Papers

New venture start-ups need to “survive” through multiple stages of reaching milestone targets. We investigate the strategies for start-ups in a milestone-oriented setting. We examine a model of an entrepreneurial start-up firm, where its state is captured by a diffusion process. The entrepreneur can choose between multiple activities (or controls), which incur different cost and determine the drift and the variance of the process. Depending on whether the process reaches a fixed upper boundary or a lower one, the start-up firm succeeds or fails. Continuous-time stochastic models with multiple ($\ge 3$) controls are typically very challenging to deal with. In this work, we are able to completely solve for the optimal policy and provide an explicit characterization of its structure. In particular, the optimal policy only uses controls from a set characterized by a so-called efficient frontier curve that orders the controls by two intuitive measures: riskiness (drift-to-volatility ratio) and cost-effectiveness (drift-to-cost ratio). A unique feature of our model is that depending on the model parameters, the efficient frontier curves can be of different types, resulting in qualitatively different structures of the optimal policy. As far as we know, this is the first study that analyzes a stochastic control model which admits efficient frontier curves of different types. Our work provides start-up firms with intuitive measures to evaluate their activities and offers valuable insights on how the optimal strategies in a milestone-oriented setting change qualitatively contingent upon the specific scenario. We believe the results provide a foundational block in the study of entrepreneurial decision-making.

Recent progress in large-scale imitation learning for robot manipulation has been driven by leveraging datasets across a wide range of robot embodiments. However, achieving significant cross-embodiment transfer is often still challenging. In this work, we study the role of using behavior-aligned representations (e.g., object bounding boxes, language motions, end-effector traces of robot motion) in vision-language-action (VLA) models to promote cross-embodiment transfer. We hypothesize that by possessing invariances across embodiments while being predictive of robot actions, these representations can help unify large-scale cross-embodiment data to enhance transfer. To assess our hypothesis, we develop a simulation-based benchmark designed to assess transfer with diverse cross-embodiment data to new embodiments. Using this benchmark, we compare different representations and ways of incorporating them. We identify that end-effector traces can be particularly beneficial for transfer, representations are generally more useful with larger prior datasets, and can be used to benefit from action-free data. We also demonstrate that they can enhance sim-to-real cross-embodiment transfer, improving task completion progress of real robot policies pre-trained on simulation data by 28%. We provide videos of our evaluations at our website: https://ajaysridhar.com/barx/.

Tabular Foundation Models, such as TabPFN, have received a large amount of recent attention due to their performance on in-context tabular machine learning tasks, which often exceeds classical baselines. However, practical deployment considerations of these models has received less attention. In this paper we investigate the memory requirements for these models. We demonstrate that employing model compression approaches can enable memory reductions of up to 7.6 with similar levels of performance, reducing deployment requirements by nearly 87%. Our work provides insight to practitioners seeking efficient deployment of these models in practical settings.

An assistant can stop repeating a fact without removing it from memory. To study this difference, we install a support-vector gate in frozen Gemma 3 and record which stored keys and values belong to each exchange. A deletion request excludes the exchange's rows from the long-range readout and recalculates the gate on what remains. We check this operation against an independent refit, then compare it with running the model again on the conversation without the exchange. This second comparison matters because the exchange may already have influenced surviving memory rows. At 4B, the gated model passed checks for recall and feasible deletion on the same six of eight records admitted by the base model, at a perplexity cost under 2%. Admission fell at the smaller and larger checkpoints with the same configuration. The edited memory agreed closely with the local refit on the registered probes, and the model disclosed fewer deleted answers than when simply instructed to forget. However, an attack evaluated separately for each record could still distinguish edited memory from memory that never stored the record. Excluding an exchange's own rows therefore provides a way to edit and audit conversation memory, while leaving a measurable difference from rebuilding it without that exchange. Additional paired studies found no update-speed advantage for the current FP32 proxy and retained-answer matching below half in every tested condition.

