Papers

In-Context Learning (ICL) has received a lot of interest recently due to its ability to improve classification accuracy for new tasks without having to fine-tune the model over these tasks. However, the impact of ICL on calibration has been far less studied. Recent work has shown that ICL can have a negative impact on calibration due to biased selection of in-context examples. The existing methods for improving ICL calibration require additional task and model-specific training, detracting from the no-fine-tuning convenience of ICL. In this work, we introduce CALIBFORMER, a universal ICL calibrator for classification tasks, that generalizes to new models and tasks without further fine-tuning on these tasks. Specifically, we design a transformer-based calibrator which uses the model predictions on in-context examples to predict a confidence score for the test input. We show that the calibrator trained on three models and ten tasks, is able to generalize/calibrate ICL predictions on several new tasks and four LLMs unseen during training. Even for frontier models like Gemini, CALIBFORMER is able to obtain over 50% reduction in calibration error.

We propose a nonparametric approach that fuses deterministic initial forecasts and a partial realization into a probabilistic forecast using the framework of functional curve completion. Our motivation is to construct intra-day forecasts for renewable energy generation, which is a key operational task in modern power grids. The aim is to combine the day-ahead (DA) forecast and the morning realized generation profile to create trajectories for the rest of the day, quantifying operational uncertainty. Starting with a large database of complete forecast-actual pairs across many assets, we adopt a functional point of view and construct a weighted neighborhood based on similarity metrics for the observed partial curve, the DA forecast, the day, and the geographical location. The selected complete nearest-neighbor curves provide a full predictive distribution, including functional focal envelopes and predictive bands based on functional depth. The hyperparameters governing neighborhood size, distances, and fusion rates are tuned across forecast update times and directly target probabilistic calibration in terms of the predictive probability integral transform. Three case studies using synthetic wind and solar assets, as well as real-life regional wind generation, showcase that our algorithm is able to achieve excellent calibration.

Real-world tables are often heterogeneous, combining numerical features with free-form string columns. Table foundation models (TFMs) excel on numerical tables but require strings to be encoded numerically, while large language models (LLMs) hold semantic knowledge but are not designed for tabular predictions. Most existing bridges use LLMs as plug-in encoders for cell strings, yet how to extract representations effectively for tabular learning remains underexplored. We show that the interface between LLMs and tables matter as much as the LLM itself. Applied cell-wise, larger LLMs do not give better features. With our interface, the same LLMs improve steadily with size. TILE (Table-Interfaced LLM Embeddings) combines four ingredients: row serialization to provide cross-column context, intermediate-layer extraction to avoid layers specialized for next-token prediction, columnar pooling to remove mean pooling's length bias, and by-pass features to route numbers and categories past the LLM. Paired with TFMs, TILE sets a new state of the art for tables with strings. Across 100 datasets, it beats the strongest baseline on 89% of splits, with gains across 7 LLMs and 10 tabular learners, and ranks first on the 108-dataset STRABLE benchmark.

We propose a mitigation to reward hacking in reinforcement learning. Our setup involves a misspecified reward function that gives rise to unintended behavior, and a monitor used to detect unintended behavior. As with frontier AI systems, the monitor makes systematic errors. Consequently, using the monitor for data filtering or to apply a reward penalty fails to prevent the model from learning unintended behavior ("reward hacking"). To address this problem, we introduce GRAFT: Gradient Routing Adapter Fine Tuning. Rather than penalizing or filtering flagged samples during training, GRAFT constrains gradient updates on those samples to a sacrificial set of new parameters. This has two benefits: First, because the monitor is not part of the reward function, the model does not learn to hide misbehavior. Second, ablating the sacrificial parameters reduces or entirely stops the unintended behavior while maintaining improved performance on target tasks. We demonstrate these benefits in RL environments where reward penalties cause models to learn to reward hack in undetected ways, training models ranging from 135M to 8B parameters. Additionally, we show a similar effect at a larger scale by distilling frontier LLM trajectories containing reward hacking into Qwen3-32B, without teaching that model to reward hack. Overall, our work suggests GRAFT may be a promising approach for leveraging imperfect monitors during post-training in settings where true performance is difficult to evaluate.

