arXiv:2606.09409v1 Announce Type: new Abstract: Pairwise comparisons combined with aggregation methods like Elo have become central to evaluating generative models, yet concerns remain that they reward superficial stylistic cues or display judge biases. In a more positive turn, we show that model rankings from pairwise comparisons strongly agree with ground-truth-based accuracy rankings when such ground truth is available for comparison. By converting five well-known benchmarks into free-form generative evaluations, we find that Elo rankings achieve a Spearman correlation above 0.9 with accuracy rankings and substantially outperform direct evaluation when the judge is weak. Furthermore, style and judge bias have only minor effects on model rankings, despite most judgments occurring on pairs where both candidate answers are correct (or incorrect). On such pairs, we find that repetition after the final answer (echo) is a causal driver of judge preference.
Science Journals
arXiv:2507.19700v2 Announce Type: replace Abstract: We propose a new framework for generating tabular synthetic datasets via disjoint generative models. In this paradigm, a dataset is partitioned into disjoint subsets that are supplied to separate instances of generative models. The results are then combined post hoc by a joining operation that works in the absence of common variables/identifiers. The success of the framework is demonstrated through several case studies and examples on tabular data that help illuminate some of the design choices that one may make. The advantages achieved by the disjoint generation include: i) An observed increase in the empirical measurement of privacy. ii) Increased computational feasibility of certain model types. iii) Ability to generate synthetic data using a mixture of different generative models. Specifically, mixed-model synthesis bridges the gap between privacy and utility performance, providing highly competitive performance on Accuracy and Area Under the Curve for downstream tasks while significantly lowering the empirical re-identification risk.
arXiv:2508.05163v2 Announce Type: replace Abstract: Security of supply is a common and important concern when integrating renewables in net-zero power systems. Extreme weather affects both demand and supply leading to power system stress; in Europe this stress spreads continentally beyond the meteorological root cause. We use an approach based on shadow prices to identify periods of elevated stress called system-defining events and analyse their impact on the power system. By classifying different types of system-defining events, we identify challenges to power system operation and planning. Crucially, we find the need for sufficient resilience back-up (power) capacities whose financial viability is precarious due to weather variability and weather-induced risk. Furthermore, we disentangle short- and long-term resilience challenges (from multi-day to annual scale) with distinct metrics and stress tests to incorporate both into future energy modelling assessments. Our methodology and implementation in an open energy system model (PyPSA-Eur) can be re-applied to other systems and help researchers and policymakers in building more resilient and adequate energy systems.
arXiv:2604.20897v2 Announce Type: replace Abstract: We develop a thermodynamic theory of algorithmic catalysis within the watts per intelligence framework, identifying reusable computational structures that reduce irreversible operations for a task class while satisfying bounded restoration and structural selectivity constraints. We prove that any class specific speed-up is upper-bounded by the algorithmic mutual information between the substrate and the class descriptor, and that encoding this information incurs a minimum thermodynamic cost via Landauer erasure. Combining these results yields a coupling theorem that lower-bounds the deployment horizon required for an algorithmic catalyst to be energetically favourable. The framework is illustrated on an affine SAT class and situates contemporary learned systems within an information thermodynamic constraint on intelligent computation.
arXiv:2606.09410v1 Announce Type: new Abstract: Prior work treats structured output as a reasoning tax, but this framing is incomplete: the cost of formatting depends strongly on a model's spare capacity. Using information-matched prose controls and a four-level schema complexity gradient, we separate format-specific effects from prompt-length confounds across 4 models and 5 benchmarks with 0% parse failures on successfully generated responses. We find that structured formats are capacity-dependent. Models with sufficient headroom absorb JSON constraints without degradation (Sonnet: $88.7\pm4.0$% JSON vs. $89.3\pm1.7$% CoT on MATH-Hard). In contrast, formats severely degrade models operating near their limits through two distinct mechanisms. First, under standard token budgets, Haiku drops 36.2pp ($p < 0.0001$) largely due to truncation. Second, even with extended budgets eliminating truncation, GPT-4o-mini drops 28.0pp ($p < 0.001$), revealing pure capacity competition independent of token exhaustion. This format penalty scales with schema complexity (McNemar $p < 0.0001$) and cannot be explained by prompt length alone. Furthermore, these results qualify claims of frontier model immunity: on AIME competition math, Opus 4.7 drops from 96.2% to 91.0% under JSON ($-5.3$pp; the displayed percentages are independently rounded, exact difference is $7/133 = 5.26$pp $\approx 5.3$pp). A delayed-structure ablation -- reasoning freely before formatting -- recovers most of the lost accuracy (3-run mean: 80--87%), supporting the capacity competition mechanism. The practical implication is not to avoid structured output, but to match it to capacity: when a model is near its limits, think first, format later.
