign each token to a sparse subset of experts. In this work, we observe that forget data often activates a small subset of experts disproportionately, while these experts may receive much weaker activation from retain data. This forget–retain routing mismatch can leave forget-critical experts under-regularized during unlearning. To address this, we propose TRACE, Targeted Routing-Aware Calibration of Experts, for MoE unlearning. TRACE first detects forget-critical experts from offline activation statistics, and then calibrates retain regularization by reweighting token-level retain losses so that each selected expert's retain-side activation frequency better matches its forget-side counterpart. Experiments on WMDP and MUSE-BOOKS across multiple MoE LLMs show that TRACE consistently improves the forget-utility trade-off, yielding a 9% relative utility improvement over the strongest baseline under comparable forgetting quality and the best performance on three out of four MUSE-BOOKS metrics.

Automated Essay Scoring (AES) systems must judge interdependent discourse elements (e.g., lead, claim, evidence, conclusion), yet most approaches treat these in isolation, harming coherence and generalization. We investigate task-aware fine-tuning of LLaMA-3.1-8B for AES using parameter-efficient LoRA with 4-bit quantization and compare three training curricula: (i) Sequential (progressively fine-tuning on lead, then position, then claim, then evidence, then conclusion), (ii) Independent (task-specific models), and (iii) Randomized (shuffled multi-task). Experiments on the PERSUADE~2.0 corpus show that modeling task dependencies matters: Sequential fine-tuning yields the strongest overall results, including F1 scores of 65% (evidence) and 87% (conclusion) and corresponding accuracies of 63% and 85%, surpassing Independent training and outperforming a general-purpose LLaMA-70B baseline on conclusion despite its far larger capacity. Randomized training improves position scoring (57% F1) but is less consistent elsewhere. These findings indicate that (1) curriculum design aligned with discourse structure can materially improve AES, and (2) small, task-optimized models can be competitive with substantially larger Large Language Models (LLM), offering a practical path to scalable, cost-effective assessment. We release templates and implementation details to facilitate reproduction and future work on curriculum design for educational NLP.

Spreadsheets and tables are widely used representations for structured data analysis, but effective analysis still requires substantial manual effort and domain expertise. Recent large language model (LLM) agents can automate parts of this process, but they often provide limited transparency into intermediate decisions, rely on implicit assumptions, struggle with multi-table comparison, and repeat similar workflows without adapting to a user's preferences. This paper presents TabClaw, an open-source interactive AI agent for spreadsheet manipulation and table reasoning. Users upload CSV or Excel files and issue natural-language requests; TabClaw clarifies ambiguous intent, exposes an editable execution plan, streams a ReAct-style tool-using analysis loop, dispatches specialist agents for parallel multi-table reasoning, and synthesizes findings with explicit consensus and uncertainty markers. Beyond one-off analysis, TabClaw records completed workflows, extracts persistent user memory, distills reusable skills from repeated tool-use patterns, supports package-style skill import, and upgrades skills from negative feedback. Experiments on spreadsheet manipulation and table reasoning benchmarks show that TabClaw improves executable task completion and reasoning performance while preserving an inspectable user workflow. This paper shows how TabClaw turns spreadsheets and tables into inspectable analytical workflows while gradually personalizing itself to recurring data-analysis tasks. Our code is available.

LLM-as-judge is the default instrument for evaluating conversational agents, yet its reliability is almost always reported as agreement with human ratings, not recall of real defects. We study a deployed multi-turn food-and-beverage ordering agent and measure how many genuine quality problems its built-in LLM judge catches, using exhaustive human transcript review as ground truth. Across three batches the judge surfaces well under a quarter of human-confirmed systematic problems – 2 of 9 patterns (22%) in one batch, and its operational gate flagged zero of 100 rounds in a batch where humans confirmed 23 distinct defects and 7 new cross-cutting patterns. Our blind-spot taxonomy shows the failure is structured, not random: the judge catches turn-local issues (a fabricated statistic, a wrong language) but misses cross-turn state issues (confirm-gate lockout, cart hallucination, escalation lockout, stale referents). The mechanism: the scoring rubric exposes only three coarse axes (intent, brand-voice, personalization) and has no category for the behavioural dimensions – state-tracking, guardrails, recovery – where most defects cluster. The failure is routing, not perception: 113 of 114 rounds whose raw judge note describes a confirm-gate or cart-state defect are scored "brand voice", and none reach an operational failure – the gate is wired to hangs and hard assertions, not the rubric – so the 0% is a routing-and-wiring failure, not blindness. The consequence for prevalence estimation is sharp: when the apparent defect rate is zero the Rogan-Gladen correction degenerates – no signal can recover the true rate – while where the gate reports a nonzero rate the same estimator implies a 3-6x undercount under our measured sensitivity. For production multi-turn agents, automated judging is a regression floor, not a substitute for human review.