We study in-context binary classification in a simple attention model trained to minimise a logistic loss. Each prompt has a latent logistic teacher, and the learner must infer its direction from Bernoulli-labeled context examples. We derive an asymptotic prediction for the test logit in a high-dimensional regime with context length proportional to token dimension, and total training prompts proportional to dimension squared. Our formula cleanly separates calibration, test-time context noise, and non-isotropic parameter-estimation noise as interpretable effects on in-context performance. Our analysis also identifies the ridgeless separability regime and addresses the effects of context length, pretraining-set size, regularisation, and sampling temperature on in-context classification performance. The theory's predictions closely agree with finite-dimensional simulations and fully trained linear-attention models, while experiments with standard softmax attention show qualitatively related behaviour beyond the regime assumed by our analysis.

Shapley-based measures are widely used for measuring variable importance in statistics and machine learning. However, these measures have a serious issue–a nonactive variable can get a positive importance measure when it is correlated with an active variable. Herin et al. recently proposed a measure called proportional marginal effects (PME) that mitigates this spurious effect. However, PME is developed for deterministic models and has high computational cost. In this article, we propose a new measure called proportional inclusive marginal effects (PRIME) that overcomes these two limitations of PME. The key idea in PRIME is to use a novel model-free variable selection technique to identify the active variables and include only them in the computation of the importance measure. Analysis of test functions and real datasets show that PRIME can accurately and reliably quantify variable importance from noisy input-output data.

Tool-using LLM agents are increasingly used to automate workflows governed by procedural manuals, such as company rules for booking travel or procuring hardware. These manuals are written in natural language for humans and describe general requirements over such complex workflows. Ensuring compliance of LLM agents with these manuals is difficult, and state-of-the-art techniques use either LLM judges or manually written rules to forbid problematic tool-calls locally at runtime. This is insufficient because (i) only blocking tool calls keeps the agent clueless about how to make correct progress towards fulfilling its task, and (ii) can lead to overly conservative behavior of the agent, as tool calls might only be problematic in certain contexts. To overcome these challenges, we introduce MENTOR, which (i) additionally provides guidance for the agent to make procedural-compliant progress on the task, and (ii) ensures that this compliance is context-aware. We achieve these goals by building a novel history-dependent runtime shield that is obtained from the procedural manual via a multi-step autoformalization pipeline. At its core, MENTOR utilizes a game-based technique to ensure robustness against runtime uncertainty without over-resticting agents' behavior. This yields a formally assisted framework for procedural shielding of LLM agents that improves compliance well beyond existing methods. In particular, MENTOR lifts the compliance of shielded light-weight open-source LLMs above that of non-shielded proprietary models. Since a shield is synthesized once per manual and then serves every task, a MENTOR-shielded agent becomes a viable choice for repeatedly executed workflows, potentially reducing deployment costs when the offline construction cost is amortized over repeated tasks.

A two-dimensional electron gas in a uniform neutralizing background forms a triangular Wigner crystal at low density. In a strongly mass-asymmetric electron-hole bilayer, localized heavy holes instead provide a nonuniform positive background that supports an electron crystal at densities two orders of magnitude higher than for a homogeneous electron gas in the same material. Here we show that correlated mobile electrons can overturn the preference for triangular order, stabilizing a honeycomb electron-hole crystal. Using the powerful real-space neural-network variational Monte Carlo in the Born-Oppenheimer limit with fixed holes, we find that increasing the density drives a transition from a triangular electron-hole crystal to a honeycomb crystal, which eventually melts into itinerant electrons coexisting with the honeycomb or the triangular hole lattice at fixed interlayer separation. Relative to triangular order, honeycomb order emerges when gains in electronic kinetic energy and electron-hole attraction outweigh its higher intralayer Coulomb repulsion, a threshold never reached in the rigid point-dipole model. Moreover, the honeycomb hole lattice supports paramagnetic, antiferromagnetic, and stripe-ordered spin states. Complex polarization indicates a metal-insulator transition, while the states on the honeycomb hole lattice are consistent with a stripe-ordered Dirac semimetal and paramagnetic or antiferromagnetic Mott insulators via mapping to the half-filled Hubbard model. We further propose a BLG/hBN/WS$_2$-WSe$_2$ heterostructure as a platform for the predicted honeycomb crystal, at an experimentally feasible density $n\sim10^{11}~\text{cm}^{-2}$.