arXiv:2606.09411v1 Announce Type: new Abstract: Large language models can be fine-tuned to encode prompt-borne secrets into fluent, seemingly benign outputs. This creates a steganographic exfiltration risk that is difficult to detect with output-level steganalysis. Recent work proposes mechanistic detection using linear probes that recover the secret from internal activations. We show that this defense can be systematically evaded, but that detectability can be recovered through a targeted data-level intervention. First, we extend the detection setup to include a non-linear MLP probe. We then adversarially fine-tune steganographic trojans across five base models: Qwen3-8B, Llama-3.1-8B, Ministral-8B, Qwen3-14B, and Phi-4-14B. The resulting models retain $58$--$79\%$ exact-match secret recovery while evading both ridge and held-out MLP probes, with $1$--$8\%$ average capability degradation across six benchmarks. We then give an information-theoretic characterization of this evasion. Successful evasion preserves recoverability while reducing low-order extractability of the secret from the content-aligned representation, forcing the payload into synergistic interaction with residual degrees of freedom. This motivates a recontextualization dataset that restricts these residual degrees of freedom. On this distribution, both ridge and MLP detectability are restored across all five evasive trojans. Overall, our findings show that activation-based steganography detection is vulnerable to adaptive evasion, but also that theory-guided evaluation distributions can expose otherwise hidden payloads.
arXiv:2507.12612v4 Announce Type: replace Abstract: Supervised fine-tuning performance for large language models depends strongly on how training budget is distributed across a heterogeneous set of tasks. In practice, mixtures are often fixed using simple heuristics (e.g., uniform or size-proportional sampling) that ignore task interactions, which can hurt transfer and waste budget on redundant sources. We introduce TaskPGM, a framework for learning continuous task mixtures via an energy-based model over tasks. Tasks form the nodes of a Markov random field: unary potentials capture per-task utility, and pairwise potentials encode inter-task relationships using behavioral divergences computed from predictive distributions of single-task fine-tuned models (e.g., Jensen--Shannon divergence and pointwise mutual information). Optimizing this objective yields mixtures that balance coverage against redundancy. We show that the resulting set function is weakly submodular under budget constraints, enabling approximation guarantees for discrete selection variants. Across multiple model families (LLaMA-7B, Qwen2-7B) and evaluation suites (BIG-Bench Hard), TaskPGM improves over standard mixing strategies and provides interpretable structure over task interactions.
arXiv:2508.06336v2 Announce Type: replace Abstract: We introduce Unsupervised Partner Design (UPD), a population-free multi-agent reinforcement learning method for robust ad-hoc teamwork. UPD generates training partners on-the-fly and selects them adaptively based on a learnability criterion, removing the need for pre-trained partner populations or manual parameter tuning. We show that this simple mechanism enables effective partner diversity and can be extended to joint partner-environment selection when a procedural level generator is available. Across Level-Based Foraging, Overcooked-AI, and the Overcooked Generalisation Challenge, UPD consistently achieves strong performance compared to both population-based and population-free baselines. In a human-AI user study, agents trained with UPD achieve higher returns and are rated as more adaptive, more human-like, and less frustrating than all evaluated baseline methods.