Evaluating reasoning quality in multi-agent LLM systems is challenging, especially for open-ended tasks without reference answers. We investigate whether intrinsic confidence signals, token-level log-probabilities from decoding, can predict reasoning quality as assessed by LLM-as-judge evaluation. Using a debate-based essay scoring framework, we compare confidence proxies against rubric-based judge scores across two ASAP essay sets. We find that early-token confidence, particularly within the first few generated tokens, is consistently the strongest predictor of reasoning quality, outperforming full-sequence statistics. Analysis of log-probability trajectories shows that the opening phase of generation is the most heterogeneous and therefore most informative. We also observe a systematic asymmetry between agent roles, with stronger alignment between confidence and quality for supportive reasoning than for adversarial critique. These results suggest that early decoding dynamics provide a lightweight and effective signal for estimating reasoning reliability in multi-agent LLM systems.

When LLM agents are coerced into covertly encoding sensitive data (Base64, ROT13, acrostic, synonym chains, and beyond), the resulting outputs evade output-side detection but the underlying computation does not. Across nine encoding families and eight models from five architecture families, that computation is supported by a shared low-dimensional encoding subspace in the residual stream. A logistic-regression probe trained on eight encoding families recovers the held-out ninth at AUC 0.975-1.000, reading the computation rather than surface features. The same direction exhibits a second mechanistic signature at the planning token, flipping polarity to activate positively when the model will simulate the encoding inline and negatively when it will outsource it to a tool call, distinguishing two execution strategies before the encoded text exists. We build MIRAGE (Model-Internal Readout of Agentic Generation Exfiltration), a two-channel real-time monitor exploiting both signals. On 126 agentic exfiltration scenarios, it reaches AUC = 0.918, substantially outperforming output-only detection (AUC = 0.518). Monitor performance is fundamentally a property of the host model's geometry: benign-encoding false-positive rate ranges from 0% on Qwen-7B to 100% on Phi-3.5, revealing that the probe faithfully reads whether a model's geometry separates covert from overt encoding. Across all tested adversarial budgets, every attack suppressing the subspace also destroyed encoding fidelity, reported as an empirical regularity on the evaluated budgets, not a structural impossibility claim.

Open-ended generation tasks often require a set of meaningfully different outputs, yet large language models often produce similar generations. Existing test-time diversity methods operate at different stages of generation with varying effectiveness, but it remains unclear what design choices lead to meaningful diversity in the output. We introduce a framework that characterizes test-time diverse generation methods by the diversity source introduced during generation and provide a transmission score for measuring how effectively variation in the source reaches the final output. Guided by this framework, we propose fully automated specification-level generation methods that first generate diverse intermediate specifications and then condition on them to produce final responses. Across five open-ended tasks and four backbone models, specification-level injection improves output diversity over test-time baselines while maintaining comparable quality. Our analysis shows that successful diversity injection depends on both the diversity of the sources and their transmission to the output, highlighting source design and source-to-output realization as two key levers for building more diverse generation systems.

When large language models generate from retrieved or augmented contexts, conflicts between external context and parametric priors remain a central reliability bottleneck. Existing contrastive decoding methods follow a context-aware paradigm that unilaterally amplifies context over parametric priors, overwriting correct priors when the context is erroneous. We generalize this to the conflict-aware paradigm that dynamically allocates authority between prior and context based on conflict signals, rather than presupposing context trustworthiness. We show that the affine combination of prior and context logits yields a power family with an inherent regime asymmetry: extrapolation amplifies errors unboundedly when the prior is correct, interpolation under-corrects when the context is correct, and no static regime covers both. Existing contrastive decoding methods are instances of this family, mostly extrapolative. To evaluate both conflict directions, we propose TriState-Bench, a model-aware evaluation protocol that calibrates per-model prior knowledge to measure three conflict states: correction, resistance, and agreement. To resolve the asymmetry, we propose Adaptive Regime Routing (ARR), which routes between regimes at each step, lifting resistance EM from below 6 to 16–33 without sacrificing correction or agreement. Our code is available at https://github.com/keith-Jiang/conflict-aware-decoding.