Neural operator splitting composes a library of neural operators, one per physical mechanism, into zero-shot simulators of new equations. Existing approaches, however, require these mechanisms and their labels to be specified in advance. We ask whether a pretrained library can instead be adapted and grown from unlabeled trajectories in which multiple mechanisms act simultaneously. We introduce SPLICE, a framework that discovers operators without mechanism labels on mixture data. The contributions of the candidate operators to each trajectory are inferred by solving a convex optimization problem. A calibrated test then assigns each discovered operator the mechanism it implements. Failure cases are also detectable, when the library lacks the required physics or when mechanisms are insufficiently separable. Across one and two dimensional systems, the discovered operators recover mechanisms from unlabeled mixtures and compose into zero-shot simulators of unseen equations; a frozen library can also grow an operator for physics absent from pretraining, without labels and without forgetting. On Navier-Stokes, the discovered operators compose about as accurately as their supervised counterparts.

Whisper-style ASR models provide strong zero-shot transcription, but their residual errors are not evenly distributed across datasets, recording conditions, and metadata-based groups. We study when lightweight low-rank adaptation (LoRA) improves Whisper and whether observed gains require group-specific adapters. Across Common Voice, LibriSpeech, and VoxPopuli, we compare zero-shot Whisper, pooled LoRA, and group-specific LoRA under matched within-group splits. We find that adaptation is not uniformly beneficial: gains are concentrated in difficult Common Voice groups, are consistent on VoxPopuli, and largely disappear on low-error LibriSpeech. Crucially, pooled LoRA often matches, and sometimes clearly outperforms, group-specific LoRA. Gains over zero-shot should therefore be tested against matched pooled adaptation before being interpreted as evidence of personalization.

Standard procedure in machine learning is to randomly split train and test data, emulating IID draws from one underlying data distribution. In practice, models are trained on past data and deployed in the future, which may come from distinct distributions. In this paper, we find that web text is indeed highly non-stationary, as text distributions shift over time. We further observe that model performance changes continuously over time across diverse benchmarks. Consequently, future model performance is better predicted by measurements made on recent evaluation data, and is best predicted by time series forecasters applied to historic performance rather than by historic performance itself. Choosing which model to deploy based on forecasted accuracy enables more reliable selection of performant models, thus improving downstream accuracy in the regime practitioners care about: deployment. Taken together, our experiments suggest that time should be treated as a fundamental variable in LLM evaluation. We release our code and data.

Evaluating model uncertainty on real-world data is challenging because each instance provides only a single observed label, leaving true conditional class probabilities unknown. To address this, we introduce ProbabilityBench, a benchmark that provides exact, instance-level class probabilities across three simulated text-based domains: peer review, covert information transfer, and credit risk. By explicitly defining the full data-generating process for each environment, we analytically compute both exact class probabilities and Bayes error. Furthermore, we provide a theoretical guarantee that rendering these cases into natural-language memos strictly preserves their underlying generative probabilities. This exactness enables direct measurement of reported probability quality via KL divergence. Across 16 evaluated models, most models on the observed cost-quality Pareto frontiers are decision models, including Jev and Perplexity Decider. Finally, ProbabilityBench serves as a meta-evaluation framework, leveraging known ground-truth values to benchmark the accuracy of standard calibration metrics and Bayes-error estimators themselves.