arXiv:2508.06659v2 Announce Type: replace Abstract: Reinforcement learning (RL) agents often struggle to generalize to new tasks and contexts without updating their parameters, mainly because their learned representations and policies are overfit to the specifics of their training environments. To boost agents' in-context RL (ICRL) ability, this work formulates ICRL as a two-agent emergent communication problem and introduces CORAL (Communicative Representation for Adaptive RL), a framework that learns a transferable communicative context by functionally separating latent representation learning from control. In CORAL, an Information Agent (IA) is pre-trained as a world model on a diverse distribution of tasks. Its objective is not direct return maximization, but world modeling and distilling its understanding into concise messages. The emergent communication protocol is shaped by a novel Causal Influence Loss, which measures the effect that the message has on the next action. During deployment, the previously trained IA serves as a fixed contextualizer for a new Control Agent (CA), which learns to solve tasks by interpreting the provided communicative context. Our experiments demonstrate that this approach enables the CA to achieve significant gains in sample efficiency and successfully perform zero-shot adaptation with the help of pre-trained IA in diverse online and offline environments, validating the efficacy of learning a transferable communicative representation.
arXiv:2508.11874v2 Announce Type: replace Abstract: Designing polynomial-time algorithms for approximate Nash equilibria (ANE) with provable worst-case guarantees is a fundamental open problem in algorithmic game theory. While large language models (LLMs) can generate candidate algorithms at scale, certifying worst-case guarantees requires formal analysis over all game instances -- a task for which no automated system previously existed. Here, we present LegoNE, a framework encoding expert proof strategies into a symbolic language that automatically compiles any candidate algorithm into a finite optimization problem certifying its worst-case guarantee. Integrating LegoNE with a reasoning LLM, we rediscovered an algorithm matching the best polynomial-time guarantee for two-player games, and discovered a three-player algorithm improving the best guarantee from $0.6+\delta$ to $0.5+\delta$ -- provably beyond the reach of the extension technique, the only previously known multi-player ANE design paradigm. These results show that encoding domain-specific proof strategies into a machine-tractable language can support LLM-driven discovery of algorithms outside known human design paradigms.
arXiv:2605.11309v2 Announce Type: replace Abstract: We consider the problem of reaching consensus in communication networks that are modeled by directed graphs. We assume the existence of a message authentication mechanism (such as digital signatures) to verify the integrity of messages. We identify the necessary and sufficient conditions on the directed communication graph for the following problems to be solvable: (i) exact consensus in synchronous systems; and (ii) approximate consensus in asynchronous systems.
arXiv:2606.09449v1 Announce Type: new Abstract: Complex reasoning tasks increasingly require systems to produce outputs whose correctness cannot be judged by exact match against a single reference. Autoformalization (AF) is a representative example; it asks a model to translate informal mathematical or logical reasoning into a formally checkable object, yet expert-validated formalizations do not scale beyond toy cases and a single informal argument can admit many valid formal renderings. Progress therefore depends on whether partial, structured proxies can substitute for exact references. We introduce a reference-free proxy-judge framework for AF that replaces gold-standard matching with a vector of per-axis property checks. The framework organizes the proxy along three structural scopes that cover global properties of the elicited object, per-module properties internal to its sub-components, and cross-domain properties that re-align it to the informal source, and aggregates each axis into a verdict vector. The vector drives a reflective refinement loop in which a violated coordinate routes the controller to a matching repair target, so each iteration changes only what is judged wrong. Under bounded judge noise, the expected intrinsic gap contracts geometrically to a noise-dependent plateau. Across seven formalization backbones on miniF2F, ProofNet, e-SNLI, and ProntoQA, refinement consistently lifts Pass Rate over the single-shot ICL baseline, and the per-axis proxy outperforms a matched scalar proxy on benchmarks where the baseline has room to improve. Structured proxy judgments therefore provide both a practical refinement signal and a theoretical handle on convergence when exact references are unavailable.