Learning a strategic task changes more than what is directly taught: fine-tuning on one game can either enhance or degrade an agent's ability to reason in another. Understanding and predicting this transfer of strategic capabilities, however, remains a key challenge for large language models (LLMs). Normal-form games provide an ideal testbed for analyzing this phenomenon, as they feature explicitly defined payoffs and well-characterized equilibrium behaviours. In this work, we investigate whether game embeddings can explain and predict changes in LLM strategic capabilities following fine-tuning across different games. We propose a lightweight two-feature embedding that captures fundamental behavioural demands: the entropy of the Nash equilibrium and the sensitivity of optimal responses to an opponent's action. We show that while existing published structural embeddings primarily memorize game identities and fail to generalize, our behavioural embedding reliably predicts performance changes on held-out games. These results demonstrate that the transfer of strategic capabilities in LLMs is not dictated by the payoff geometry of a game, but by the underlying structure of the decision-making behaviour it requires.

The committor function describes the probability that a stochastic dynamical system, while transitioning between two metastable states, will first reach one state rather than the other. It is a central object in transition path theory, providing the accurate reaction rate and the characterization of the transition mechanism through reactive probability flux and the transition state ensemble. While the numerical methods for this function are well studied for over-damped Langevin dynamics where the friction damping coefficient goes to infinity, its computation in the general under-damped case remains an open challenge due to the lack of a variational structure in the underlying backward Kolmogorov equation. This restricts the quantitative investigation of the impact of the finite damping coefficient on rare transitions in real systems. In this work, by decomposing the infinitesimal generator of the underdamped Langevin dynamics into a self-adjoint and a skew-adjoint part, we develop an iterative scheme in which each subproblem has a variational form and can be solved efficiently using neural networks. Fundamentally, this approach can be viewed as a generalized Concus-Golub-Widlund (CGW) splitting method and each subproblem has a variational form, which can be solved efficiently. This iterative scheme also has the flexibility of introducing a preconditioning operator associated with a new overdamped Langevin dynamics, which naturally leads to the design of enhanced adaptive sampling for the subproblems. We validate the efficiency and accuracy of our method on benchmark systems including a one-dimensional double-well potential, the Muller-Brown potential, and alanine dipeptide. The proposed iterative variational framework provides a practical and extensible numerical tool for probing rare transition events in Langevin dynamics across varying damping coefficients.

Electrical submersible pumps (ESPs) are essential in offshore oil production, where unexpected failures can result in significant operational and financial losses. Accurate predictive maintenance for ESP systems remains challenging due to nonlinear operating conditions, class imbalance, and variability among pump units. To address these issues, this study presents a fault diagnosis framework that incorporates class imbalance awareness by employing Siamese contrastive representation learning and prior-corrected k-nearest neighbor (KNN) classification. The method first extracts discriminative features relevant to fault detection from vibration-domain indicators and engineered harmonic relationships. A Siamese neural network is trained with class-balanced contrastive pairs to construct an embedding space that clusters samples of the same fault type and separates different fault classes. To further mitigate class imbalance during classification, a prior-corrected distance-weighted KNN is applied. The framework is validated using a Leave-One-ESP-Out (LOEO) strategy to evaluate generalization to previously unseen ESP units. Experimental results indicate that the proposed framework delivers robust and consistent fault classification performance under realistic industrial conditions, supporting its potential for reliable predictive maintenance and intelligent ESP system monitoring.

Physiological signals support diverse clinical and monitoring tasks, yet existing physiological signal foundation models typically require task-specific adaptation for each task. Natural language provides a common interface for specifying different prediction objectives, but the ability of current models to follow such instructions across physiological signal modalities remains insufficiently evaluated. To address this gap, we introduce PhysioBench, a unified benchmark for physiological signal question answering. PhysioBench harmonizes annotations from 22 public datasets into 61.4 million questions across 30 tasks. Each question-answer pair is grounded in a signal segment and traceable to its source annotation. We evaluate 21 representative models, including large language models, vision-language models, time-series language models, and physiological signal foundation models under three complementary settings. The results show that none of the evaluated models achieves consistently strong performance across physiological signal modalities and tasks. The incorporation of natural language supports unified prediction across tasks, although performance remains sensitive to question formulation. Beyond these findings, PhysioBench offers an extensible platform for fine-grained analysis and future research on physiological signal understanding. Our codes are available at https://github.com/Leanna97/PhysioBench.