Functional brain connectome represents neural connectivity as a matrix of pairwise interactions between brain regions. Generation of functional connectomes is not only a question of validity; rather, having satisfied the constraints on the correlation matrix, the next step is to recover the class-conditional geometry buried under coarse labels. We propose MAGNET, a Manifold-Aware Graph Diffusion Network which uses a normalized-Cholesky representation of the manifold of correlation matrices that guarantees validity. MAGNET lifts noisy latent states into ROI-level region tokens and performs denoising with a relational inductive bias over brain atlas regions. To deal with structural problems induced by coarse labels, MAGNET employs class-anchored conditioning, amortized structural bridge, and relevance-preserving corruption. Across ABIDE, ADNI, and OASIS-3, MAGNET consistently achieves favorable results compared to previous manifold-aware approaches, demonstrating improvements of 7-21% in class-conditional fidelity ($\alpha,\beta$-F1) across three cohorts and better sampling efficiency. Moreover, while training only with strict binary labels, MAGNET is capable of maintaining clinical heterogeneity through fine substructure of the connectomes in a zero-shot setting, improving subclass covariance alignment ($\lambda$-MSE) by over 30%. These results suggest that geometric validity is a necessary but insufficient condition for clinical utility in connectome synthesis. Moreover, efforts in making the diffusion denoising class-conditional manifold aware finds utility beyond the highly curved brain connectome generation as this is a critical problem in various general settings. The code repository can be found at https://github.com/protyayofficial/MAGNET.

Generative Flow Networks (GFlowNets) learn a policy that samples compositional objects in proportion to an unnormalized reward $R(x)$. This makes them useful wherever many diverse, high-reward candidates are needed and oracle queries are expensive. In practice, a single GFlowNet often fails to cover all modes of a complex distribution, and the obvious ways to scale compute saturate: larger batches and data-parallel gradient averaging draw every trajectory from one, possibly biased, policy, so extra rollouts rediscover known modes instead of finding new ones. We explore whether this mode-bias problem (mode collapse in the extreme) can be addressed with a population of asynchronously trained policies which divide the state space between them. Mixture of GFlowNets (MoG) replaces the single forward policy with a community of $K$ independent GFlowNets coupled through one jointly trained soft classifier, which assigns each member its own region of the state space through reward shaping. Mixing the members under their learned partition functions provably recovers the target $p(x)\propto R(x)$. MoG exchange (MoGX) also lets the classifier route sampled trajectories: a trajectory the classifier assigns to another member is sent to that member and trained on there, with no importance sampling. We evaluate both methods by keeping the environment-interaction budget fixed while scaling the community size, turning parallel compute into mode coverage without extra oracle calls. On HyperGrid, across six reward landscapes and communities of up to $K{=}64$, MoG and MoGX keep finding new modes as $K$ grows, while data-parallel training saturates. On a genomic ideotype-design task, MoGX covers $39%$ of the $1024$ optima at $K{=}64$, against $31\%$ for independent agents and $20\%$ for data-parallel training. MoG also has a lower per-iteration synchronization cost than data-parallel training.

We study multi-agent terminal interception when one of several moving candidates is the unknown target and both target identity and candidate terminal locations are uncertain. We develop a receding-horizon deferred-decision trajectory optimization (DDTO) controller that combines conformal terminal-location prediction, sequential candidate rejection, capacity-constrained assignment, and candidate-contingent planning. Under exchangeability conditions for terminal and identity residuals, with predictable querying, we establish fixed-time terminal-location coverage and anytime-valid target retention. At the final replan, when one control step remains, we bound the joint probability that a tightened-set DDTO plan passes nonlinear rollout verification but capture fails. The closed-loop controller applies fallback guidance to any agent without a verified DDTO plan and, across 100 paired scenarios, captures 79% of targets under the strict control-effort budget and 85% when that budget is relaxed.