arXiv:2606.09450v1 Announce Type: new Abstract: LLMs have recently achieved strong results on formal proving benchmarks. However, existing evaluations remain heavily concentrated on competition-style problems and often fail to capture how models behave on longer, more dependency-rich mathematical developments. We introduce TheoremBench, a Lean4 benchmark designed to evaluate theorem provers beyond contest settings. The benchmark is built from nearly one hundred classical theorems and is released in two complementary forms: a plain main version containing one target theorem per instance, and a premised version that expands each theorem into a structured family of related proving tasks consisting of the main theorem together with automatically extracted supporting subtheorems. This design enables evaluation of not only whether the final theorem was proved from scratch, but also of partial progress through the internal proof structure of a theorem. Our experiments show that explicit premises substantially improve performance for Lean4-capable prover models. To provide a comprehensive evaluation, we introduce theorem-level coverage and token-efficiency metrics that expose qualitative differences in proof behavior. The results show that current provers remain strongly biased toward easy subtheorems and often solve theorems through long and inefficient tactic traces rather than compact proof plans. TheoremBench therefore provides a more fine-grained view of formal reasoning ability and highlights the importance of structural benchmark design for evaluating Lean4 theorem provers.
arXiv:2606.08106v1 Announce Type: new Abstract: Self-evolving agents improve by repeatedly proposing changes to their own prompts, skills, or workflows and keeping those that score higher on a small held-out set. Almost all effort has gone into the proposer that generates candidates; we argue the weak point is the acceptor, the rule that decides whether to commit a change. Applied hundreds of times against the same noisy dev estimate, the ubiquitous "keep it if the score went up" rule is uncontrolled adaptive multiple testing: the agent effectively p-hacks itself, accumulating false commits that make it churn and drift rather than improve. We recast committing as a sequential hypothesis test and propose PACE (Paired Anytime-valid Commit Evaluation), a training-free, anytime-valid commit gate. Each candidate is compared to the incumbent on identical instances and committed only when a testing-by-betting e-process accumulates decisive evidence, stopping early to save evaluations and controlling each candidate's false-commit probability at a user-set level even under optional stopping (a per-decision guarantee). On Qwen2.5 agents (0.5B-3B) self-evolving at the prompt level on GSM8K, SVAMP, and ARC-Challenge, greedy acceptance commits 30-42% false and 10-33% harmful edits when a genuine improvement is hidden among noisy proposals, while PACE commits the real one and essentially nothing else, matching greedy's held-out accuracy at sharply lower variance and about 18% lower evaluation cost. With no real gain available, greedy commits 13-21 spurious self-modifications per run (72-100% false) and degrades the most fragile agent by 4.9 points, while PACE holds at baseline. Reliability of self-evolution depends on the acceptor, not only on the proposer.
arXiv:2606.08266v1 Announce Type: new Abstract: James Dixon introduced the Data Lake in 2010. The pitch was simple: store data raw, postpone schema, cut up-front transformation. It promised flexibility and easier analytics. Fifteen years on, that promise has mostly gone unmet: survey after survey reports high failure rates, whether a big data program, a Data Lake, or a data science effort. This paper asks why. Reading 64 sources across academic work, analyst reports, and practitioner accounts, we found seven recurring anti-patterns, the Seven Deadly Sins of Data Lakes, and offer an explanation for them: Governance Debt, the compounding cost of governance decisions organizations keep deferring. A second pattern surfaced on its own: when governance gets hard, organizations drift back toward structured, warehouse-style approaches, a pull we name governance gravity. The term Data Swamp is used loosely in the literature, so we give it a working definition with measurable indicators, plus a qualitative rubric, the Governance Debt Assessment Model, for catching decay early. The root causes are organizational far more than technical. We also asked whether the newer paradigms, Data Lakehouse and Data Mesh, absorbed the lesson; the technology advanced, the organizational record barely moved. For practitioners we provide two tools, a Reality Check Framework and a Stage-Based Intervention Matrix. The paper rests on more than the analyst literature: it draws on a primary catalogue of close to five hundred field reality checks recorded over fifteen years of building and rescuing enterprise Data Lakes in financial services and telecommunications across Morocco and West Africa. Assembled independently of that literature, the catalogue lands on the same anti-patterns, surfaces two dimensions the literature under-reports, operational debt and engineering-discipline debt, and reads the problem from an emerging-market vantage.