Detecting crop disease early is central to sustainable agriculture and food security under United Nations Sustainable Development Goal 2 (Zero Hunger), and is especially urgent in resource-constrained regions where expert diagnosis is scarce but low-cost mobile devices are widespread. This paper presents TinyCNN, a lightweight convolutional neural network for on-device plant disease classification. TinyCNN uses depthwise separable convolution blocks and contains only 193,190 trainable parameters with 110.05M MACs for a 224x224 input image. On the 38-class PlantVillage benchmark, TinyCNN achieves 98.88% test accuracy and 98.03% macro-F1 while being approximately 58x smaller than ResNet18 and 11.8x smaller than a MobileNetV2 teacher, directly reducing the energy, memory, and cost footprint of inference in line with Green AI principles. The paper further analyzes vanilla knowledge distillation as a sustainable model-compression strategy; an ablation over alpha in {0.3, 0.5, 0.7} and T in {2, 4} selects alpha=0.3, T=4, producing a distilled TinyCNN with 98.81% test accuracy. Finally, cross-dataset evaluation from PlantVillage to PlantDoc reveals a substantial robustness gap under real-world conditions, which a Grad-CAM analysis attributes to off-leaf, background-driven attention consistent with shortcut learning. TinyCNN is thus an energy-efficient, deployable building block for sustainable agricultural intelligence, while field robustness remains the key barrier to durable real-world impact.

Current Transformer interfaces index tokens by one or more integer coordinates, which determine their addresses inside attention. In RoPE and its multi-axis or hierarchical variants, the resulting address has the form $A(i)=R_1^{i_1}R_2^{i_2}R_3^{i_3}$, where the exponents are integer coordinates assigned after choosing a serialized token layout. When Transformers process new or large collections of data, this addressing scheme can produce unseen offsets or coordinate combinations, push repositories toward retrieve-and-serialize pipelines, and force new entities, records, or repository items to be represented by long token strings or identifier embeddings not seen in training. We introduce labeled incidence structures (LIS), in which each participating token or value is an endpoint with content $x$ and a structural index $i$. The index can include local position, relation role, relation instance, text unit, field, or content-derived identity. The model maps this index to a structural address $A(i)$, so adding new tokens, facts, text units, or repository items applies the same learned address rule to structural and content coordinates rather than requiring larger integer coordinates, unseen coordinate combinations, or new identifier embeddings. Attention scores endpoints $i,j$ using $q_i^\top P_{j\to i}k_j$, where journey consistency forces $P_{j\to i}=A(i)^{-1}A(j)$. When $i$ has several coordinates, such as position, role, and instance, coordinate independence is equivalent to factoring $A(i)$ into one address factor per coordinate. This recovers RoPE, RoPE-2D, and HiRoPE as special cases. This allows knowledge-graph (KG) roles, fact instances, and text units to enter the attention score directly. In controlled shallow diagnostics, the LIS address interface is implemented inside ordinary Transformer attention and yields promising results across text, KG, and $n$-ary settings.