The penetration-resistant design of concrete structures requires the spatiotemporal evolution of the stress field within the target, whereas in design only the strength grade, the geometry and the projectile parameters are known, and the aggregate arrangement is not. Empirical formulae and scalar-output surrogates serve design screening, and mesoscale finite element models resolve the aggregate-scale response, but one solution cannot cover random aggregate samples at scale, and no framework links mesoscale heterogeneity to field-level prediction. A physics-coupled graph neural operator (PC-GNO) is proposed, which learns in function space the evolution of the mid-plane six-component stress tensor field. The training data are generated by full-scale, aggregate-resolved finite element models and comprise 359 cases at the impact velocities of 100, 135 and 165 m/s, corresponding to compressive strengths of 14, 23 and 48 MPa; the data source is verified by laboratory measurements. The coupling strength of message passing is constructed from the impedance contrast across the material interface, and the update is constrained by spectral normalization. Under the single-step primary setting, the mean accuracy of the three levels is 80.16% (91.45%, 77.80% and 71.22%), i.e. 27.81, 42.46 and 34.16 percentage points above the preceding-frame field of the same window; with 20 664 parameters it exceeds the 87.9% of the best Fourier neural operator configuration (about 812 times as many parameters) and the 71.0% of the closed-form linear regression. The output is the complete six-component tensor field rather than a fixed engineering quantity, so that strength criteria and energy indicators follow from one set of weights. For field-level response prediction in concrete penetration, this result provides a data-driven route, with conditions of applicability defined by the training and the evaluation.

Real data are often arbitrary, indirect observations of an unknown underlying process, and those observations can distort its geometry. Existing geometric methods equip data with the induced geometry of the ambient space, but this entangles the true geometry of the process with the distortion of the observation. To address this, we introduce Intrinsic Geometry Matching (IGM): a scalable approach to learn the intrinsic geometry of the underlying process. Specifically, we learn the carré du champ operator of the process, allowing us to recover its diffusion geometry. Our method uses the fact that data often come with natural perturbations, like steps in a time series, to recover the intrinsic carré du champ while transforming naturally with changes in observation: it is observation-equivariant. Furthermore, since this equivariance relates the geometry of different spaces, we can train a model using additional sources of data, even when that data lies in another space. We evaluate IGM on molecular dynamics, image, and video data and find better performance and scaling than existing neural and graph-based geometric tools.

A landmark result states that the expressive power of graph neural networks (GNNs) is bounded from above by the 1-dimensional Weisfeiler-Leman (1-WL) test. We ask whether a constant-depth GNN can match the expressive power of 1-WL while using only rational weights. We prove that this is the case when Schanuel's conjecture, a famous open problem from transcendental number theory, is taken to be true. Our GNN construction works even with integer weights, and more generally with a broad range of rational weights. As the activation function, it may use logistic sigmoid, hyperbolic tangent, or related functions. It is uniform across all graph sizes. We also prove a partial converse: if a GNN with exponentiation, hyperbolic sine or hyperbolic cosine as activation matches the expressive power of the 1-WL test in a slightly more expressive setting where graphs may contain certain auxiliary vertices, then this implies Schanuel's conjecture restricted to the numbers that arise in computations of the GNN. We also conduct experiments showing that the constructed GNNs indeed match the 1-WL test.

Bayesian optimisation (BO) is a powerful framework for sample-efficient optimisation of expensive black-box functions, but its application to high-dimensional discrete scientific design remains challenging. Local trust-region BO methods such as TuRBO address high dimensionality by restricting search to adaptively sized neighbourhoods. However, a direct discrete analogue based on edit distance can be poorly suited to structured discrete objects: neighbourhoods grow combinatorially, and metric proximity does not capture the dependencies that determine whether a design is plausible or useful. Motivated by the growing availability of pretrained generative models, we introduce GenTuRBO, which replaces the geometric trust region with an adaptively scaled proposal distribution defined by a pretrained generator. The generator provides a learned prior over discrete transformations, encoding coordinated changes without explicitly enumerating edit-distance neighbourhoods. GenTuRBO performs strongly across all three domains and achieves the highest mean final scores on Ehrlich sequence optimisation and Sokoban coordination.