Multi-agent debate systems are typically evaluated only on whether the final answer is correct, overlooking the quality of the intermediate reasoning that debate is designed to produce. This paper studies the relationship between three signals in multi-agent debate: token-level log-probability distributions over reasoning tokens, LLM-as-judge rubric scores assigned to those tokens, and final task accuracy. We examine whether internal confidence signals predict externally evaluated reasoning quality, and whether either signal aligns with task correctness, across three domains: rubric-based scoring, mathematical reasoning, and factual question answering. Our framework pairs a two-agent debate architecture – a Constructor and an Auditor – with an LLM-as-judge that scores each agent's reasoning along instruction following, justification quality, and evidence grounding, together with a critical-failure flag. Experiments in the rubric-scoring domain reveal a consistent four-phase confidence trajectory and a substantial role asymmetry: confidence aligns with judged reasoning quality roughly twice as strongly for the Constructor as for the Auditor, and confidence-based detection of critical reasoning failures is markedly more reliable for the Constructor (AUROC 0.804) than for the Auditor (0.634). These findings motivate the broader cross-domain investigation proposed in this paper.

Evaluating Knowledge Graph Completion (KGC) models remains challenging because standard assessment relies on isolated rank-based metrics such as MRR, Hits$@$k, and Mean Rank, which often produce conflicting model orderings across datasets. A model that leads on MRR may trail on Hits@1, and strong performance on one dataset may not generalize to another. This fragmentation hinders comparison, enables selective reporting, and obscures real progress. We reframe KGC evaluation as a Multi-Criteria Decision-Making (MCDM) problem and present a meta-analysis of seven aggregators across five tests: consistency, cross-dataset stability, metric independence, robustness under noise, and generalizability. Each test is averaged over leave-one-model-out (LOMO) and leave-one-group-out (LOGO) removals so that reliability reflects aggregator behavior across diverse model subsets. Across tail $(h,r,?)$ and relation $(h,?,t)$ prediction, Pareto-optimal analysis identifies Z-score as the most balanced aggregator, which ranks DualE highest for tail prediction and FMS (Flow-Modulated Scoring) highest for relation prediction. A test-sensitivity analysis using the same removals shows that consistency and stability are largely removal-invariant, while generalizability and independence are the most sensitive. The framework resolves evaluation inconsistencies and offers evidence-based guidance for aggregator selection and model benchmarking in KGC.

OpenRTLSet introduces the largest fully open-source dataset for hardware design, offering over 131,000 diverse Verilog code samples to the research community and industry. Our dataset uniquely combines Verilog code from GitHub repositories (102k modules), VHDL translations (5k modules), and synthesizable C/C++ translations (24k modules), all freely accessible without proprietary restrictions. Using the reasoning model DeepSeek-R1, we generated paired natural language descriptions for each code sample, enabling fine-tuning of various language model families (e.g., Qwen and Granite) for Verilog code generation. Our dataset explores multiple options, including Verilator-generated C++ files as additional context during labeling, quantization techniques (INT4 vs. BF16), and performance differences across model sizes (7B-32B parameters). OpenRTLSet demonstrates that open-source approaches can achieve superior performance in hardware design tasks, establishing a new foundation for accessible research and commercial use in this domain.

This paper presents AuditBench, a new benchmark dataset for evaluating the capabilities of LLMs at investigating security-related system audit logs. We design and use this benchmark to explore the performance of LLMs on four log-investigation tasks that incident response teams commonly perform, ranging from triaging alerts generated by detectors to identifying persistence mechanisms on compromised systems. AuditBench consists of system audit logs collected from Linux and Windows machines, and spans over 50 different security investigation scenarios, including both malicious and benign activity. Using our benchmark, we evaluate and analyze the performance of five frontier LLMs at analyzing audit logs for attack investigations. Our analysis illuminates how LLM performance and error profiles vary according to different design choices, such as differences in model size, data representation, prompt construction, and specific investigation tasks. Additionally, we characterize the quality of the explanations produced by LLMs and the types of errors that models make across our benchmark. Collectively, our work provides a foundation for assessing the capabilities of LLMs for investigating security logs, novel insights for practitioners using LLMs in security operations, and important directions for future research.

Supervised fine-tuning with synthetic rationales is widely assumed to improve clinical prediction by teaching models not only what to predict but why. We test this assumption on five-year Alzheimer's disease and related dementias (ADRD) prediction from longitudinal health histories. In this setting, we generate training rationales using the correct labels, then fine-tune models to output a rationale followed by a label. Across a large-scale controlled experiment of 504 configurations, we find that Rationale First SFT consistently and substantially hurts predictive performance relative to Label Only fine-tuning. The degradation persists across model families and data scales, and is not resolved by using a reasoning-oriented base model. A fixed-protocol cardiovascular disease (CVD) validation reproduces similar findings. Crucially, the failure is not explained by poor rationale quality: human expert annotation confirms that the generated rationales are medically accurate and faithfully grounded in patient-specific evidence, and few-shot experiments show that the same rationales improve performance when used as inference-time demonstrations rather than training targets. A simple label-ordering intervention clarifies the mechanism: placing the prediction before the same rationale restores predictive performance. This suggests that conditioning predictions on generated rationales contributes to the degradation. We hope our work paves the way toward a more precise understanding of when and how rationale-based supervision helps and when it does not, guiding the responsible development of language models for high-stakes clinical prediction.