Memoizing an expensive function of a sorted score vector is a data-structure problem before it is a numerical one: at a billion gridpoints, a hash map or a search tree spends most of its space on keys the grid already determines. We describe an implemented memo table that stores none. An entry's address is computed in closed form from the sorted argument itself, so $N$ values occupy $N$ slots, the argument is recoverable from the index, and the table can be memory-mapped and served from a file larger than RAM. Against a chained hash map it uses $5.7\times$ less memory at $37$M entries and $10.2\times$ less at $1.9$B, answers queries up to $2.9\times$ faster and builds up to $250\times$ faster; on 64 threads its construction needs no coordination; a sharded hash gains only $1.17\times$. Against an open-addressing table with inline keys it is $4$–$7\times$ smaller and $100\times$ faster to build but $1.5\times$ slower to query, a deficit we trace to the $O(d)$ index arithmetic. At $22$ GB on a $16$ GB desktop it serves each query in one disk access, where no key-storing container can be built; and its order-preserving addressing keeps a perturbation workload on the same pages that a hashed layout scatters. The closed form exists because the key set is the multiset combinations, whose index is the combinatorial number system. Memoizing Plackett–Luce normalization runs $25$–$55\times$ faster than Newton's method; memoizing $\alpha$-entmax thresholds does not pay. The contrast says when this structure is worthwhile.

Reliable and low-latency communication is a fundamental requirement for smart city services and Industry 4.0 applications enabled by NR-V2X networks. However, limited Road-Side Unit (RSU) deployment and complex urban propagation conditions often prevent Connected and Automated Vehicles (CAVs) from maintaining stable connectivity. This paper proposes an AI-driven Learning-to-Optimise (L2O) framework based on Graph Neural Networks (GNNs) for real-time multi-hop relay selection in NR-V2X systems. The vehicular network is modelled as a graph, where nodes represent CAVs and RSUs, and edges encode radio-link characteristics. An offline Mixed-Integer Linear Programming (MILP) formulation provides optimal relay decisions used as supervision for training an edge-aware Graph Isomorphism Network with Edge Features (GINE). Extensive experiments on realistic urban datasets demonstrate that the proposed approach achieves near-optimal connectivity performance, recovering up to 11.3% connectivity gain, while reducing execution time by orders of magnitude (up to 100 x speed-up) compared to MILP. The framework enables scalable and real-time network control, making it suitable for smart city and Industry 4.0 deployments.

Since GPT, most Transformers have repeated the same attention mechanism at every layer. Yet this design is largely a convention rather than a tested conclusion. When multiple sequence mixers are combined in one stack, improvements may arise from mechanism choice, placement, or both, making causal attribution difficult. We introduce Aether-7B-5Attn, a 6.59B-parameter mixture-of-experts model ($\approx$2.98B active) whose 49 layers contain seven sequence-mixing mechanisms arranged as a $7\times7$ Latin square. Because each mechanism appears exactly once in every row and column, the design guarantees balanced exposure across depth while eliminating placement confounds. To evaluate this principle, we build a parameter-matched proxy with four mechanisms arranged as a $4\times4$ Latin square over sixteen layers, matched to 700.9M parameters and trained with eight seeds per arm. The results reveal a clear dissociation. Rearranging a distributed heterogeneous stack into a balanced periodic cycle changes validation loss by only 0.16%, indicating that exact placement has little effect. In contrast, clustering the same mechanisms into contiguous depth bands incurs a 0.59% penalty, while replacing the heterogeneous stack with a homogeneous one incurs a 1.68% penalty. These results indicate that performance depends primarily on heterogeneous composition distributed across depth rather than on any particular permutation. We confirm this finding at 2.16$\times$ larger scale (1.514B parameters), where the homogeneous-stack penalty increases to 2.63% and removing the SSM-family mechanism produces a 3.20% degradation. We further report per-mechanism cost profiles, English and Korean evaluations, and a causal-safety audit of all 49 layers. We release model weights, training recipes, training code, logs, and architecture source code.