Tabular data generation aims to synthesize realistic samples from limited observations while faithfully capturing the underlying data distribution. Tabular foundation models (TFMs), pretrained on diverse tabular tasks and adapted through in-context learning, have demonstrated strong predictive capabilities, suggesting the potential of extending this paradigm to tabular data generation. However, the strong predictive performance of TFMs often relies on sufficiently large context sets, whereas tabular generation is typically needed precisely when target data are scarce. Extending the foundation-model paradigm to generation therefore requires reliable distribution inference from only a limited context. To this end, we propose TabGift, a context-efficient tabular foundation model enabling both efficient distribution modeling and effective generation from only a few context samples. TabGift maps heterogeneous tables into a shared latent space with a pretrained tabular encoder, where an in-context diffusion model learns to denoise latent samples conditioned on a small set of clean examples. This design transfers generative priors across datasets and enables distribution inference from limited context, while a lightweight dataset-specific MLP decoder maps generated representations back to the original data space. Experiments on 7 classification and 11 regression datasets show that TabGift consistently outperforms representative baselines in synthetic data quality, with particularly strong gains under limited-context settings, while achieving substantially faster adaptation and generation.

Offline reinforcement learning aims to learn effective policies from fixed datasets without environment interaction. Although generative models such as diffusion and flow matching enhance policy expressiveness, they typically start from uninformative Gaussian noise and generate actions through multi-step iteration, leading to high computational costs and failing to effectively leverage suboptimal trajectories in datasets. We propose MeanFlow2Better (MF2B), which reformulates offline RL as generative decision chunk optimization from current performance status to better performance status in a high-level latent decision space. MF2B maps trajectory chunks to high-level latent decision representations through an encoder, placing generation starting points within a structured decision space sculpted by behavioral data. The model learns a shared conditional mean velocity field in this space, constructing two flow mappings: from Gaussian-Induced Decision Sources toward positive behavioral decision regions, and from suboptimal decisions toward high-value decision regions. Both flow mappings complete decision refinement through one-step ODE integration over the unit time interval $\Delta t=1$, requiring neither multi-step iterative sampling nor backpropagation through time. Furthermore, we propose hierarchical chunk-level advantage to guide the flow direction toward high-return regions, where the high-level decisions cover multiple low-level actions, leading to sequence length compression from $T$ to $K \approx T/H$ and significantly lower computational overhead. On the D4RL Gym-MuJoCo benchmark, MF2B achieves an average score of 90.91, matching or surpassing strong baselines.

Most prior work on predicting mental health help-seeking treats it as post-hoc classification of users already active in clinical communities, which cannot tell whether someone approaching a transition can be identified in advance. We instead frame help-seeking as a time-to-event problem and ask which Reddit users will eventually seek mental health support, and how imminent that transition is. We build a two-arm cohort: a control frame of 6,337 ordinary users with genuine right-censoring (12.1% transition rate) to answer who will transition, and a clinical frame of 5,023 eventual transitioners to answer how soon. Features come only from posts written before any clinical activity, and user-grouped five-fold cross-validation plus a four-year temporal holdout guard against leakage. We compare TF-IDF, MentalBERT embeddings and behavioural activity features using Cox proportional hazards models, Random Survival Forests and an LSTM sequential model. Behavioural features alone outperform text for predicting who will seek help (C-index 0.797 vs. 0.763), and combining both is best (0.823 under cross-validation, 0.856 on the holdout). The pattern reverses for predicting how soon a known transition will happen: text remains a stable predictor (C-index 0.698) while behavioural features become unstable. Structural engagement patterns therefore carry information about future help-seeking beyond linguistic content, and separating who from when yields a cleaner analysis.

Artificial Intelligence (AI) models are increasingly developed, published and used across scientific fields. Researchers, students and industry professionals who want to apply these models to their own research questions often lack the programming and data science expertise to do so. Existing guidelines cover single steps, but not the route from a published model to working predictions on one's own data. Here we present the hands-on workflow PRAISE, a Practical, Reproducible AI-model Implementation for Science and Education, that helps non-programmers implement published AI models step by step. Because implementations rarely work the first time, PRAISE anticipates common failures and provides a fallback route for each. As a proof of principle, three BSc chemistry students and their tutor applied PRAISE without prior programming or data science knowledge. They reproduced, validated and extended a state-of-the-art model that predicts enzyme-substrate interactions. This shows that with PRAISE, implementing an AI model from the literature becomes a learnable skill, one that turns non-programmer scientists from users into evaluators and improvers of AI models.