Frontier large language models can often recognize when they are being evaluated, a capability known as evaluation awareness. If models behave differently in evaluations than in deployment, this undermines the validity of evaluation results, which are a crucial component of current AI safety frameworks. We introduce EvalDetectBench, an open pipeline and benchmark for measuring evaluation awareness that works with any Inspect-compatible evaluation, allowing practitioners to test against current and future benchmarks. EvalDetectBench ships with a newly curated transcript suite covering current frontier system-card evaluations and diverse deployment sources. The benchmark serves two purposes: measuring how reliably frontier LLMs recognize that they are being evaluated, and assessing how detectable individual benchmarks are as evaluations. We identify two methodological choices in the existing literature that introduce systematic bias: the identity of the model that generated the deployment transcripts accounts for 11.25% of measurement variance and can reorder model rankings; and elicitation prompts selected for high performance on one model can perform near chance on others. EvalDetectBench corrects for both via per-model probe calibration and a stratified generator-harmonisation procedure.

Interactive AI agents must acquire the right context as efficiently as possible. When a user omits a constraint, preference, file, or task variable, an agent can proceed with a default assumption or spend tokens on a clarifying question, retrieval call, tool call, or prompt trial. We formulate this tradeoff as active inference for context acquisition. An inner inference step updates beliefs over a latent task state, and an outer decision selects the next context action, task action, or stop action to minimize expected free energy under cost. In deterministic settings, the epistemic term reduces to expected information gain, optionally normalized by token cost. We instantiate the framework in Optimal Question Asking (OQA), with exact posteriors and a dynamic programming oracle, and benchmark frontier language models on binary and multiway categorical tasks from 25 to 300 candidates. We also study clarification before generation and automated prompt optimization under token budgets. The formulation is model-agnostic and views active inference as a design principle for the context-acquisition layer of AI agents.

We present FrenchNews-7, a cross-publisher France-based French-language news editorial desk classification benchmark combining a large multi-outlet corpus, a URL-derived seven-class taxonomy, and a fine-tuned CamemBERT classifier. Labels are assigned via a hybrid pipeline combining publisher URL slugs with LLM annotation for structurally ambiguous cases, audited through an inter-rater study (2 humans + 2 LLMs; pairwise $\kappa \geq 0.766$, human–human $\kappa = 0.806$). We evaluate lexical, multilingual, and French-specific trained classifiers under both in-distribution and held-out-publisher settings, with additional comparison against zero-shot LLM baselines (GPT-OSS-120B, Mistral Small 3.2, Llama-3.3-70B) on the held-out pool. The strongest model, CamemBERT-base on full article text, outperforms headline-only input, generalizes to unseen outlets, and exceeds all three zero-shot LLM baselines on overall recall (0.799), with the gap concentrated in the ambiguous editorial-boundary categories Economie and Societe. Cross-publisher evaluation reveals uneven boundary stability: Sport, Culture & Loisirs, and International transfer cleanly, while Economie (recall = 0.517) is statistically indistinguishable from blinded human agreement on the same articles (0.55, $n = 40$), and Societe (precision = 0.577) absorbs boundary ambiguity, both suggesting editorial conventions rather than recoverable classifier headroom. The fine-tuned CamemBERT-base model, labeled manifest, reference collection scripts, and a reliability-tier guidance table are available at https://huggingface.co/LeFrenchNewsLab/camembert-base-frenchnews7 (model) and https://huggingface.co/datasets/LeFrenchNewsLab/frenchnews-7 (dataset).

Assessing whether multimodal content aligns with macro-societal values, such as peace, justice, and freedom, has become an increasingly urgent challenge. Existing frameworks are largely confined to safety-oriented taxonomies, text-only psychometric probes, or single-label classification. Therefore, we propose MAVEN, a hierarchical framework for macro-societal value evaluation of multimodal content, grounded in international human-rights instruments and cultural value theory. MAVEN organizes values into 6 primary dimensions and 72 secondary indicators, supporting multi-level quantitative scoring. Building on MAVEN, we construct a human-verified multimodal benchmark and a soft-match metric to evaluate VLMs' assessments across value dimensions. For evaluator optimization, we propose a span-adaptive variant of multi-level preference optimization for evaluator distillation, together with a training-free multi-role consensus strategy at inference time. We evaluate existing open- and closed-source VLMs on our benchmark, revealing shared tendencies and clear differences in macro-societal value judgments. Experiments show that our compact 2B evaluator matches its 8B counterpart in the same family and approaches frontier closed-source VLMs, offering a practical path toward scalable macro-societal value evaluation. Our SA-MDPO implementation and MacroValue-Bench are available at https://github.com/zzzzzzzzjj/MAVEN.