Online reinforcement learning (RL) algorithms frequently exhibit poor sample efficiency and unstable learning dynamics, stemming from systematic critic estimation errors that are exacerbated by greedy policy updates. Existing behavior-prior reinforcement learning methods attempt to alleviate this issue by relying on offline pre-training to learn behavior models from fixed datasets and using policy priors to constrain online policy updates. However, the limited quality of offline datasets often hinders the ability to provide high-value policies that can effectively guide policy updates. The absence of expert trajectories significantly impairs online policy learning, leading to low sample efficiency and suboptimal performance. To address these challenges, we depart from conventional behavior prior approaches and propose a Bidirectional Behavior Prior Distillation (B2PD) algorithm. B2PD leverages action-value priors to guide a conditional variational autoencoder (CVAE) in generating a high-value behavior support set. The resulting expert behavior priors are further distilled into the agent, effectively reducing inefficient exploration and enabling stable policy optimization, while establishing a bidirectional knowledge flow mechanism. Empirical evaluations on both state- and pixel-based tasks verify that B2PD substantially improves sample efficiency while maintaining stable policy optimization. More broadly, this work shows that enforcing high-quality behavioral support during online learning effectively mitigates critic-induced error amplification, enabling structured behavior priors to guide policy updates in a principled and sample-efficient manner.

Zero-shot essay scoring with large language models is usually demonstrated with proprietary API models, yet the settings where automated scoring is most needed, such as public schools grading thousands of essays under strict privacy rules, are often those where sending student writing to a third-party API is unacceptable. We ask how much capability survives when the model must be a sub-3B open model running fully locally in FP16, with a controlled study of four instruction-tuned models from two families (Qwen2.5 at 0.5B/1.5B/3B, SmolLM2 at 1.7B) on all eight ASAP-AES prompts on a single 8 GB consumer GPU, with bootstrap confidence intervals, Holm-corrected paired tests, and deployment-realistic variants of the key design choices. Three findings emerge. (i) Rubric-decomposed prompting beats holistic prompting for every model under batch min-max aggregation (though Qwen2.5-3B drops significantly on one prompt), and under mean aggregation two unrelated families land within 0.01 at the 1.5-1.7B scale. (ii) Mapping trait scores into the prompt range is fragile to grader calibration: one model compresses traits into a narrow low band (2-4 on 0-10) and naive mean aggregation collapses, while the min-max normalization of Multi-Trait Specialization repairs it (macro QWK 0.204 to 0.388) and stays within 0.03 when its statistics are frozen on 30 held-out essays. (iii) Signed error falls with essay length in eleven of twelve configurations, opposite to the verbosity bias reported for large LLM judges; normalized rubric decomposition largely flattens this slope for well-calibrated models. We anchor results honestly: the best local configuration (0.388) remains far below both the human inter-rater ceiling (0.769) and a length-only baseline (0.523), so we position sub-3B local models strictly for formative, human-supervised feedback.

Financial sentiment measures are widely used in empirical finance, but it remains unclear whether general-purpose large language models (LLMs) improve on existing finance-specific methods. This paper evaluates twelve sentiment models, including dictionary-based methods, finance-specific transformers, and open-source LLMs, using two criteria: linguistic validity and economic validity. We find that general-purpose LLMs achieve classification performance comparable to finance-specific transformer models without task-specific fine-tuning. However, higher classification accuracy does not translate into stronger economic relationships. Several models produce sentiment measures that are significantly associated with earnings surprises, but none is significantly associated with next-day stock returns. Model performance is strongest for announcements with large earnings beats or misses and substantially weaker for announcements with more moderate earnings surprises. These findings suggest that financial sentiment captures information about firms' economic performance but has limited ability to explain short-run market reactions