arXiv:2606.08118v1 Announce Type: new Abstract: In millimeter-wave (mmWave) vehicular networks, dense base station (BS) deployments expand the user association (UA) decision space while dynamic blockages cause link quality fluctuations, posing critical challenges for effective mobility management. Traditional Multi-Armed Bandit (MAB) frameworks assume stationary reward distributions and fail to handle the rapid context-reward mapping shifts caused by vehicle mobility and transient blockages. To address this, we propose Blockage-Aware Non-stationary Dynamic Bandit (BAND), a fully distributed, channel state information (CSI)-free mobility management framework for mmWave vehicular networks, formulating UA as a non-stationary contextual bandit problem, enabling online adaptive optimization without requiring central coordination or offline training. BAND employs a cumulative sum-based change detection (CUSUM-CD) to dynamically narrow the active BS set, reducing exploration overhead while tracking reward distribution shifts. Proactive blockage detection suppresses transient signal degradation in the reward estimation process. Simulations demonstrate over 40% regret reduction and up to 33.1% network communication rate improvement compared with hypercube-based contextual bandit baselines, with robustness validated across varying blockage rates and network configurations.
arXiv:2507.22876v2 Announce Type: replace Abstract: The Satisfiability problem (SAT) is fundamental in computational complexity theory and has a wide range of industrial applications. Optimizing modern SAT solvers in real-world settings is quite challenging due to their intricate architectures. While automatic configuration frameworks have been developed, they rely on manually constrained search spaces. Here we develop AutoModSAT, a framework that uses large language models (LLMs) to automatically optimize SAT solvers. AutoModSAT combines an LLM-compatible modular solver design, unsupervised prompt optimization to diversify generated functions, and an efficient search procedure based on presearch strategy and a $(1+\lambda)$ evolutionary algorithm. Extensive experiments across a wide range of datasets demonstrate that AutoModSAT achieves $40\%$ performance improvement over the baseline solver and $30\%$ improvement over the state-of-the-art solvers. Moreover, AutoModSAT also attains a notable speedup compared to the parameter-tuned alternatives of the state-of-the-art solvers over most of the test datasets. These results demonstrate the potential of LLM-guided heuristic discovery for optimizing complex SAT solvers.
arXiv:2508.00917v2 Announce Type: replace Abstract: Connected autonomous vehicles (CAVs) must simultaneously perform multiple tasks, such as perception, prediction, planning, and control, to ensure safe and reliable navigation in complex environments. Moreover, through vehicle-to-everything (V2X) communication, cooperative perception and driving among CAVs can be enabled, thereby mitigating the limitations of individual vehicles, while it also introduces stringent latency, reliability, and bandwidth constraints. Traditionally, tasks are addressed using separate models, which leads to high deployment costs, increased computational overhead, and challenges in achieving real-time performance. Multi-task learning (MTL) has recently emerged as a promising solution that enables the joint learning of multiple tasks within a unified model. This offers improved efficiency and resource utilization. To the best of our knowledge, this survey is the first comprehensive review focusing on deep MTL in CAVs. We begin with an overview of CAVs and MTL to provide foundational background. Then, we review MTL approaches across key functional domains in CAVs, including perception, prediction, planning, control, as well as V2X communications and radio resource management (RRM). For the first four domains, we categorize existing works under ego vehicle-only (onboard-only) and V2X-enhanced cooperative (multi-agent) paradigms. We further discuss V2X communications and RRM as communication-centric MTL problems. Finally, we discuss the strengths and limitations of existing methods, identify key research gaps, and provide future research directions aimed at advancing MTL methodologies for CAV systems.