Recent advances in deep learning have significantly enhanced the capabilities of Natural Language Processing (NLP) and Vision-Language Models (VLMs). However, these advancements come with increased vulnerabilities, notably through backdoor attacks that pose severe security threats. This thesis addresses two critical dimensions of Trustworthy AI and Efficient Multimodal Representation Learning: (1) security through analyzing, detecting, and designing backdoor attacks in NLP and VLMs, and (2) efficiency through advanced multimodal representation methods tailored for clinical and medical imaging applications.

Retrieval-Augmented Generation (RAG) systems have emerged as a powerful process for allowing large language models (LLMs) to retrieve relevant information to use as source material during text generation. A critical yet under-explored component of these systems is the granularity at which source documents are segmented into retrievable chunks. The size of these chunks has the potential to significantly influence generation quality, contextual correctness, retrieval precision, and computational efficiency. Despite its importance, chunk size is often selected without proper evaluation of its impact on generation quality. Smaller chunks, such as individual sentences, may allow for precise retrieval by narrowing the focus of each chunk. However, they contain less information, which may limit the model's ability to generate coherent responses. Larger chunks, such as entire chapters, contain lots of broad information that may improve correctness, but also introduce additional noise and increase computational cost. Because larger chunks contain more information, the number of chunks returned to the model must also be considered. This paper evaluates how chunk size, along with the number of retrieved segments, influences generation quality and retrieval effectiveness. By comparing these configurations, this study seeks to better understand how document segmentation affects the performance and efficiency of Retrieval-Augmented Generation systems. segmentation affects the performance and efficiency of Retrieval-Augmented Generation systems.

Standard Retrieval-Augmented Generation pipelines rank retrieved documents by semantic similarity alone, without accounting for source provenance or credibility. This work evaluates a simple and interpretable modification to RAG retrieval ranking that incorporates domain-informed source reliability priors. Each document is assigned a prior lambda(s) based on its source type, and retrieval scores are reweighted using score(q, d) = sim(q, d) * lambda(s). The framework is evaluated against a similarity-only baseline on a 120-document health-domain corpus. In this controlled setting, source-aware reranking improves Precision@5 from 0.48 to 0.72 and reduces average adversarial document retrieval under the evaluated threat model, where low-credibility sources are identifiable via metadata. All experiments were executed on Rosie, the high-performance computing cluster at the Milwaukee School of Engineering, which provided the GPU-accelerated infrastructure necessary to run the full experimental pipeline reliably and reproducibly. These results suggest a potential mitigation strategy for source quality degradation in RAG pipelines, within the limits of the experimental setup described.

Large Language Models (LLMs) have achieved widespread adoption because of their strong reasoning and query-response capabilities. However, deploying them in embedded and edge computing environments remains challenging because of strict latency, memory, and energy constraints. Their large parameter counts and computational demands hinder efficient execution on resource-constrained platforms. Although model pruning has emerged as a viable solution for reducing scale while preserving performance, jointly optimizing layers, attention heads, and Multi-Layer Perceptron (MLP) dimensions remains highly complex. Exhaustively exploring this combined design space is computationally expensive and often leads to local optima or unstable configurations. To address these limitations, we propose a hardware-aware, multi-objective structured pruning framework. The proposed two-stage method explicitly targets latency and model size for efficient deployment on edge devices. In the coarse-grained stage, multi-objective depth pruning removes entire attention and MLP blocks to reduce computational load and memory usage. In the subsequent fine-grained stage, Parallel Bayesian Optimization (PBO) searches for the optimal layer-wise pruning ratios for pruning under latency constraints, while importance-based strategies rank the specific components to be pruned within each layer's allocated budget. Experimental results show that our approach reduces model complexity with minimal impact on commonsense reasoning tasks and zero-shot performance. Our method achieves a favorable trade-off among accuracy, latency, and model size, making it suitable for edge deployment. Across multiple LLMs at 37.5% and 50% pruning ratios, the proposed approach achieves better performance on commonsense reasoning tasks than existing methods while significantly reducing inference cost.