Retrieval plug-ins supply a deep forecaster with information its lookback window cannot carry. Published evaluations report consistent gains, and each credits its own mechanism. We show that the benefit belongs instead to the operating point: the relation between window length $S$ and dominant seasonal period $L$, an axis the standard protocol never varies. Stratifying the evaluation by that relation exposes the regime. At $S{=}12$, a simple control that repeats the last observed period beats the six standard backbones, in aggregate, on four of seven benchmarks by $8\%$ to $44\%$ of MSE. It beats the strongest plug-in we run on ETTm1 and matches it on ECL. It is worse by up to $25\%$ on the three datasets whose training-split spectra lack a concentrated, shared period. A controlled synthetic sweep of horizon, period, and window shows the benefit boundary tracks the period (correlation $+0.71$), not the horizon ($-0.23$). A paired control with no phase to recover nearly erases the effect, consistent with phase starvation. Zero-shot pretraining does not escape it: a foundation model trails trained backbones by $22\%$ to $50\%$ on the periodic benchmarks. Within our instrument, exact lookup matches graph diffusion: the payoff is consulting the record, not the machinery on top. Two interpretable statistics, a trend test and a staleness rate, predict the sign of the per-cell benefit at $0.76$ accuracy under leave-one-dataset-out evaluation, a suggestive margin over the $0.69$ majority rule, where a 22-feature stack manages $0.57$. We propose no new plug-in. The contribution is the regime map, the protocol that reveals it, and two statistics that screen it before deployment. Code: https://github.com/KurbanIntelligenceLab/retrieval-regime.

Knowledge graphs are usually integrated into question answering by encoding a retrieved subgraph with a graph neural network and fusing it with the language model in the online inference path. The same subgraph is therefore re-encoded from scratch every time a pair is scored, across training epochs, seeds, and evaluation runs, even though the knowledge graph never changes. We ask whether the retrieved knowledge graphs can instead be compiled once, offline, and then accessed as read-only memory. VisKG-LM shows that it can, by decoupling graph encoding from language reasoning. It serializes each retrieved candidate-specific subgraph as Relation-Labeled Paths and renders the result as an image whose two-dimensional layout preserves the branching structure of the paths. Each image is encoded once, offline, and cached for reuse. At inference, the language model contextualizes the question and candidate from text alone, and only its final layer consults the cached visual memory, reading both its global layout and its local relational detail. The graph information thus enters only after the text has been understood. On the test sets of CommonsenseQA, OpenBookQA, and MedQA-USMLE, VisKG-LMimproves over GreaseLM by $1.2$, $0.8$, and $4.3$ points, respectively, while matching or surpassing GraphVis, a $7$B vision-language model, with only about $400$M online parameters. Against a matched text-only control that receives the identical Relation-Labeled Paths, it gains $4.2$, $6.5$, and $5.1$ points across the three benchmarks. These gains show that the complete visual-memory interface adds value beyond path textualization alone and support compiled visual memory as an alternative to online graph propagation.

Auditing vision-language models (VLMs) for societal bias requires distinguishing direct algorithmic valuation disparities from confounders embedded within archival metadata. In this study, we audit Contrastive Language-Image Pretraining (CLIP) models using historical artwork metadata from the Metropolitan Museum of Art Open Access collection (N = 1,500 total objects; N = 743 attributed works: Male n = 534, Female n = 209; n = 618 anonymous). We establish a quantitative audit framework evaluating zero-shot CLIP logit differential scores across three semantic prompt pairs (masterpiece, quality, and influence). Unadjusted evaluations demonstrate high score convergence without a statistically significant main gender effect under OpenAI CLIP (mu_F = -0.0067 vs mu_M = -0.0035, p = 0.1829) or OpenCLIP (mu_F = 0.0171 vs mu_M = 0.0237, p = 0.1224). Two One-Sided Tests (TOST) confirm statistical equivalence across Cohen's d >= 0.25 bounds (pTOST < 0.005). Multivariate OLS regression controlling for artwork medium, creation era, and aspect ratio (R^2 < 0.02) confirms that artist gender has no statistically significant conditional effect (p > 0.20). High residual embedding variance (R^2 < 2%) indicates that global zero-shot valuation metrics operate near an embedding noise floor, showing that broad zero-shot prompt logit differentials are a coarse measurement instrument rather than proving absolute model fairness. We highlight two key caveats: (i) macro-level score equivalence reflects metric insensitivity to fine-grained visual-semantic features and does not preclude localized micro-level visual biases, and (ii) excluding 41.2% unattributed holdings reflects institutional survival bias. These results demonstrate the necessity of multivariate confound control, equivalence testing, and archival provenance auditing when assessing AI fairness in cultural heritage collections.