arXiv:2606.08122v1 Announce Type: new Abstract: Predicting a user's next Point-of-Interest (POI) based on their historical check-in records is a fundamental task in location-based services. While recent methods incorporating large language models have shown strong reasoning capabilities and promising results, they typically formulate the prediction task as a one-step trajectory-to-location mapping problem, making predictions prone to shallow trajectory correlations and historical frequency bias. We argue that users rarely choose locations directly and instead, they usually first form a traveling intention and then accordingly select specific POIs. Motivated by this insight, we propose IntentPOI, a two-stage intention-guided reasoning framework. In the thinking stage, we infer users' intermediate intentions by incorporating historical mobility patterns, similar peer behaviors, and the temporal contexts. In the acting stage, we first construct a compact candidate pool, and then perform intention-guided reasoning to identify locations that best align with the inferred intention. By explicitly decoupling intention inference from location prediction, IntentPOI transforms the next POI prediction from direct trajectory matching into intention-guided reasoning. Extensive experiments on three real-world datasets demonstrate that IntentPOI consistently outperforms eleven state-of-the-art baselines.
arXiv:2508.08217v2 Announce Type: replace Abstract: Hazardous environments such as chemical spills, radiological zones, and bio-contaminated sites pose significant threats to human safety and public infrastructure. Rapid and reliable hazard mitigation in these settings often unsafe for humans, calling for autonomous systems that can adaptively sense and respond to evolving risks. This paper presents a decision-making framework for autonomous vehicle dispatch in hazardous environments with uncertain and evolving risk levels. The system integrates a Bayesian Upper Confidence Bound (BUCB) sensing strategy with task-specific vehicle routing problems with profits (VRPP), enabling adaptive coordination of unmanned aerial vehicles (UAVs) for hazard sensing and unmanned ground vehicles (UGVs) for cleaning. Using VRPP allows selective site visits under resource constraints by assigning each site a visit value that reflects sensing or cleaning priorities. Site-level hazard beliefs are maintained through a time-weighted Bayesian update. BUCB scores guide UAV routing to balance exploration and exploitation under uncertainty, while UGV routes are optimized to maximize expected hazard reduction under resource constraints. Simulation results demonstrate that our framework reduces the number of dispatch cycles to resolve hazards by around 30% on average compared to uninformed baseline dispatch strategies, underscoring the value of uncertainty-aware vehicle dispatch for reliable hazard mitigation.
arXiv:2510.01661v3 Announce Type: replace Abstract: Multi-step manipulation in dynamic environments remains challenging. Imitation learning (IL) is reactive but lacks compositional generalization, since monolithic policies do not decide which skill to reuse when scenes change. Classical task-and-motion planning (TAMP) offers compositionality, but its high planning latency prevents real-time failure recovery. We introduce SymSkill, a unified framework that jointly learns predicates, operators, and skills from unlabeled, unsegmented demonstrations, combining compositional generalization with real-time recovery. Offline, SymSkill learns symbolic abstractions and goal-oriented skills directly from demonstrations. Online, given a conjunction of learned predicates, it uses a symbolic planner to compose and reorder skills to achieve symbolic goals while recovering from failures at both the motion and symbolic levels in real time. Coupled with a compliant controller, SymSkill supports safe execution under human and environmental disturbances. In RoboCasa simulation, SymSkill executes 12 single-step tasks with 85% success and composes them into multi-step plans without additional data. On a real Franka robot, it learns from 5 minutes of play data and performs 12-step tasks from goal specifications. Code and additional analysis are available at https://symskill.github.io/ .
arXiv:2606.09499v1 Announce Type: new Abstract: World models have recently seen a rapid growth in both their popularity and capability as more data efficient tools for generating robot training data or simulating real world environments, with many works proposing their integration into the robot learning pipeline. While highly practical, in this work we demonstrate that world models introduce a uniquely stealthy and effective data poisoning entry point into the robot learning supply chain that can result in the deployment of unsafe or otherwise compromised robotic policies despite training on seemingly safe ground truth training data. In contrast to traditional data poisoning techniques which directly implant dangerous trajectories into sold or uploaded datasets, our novel attack methods inject malicious prompts or compromising transition dynamics into visibly safe teleoperated datasets which are only activated once fed through a world model as input. This can result in the generation of synthetic, dangerous robot training trajectories and subsequently unsafe or compromised robot policies. We demonstrate the effectiveness of our attacks against both state of the art action conditioned and text conditioned world models, showing a full end-to-end backdoor on a downstream DRL policy and a proof-of-concept for the VLA setting. Overall these findings necessitate research into more secure world models and reevaluating their position within the robot learning supply chain.