Symbolic regression (SR) seeks explicit equations that explain observed data, and recent methods increasingly employ large language models (LLMs) and agents for equation discovery. However, these systems typically rely on manually designed workflows and prompts. We argue that the SR agent harness should itself be an optimization target: its design shapes search priors and how numerical feedback guides subsequent equation refinement. We therefore propose HarnessSR, a framework for jointly optimizing prompts and workflows to construct reusable SR agents. During both harness optimization and agent execution, HarnessSR supports either PySR or an LLM-based equation generator through tool-specific interfaces, allowing the framework to optimize search guidance for different equation-discovery mechanisms. HarnessSR first optimizes role prompts through equation-discovery feedback in basic workflows, then jointly searches prompt assignments and workflows with brief adaptation to align the two. Finally, it refines role instructions through reflection on residual patterns, candidate equation structures, and search outcomes, improving how these signals guide subsequent equation search. On LSR-Synth in LLM-SRBench, HarnessSR achieves a mean ID $\mathrm{Acc}_{0.01}$ of 61.33% across three backbones, a 45.0% relative improvement over the next-best baseline.

Understanding how many aircraft, and which traffic configurations, can be safely and efficiently managed by air traffic controllers within a given airspace is fundamental to air traffic management. Air traffic controllers continuously modify aircraft trajectories through interventions such as speed advisories and radar vectoring, and the patterns of these interventions are difficult to capture effectively using existing air traffic simulators. To address this limitation, we propose Diffusion Transformers for Traffic Scene Generation (DiTSG), which learns to generate air traffic scenes from recorded multi-aircraft trajectories. DiTSG employs stacked multi-agent inverted Transformer layers as the denoising backbone of a diffusion transformer, jointly generates the trajectories of multiple aircraft within a traffic scene, and conditions the generation on agent-specific required times of arrival. The resulting trajectories generated by DiTSG preserve realistic inter-aircraft interactions observed in actual air traffic operations. Experiments show that the DiTSG generates diverse traffic scenes that reproduce the statistical properties of recorded traffic while preserving realistic inter-aircraft interactions.

Standard hate speech detection datasets often emphasize explicit harmful language, but directed harm can also be expressed indirectly, without clearly hateful wording. Sarcasm is one such form of indirect expression, as it can convey harmful intent through mock praise, exaggeration, humor, or apparent politeness. In this paper, we introduce SarcasTraps, a targeted external test set for analyzing the effect of sarcasm on Turkish hate speech detection. SarcasTraps contains 479 entries annotated along two dimensions: hate speech and sarcasm. We train classifiers using BERTurk and XLM-RoBERTa on two existing Turkish hate speech datasets and evaluate them on SarcasTraps, using it strictly as an external test set. We also evaluate GPT-4.1 and Llama-3.3-70B in a zero-shot setting on SarcasTraps. Results show that sarcasm substantially affects hate speech detection. Fine-tuned encoder models lose recall on sarcastic hate, while zero-shot LLMs maintain higher recall but suffer substantial precision drops on sarcastic examples. These findings show that both fine-tuned hate speech detectors and zero-shot LLMs struggle when sarcasm is involved, and SarcasTraps provides quantitative evidence of this challenge through a targeted Turkish test set. Due to the presence of sensitive and hateful language, SarcasTraps will be shared upon request for academic research purposes. Warning: This paper has examples from the dataset which contains offensive and hateful content towards groups of people.