Where in a Transformer is the change from memorization to generalization functionally expressed? We introduce Transition Games–behavior-aligned exact activation games with paired non-generalizing controls–and find distributed utility gain with a prospective block-0 attention bias; selected degree-two modes account for 67–92% of its addition contrast across replacement games, and a disjoint exact path study confirms that block-1 MLP mediates more of their effect than all other tested downstream paths in 12/12 pairs. The sharper "MLP memorizes, attention generalizes" prediction instead reverses (-.331 bits/example at the memory anchor; 0/12 in the predicted direction), while routing onset, global rank collapse, and a prime-invariant architecture ridge also fail, identifying grokking here as spectral recoding of an existing distributed circuit rather than a module switch.

While AI systems increasingly shape high-stakes societal domains, their governance is limited by the lack of risk management methods that operate on real harms, taking their severity, and not only their likelihood, into account. As a consequence, AI risk management models remain compliance-driven and provider-centric, offering limited insight into how harms are dangerous, and on what should be the priority of intervention. The problem is amplified by the nature of harm data which are typically ordinal and multidimensional. To solve the problem, and offer an effective risk assessment methodology, in this paper we propose to model harm data by means of Lorenz Zonoids and Gini indices. To this aim we propose to extend them in a multidimensional setting, and show how to practically calculate them for a real AI incident data repository, provided by the Massachusetts Institute of Technology. The empirical findings indicate that environmental, infrastructure, property, physical, and democracy-related harms attain the highest values under the two multidimensional Gini indices and therefore exhibit the strongest concentration in their joint direct, indirect, and inferred severity-frequency distributions. These concentration patterns may help identify categories that warrant closer examination when mitigation priorities are determined.

Robert Smithson's 1970 land artwork Spiral Jetty, located in the north arm of Utah's Great Salt Lake, has alternated between submergence and exposure during severe lake decline. We analyze 1,744 co-registered Landsat 4-9 and Sentinel-2 image chips spanning every year and calendar month from 1984 to 2025. A 14-feature complexity signature combines Shannon entropy, multiscale permutation entropy, fractal dimension, lacunarity, gray-level co-occurrence texture, intensity statistics, and ImageNet-pretrained ResNet50 features. These measurements are compared with a 42-year monthly climate and hydrology panel from NASA GISTEMP, USGS NWIS, Open-Meteo, and the Global Carbon Budget. Bootstrap analysis shows that Shannon entropy is a weak proxy and does not support an earlier small-sample claim of positive correlation with global temperature. By contrast, coarse-scale permutation entropy and mean intensity track lake elevation strongly, with Spearman correlations of 0.85 to 0.88 and 95 percent confidence intervals excluding zero. The third principal component of the ResNet50 embeddings emerges without supervision as an AI climate axis, correlating 0.86 with cumulative CO2 and -0.83 with lake elevation. Image complexity leads lake stage by about three years, with Pearson r = 0.58 at lag +3 and a 95 percent confidence interval of 0.40 to 0.73. STL decomposition reveals a non-monotonic trend that rises from 1984 to 2015 and declines sharply thereafter as the lake approaches record-low elevations. Partial correlations controlling for month and sensor confirm robustness to seasonal and sensor effects. These results refine the art-as-thermometer metaphor into an art-as-leading-indicator-of-hydrological-state interpretation. The dataset, feature pipeline, and analysis code are released as a public benchmark.