arXiv:2606.09451v1 Announce Type: new Abstract: Humans rely on spatially dense, geometry and force-aware tactile feedback at high temporal resolution for dexterous manipulation. While vision-based tactile sensors enable dense force estimation, they are limited by camera frame rates, motion blur, and data bandwidth. Event-based optical tactile sensors offer an attractive alternative with microsecond temporal resolution and low motion blur, but existing methods are restricted to predicting only net forces. We introduce the first framework for dense 3D force field reconstruction using event-based optical tactile sensors. Our approach estimates 3D surface displacements from event data and maps them to forces via the inverse Finite Elements Method (iFEM). Shear displacements are recovered through the proposed event-based marker tracking algorithm, while normal displacements are predicted by a convolutional neural network trained on a collected dataset of synchronized force-displacement-event data. Experiments demonstrate accurate reconstruction of physically grounded forces, achieving a mean absolute error of (0.14 N, 0.10 N, 0.93 N) over force ranges up to (4 N, 4 N, 20 N), while operating at an average of 100 Hz. This work constitutes a first step toward enabling dense force feedback for high-frequency control in robotic grasping and dexterous manipulation.
arXiv:2606.08146v1 Announce Type: new Abstract: Fraud detection in payment, e-commerce, and telecommunications systems requires accuracy at the individual level, robustness under severe class imbalance, and ease of understanding for risk managers. Existing methods fall at least one of these requirements: automated machine learning systems search a fixed numerical space without semantic awareness of the dataset; graph neural network-based methods require pre-defined relational graphs and remain opaque at the individual-decision level; and the design of general-purpose large language model (LLM) agents does not consider the recall and precision constraints specific to real-world fraud detection. In this paper, we propose SAGE, the first end-to-end LLM-driven multi-agent framework for fraud detection. SAGE coordinates three dedicated agents that make decisions based on a six-layer Data Diagnostic Tree (DDT) and a Markov decision process guided by natural-language gradients, automatically optimizing the model under a fraud-specific reward. On five fraud datasets and five LLM backbones, SAGE wins $96.00\%$ of method--dataset comparisons and improves F1 by an average of $40.86\%$ over baselines. The code is available at https://github.com/yichenC1c/SAGE.
arXiv:2606.08221v1 Announce Type: new Abstract: Designing OLED molecules with targeted optical properties remains challenging due to the scarcity of high-quality data and the limited reliability of conditional control in generative models across chemical motifs. Here, we benchmark a token-conditioned autoregressive language model for OLED molecular generation in a realistic low-data regime. A GPT2 model is pretrained on large chemical corpora, augmented with discrete property tokens, and fine-tuned using multi-task optimisation. Conditioning targets vertical absorption energy and oscillator strength, with the HOMO-LUMO gap included as an auxiliary electronic descriptor. Generated molecules are evaluated at the TDDFT level to assess distributional fidelity and controllability. The generated library reproduces the dominant optical-property support of the training distribution while shifting towards lower molecular weight and fewer heavy atoms. Token-level control is consistently directional across conditioning bins, but is not fully orthogonal and exhibits local calibration irregularities. A chemotype-resolved analysis further shows that controllability depends strongly on local electronic environments: moderately conjugated aromatic-carbon motifs are associated with improved joint target satisfaction, whereas electron-withdrawing motifs, particularly aryl nitriles, show systematic red-shifting and reduced controllability. These results establish a quantitative benchmark for conditional OLED molecular generation and show that model reliability must be assessed in chemically meaningful subspaces rather than from aggregate property distributions alone.