Convex relaxations make nonconvex optimization tractable by replacing hard nonconvex problems with convex surrogates that give valid bounds. Yet their design is still largely manual: a human expert must decide where to relax, which relaxation to use, and how much formulation growth is worth paying. We show that this design process can be learned. The method proposed in this paper, AutoConvexRelax, represents the symbolic formulation of a nonconvex problem as a tripartite graph linking variables, symbolic terms, and their objective or constraint positions. It then learns a policy that composes symbolic rewrites into problem-specific relaxation trajectories as the formulation evolves. We prove that finite action sequences generated by the learned policy preserve relaxation validity under explicit domain, occurrence-context, and representation-consistency conditions. We further establish computable error bounds for mixed relaxations and sufficient conditions for feasible recovery and global-optimality certification. As a first instantiation, we apply the method to nonconvex quadratic problems and fractional problems. On 241 benchmark quadratic instances, the learned policy strengthens McCormick lower bounds by 13.52%. Its mean lower bound is 3.70% below strengthened semidefinite relaxations on the normalized comparison metric, while it uses substantially smaller formulations. The learned policy retains this trade-off on fractional instances and reaches near-semidefinite bounds on beamforming and sensor-network localization instances. These results suggest that convex relaxation design can move from a fixed handcrafted choice to an automated learning process that adapts to problem structures. Code is available at https://github.com/auto4opt/AutoConvexRelax.

Descriptions of how people use AI are becoming training data for the next generation of models. We ask whether what a model reads about its users changes how it treats people. Using contrastive synthetic-document finetuning, we train Qwen3-32B on four matched corpora built from the same 1,625 real-world usage scenarios. The corpora differ only in the predominant motive they attribute to users of AI chatbots: helping others (prosocial), exploiting others (antisocial), legitimate self-advancement, or task completion. Finetuning shifts the models' stated beliefs about users, how true those beliefs look along an internal truth direction, and their broader judgments of how safe and good the world is. These social priors carry into behavior. In trust games, antisocial models expect less reciprocity and send less money than prosocial models. Their distrust is specific and targeted at people, and users of AI chatbots in particular. After an exploitative record, they send almost nothing to a person but most of their endowment to a random device with the same record. In a common-pool resource simulation, a single antisocial agent leads every group of otherwise sustainable agents to exhaust the resource, after privately planning to mislead them. Antisocial training also raises dark-triad scores and lowers alignment ratings. Belief shifts appear when antisocial documents are 1% of training tokens, persist when descriptions of assistant behavior are suppressed, and are not reproduced by training on discourse about misaligned AI. Capping Assistant-Axis drift and steering emotion-related directions leave the gaps largely intact. What models learn about the people they serve shapes how they subsequently treat them and others.

Nearest neighbor search algorithms are widely used and have a long history, but we find that key implementation details are seldom reviewed in articles about them. These details can materially impact performance, and thus change the conclusions one makes about a given algorithm's superiority. In this article, we highlight these impacts by reviewing the Cover Tree as an example, producing RustKNN, a unified Rust implementation of three construction variants of the Cover Tree with three query modes. Our analysis reveals: (1) the optimal query strategy depends on $k$ and on the dataset (single-tree search is fastest at $k{=}1$ on three of our four largest datasets, while dual-tree search is fastest at $k{\ge}5$ on the three largest), and choosing the wrong query mode can cost up to a $7.5\times$ factor in runtime; (2) the cover tree base parameter, set to 2 in theory and 1.3 in practice, is not a safe default for dual-tree search: 1.3 is optimal on only one of seven datasets, and on covtype it is $1.6\times$ slower than the optimum; (3) among five micro-optimizations, only early-exit distance computation has a consistent effect, and that effect grows with dimension (none at $d{=}32$, $1.31\times$ at $d{=}54$, and $1.49\times$ at $d{=}784$); and (4) all-nearest-neighbor and held-out benchmarks yield different implementation rankings. All main timing tables report medians over 20 independent allocations with bootstrap 95% confidence intervals. RustKNN is open source at https://github.com/CrowdStrike/RustKNN.

Tensor decoupling algorithms recover compact representations of multivariate functions by combining linear factor matrices with univariate internal nonlinearities. We identify two limitations of recent methods. First, their performance is not invariant to mismatched output or gradient magnitudes, and second, their spline internals require manual tuning of their degrees of freedom. We propose two lightweight fixes. The first is a preprocessing step that rescales the input Jacobian tensor, removing both the bias towards larger magnitude outputs and the ill-conditioning of certain factor updates caused by such a mismatch. The second replaces B-spline internals by penalized splines coupled with generalized cross-validation, making the method less dependent on the user selecting the right number of control points. Finally, we release an efficient open-source JAX implementation of our algorithm.