Autonomous and intelligent transportation systems operate in complex urban environments where safety depends on interactions among vehicle behavior, environmental conditions, and vulnerable road users (VRUs) such as pedestrians and cyclists. Most advanced driver assistance systems (ADAS) employ reactive mechanisms that activate only after hazards have emerged, a critical limitation underscored by rising VRU fatalities in the United States. This study introduces PRISM (Proactive Risk Intelligence and Safety Management), an agentic multi-model safety architecture that transitions from reactive crash avoidance to proactive, continuous risk management. PRISM employs inverse crash-probability modeling to convert binary crash classifiers into dynamic, interpretable safety scores. Three specialized models addressing trajectory kinematics, environmental risk, and VRU interaction operate concurrently, coordinated by a reasoning layer incorporating reinforcement learning, contextual memory, and feature-level attribution. The system provides graduated safety interventions across four tiers, from silent monitoring to emergency alerts. Unlike rule-based systems with static thresholds, PRISM dynamically adjusts safety parameters in real time. Validated across 1,296 scenarios from three naturalistic driving datasets without dataset-specific retraining, the system yielded a mean safety score of 68 out of 100, classified 77.6% of scenarios as advisory, and flagged a near-miss rate of 3.8%, with 11% of scenarios escalating to intervention or emergency response. Feature attribution consistently identified trajectory risk and VRU proximity as primary safety factors. PRISM provides a unified, interpretable framework for proactive transportation safety with emphasis on VRU risk reduction in dense urban environments.

Preference elicitation is essential for aligning AI systems with human values. Prior approaches (e.g., for organ allocation) often ask stakeholders to compare the decisions of an algorithm (e.g., patient A vs. patient B). Such a decision-level approach conflates the means with the ends. Instead, we elicit preferences directly over allocation outcomes to learn a utility function for policy optimization. We construct a novel preference elicitation algorithm for linear utilities that outperforms prior techniques in practice. Our algorithm has two phases. The first phase learns cutting planes through pairwise comparisons to rapidly shrink the space of possible attribute weights and warm-starts the second phase by eliminating dominated regions. The second phase then provably converges to the user's utility function. We apply our technique to heart transplant allocation, where a policy must balance competing objectives such as post-transplant outcomes, waitlist mortality, geographic ease, and equity. Using our algorithm, we conduct a user study to learn and aggregate a human-value-aligned utility function. We evaluate the linear representation against LLM-based methods supplied with users' natural language explanations, finding that none of these methods is more accurate than the learned linear utilities. We use the resulting utility function to optimize heart transplant policies that are significantly better aligned with human values. Compared to the hindsight optimum, the status quo policy achieves a competitive ratio of just 0.66, while our method is near-optimal with a competitive ratio of 0.95.

Market microstructure simulation aims to model how liquidity, prices, and order flow evolve in electronic financial markets. Since market data reveal only one realized trajectory, many important questions are inherently counterfactual and require realistic trajectory-level simulation. Existing financial generative models, however, often model order events and market states, such as the LOB, in isolation, overlooking the dynamic interaction between order flow and liquidity in market microstructure. We propose the M3 (\underline{M}arket \underline{M}icrostructure \underline{M}odel), a state-event generative foundation model for market microstructure dynamics. M3 learns to generate future order-flow trajectories, while accounting for the evolving interaction between order events and limit-order-book liquidity. Trained on large-scale order-level real stock market data, M3 exhibits predictable scaling behavior, reproduces key market stylized facts, and enables practical simulation-based applications including forecasting, stress testing, and market-impact analysis. These results suggest a scalable foundation-model paradigm for counterfactual market simulation at the microstructure level.

Studying climate change requires reducing uncertainties in CO2 and CH4 emission estimates to better distinguish anthropogenic from natural sources, which motivates spaceborne measurements with improved revisit frequency and spatial coverage. In this context, the Horizon Europe SCARBOn project assesses a low-cost satellite constellation featuring the NanoCarb imaging interferometer as its core sensor for monitoring CO2 and CH4 emissions in the atmosphere. However, estimating CO2 and CH4 concentrations with high revisit and spatial coverage poses significant challenges: full-physics retrieval algorithms commonly used rely on repeated high-resolution radiative transfer (RT) simulations, which are computationally expensive when using line-by-line RT models. As an alternative, we propose in this study a feedforward multilayer perceptron (MLP) surrogate designed to accurately and efficiently predict top-of-atmosphere radiances in the CO2 weak band, using a combined mean absolute error (MAE) loss on radiances and RT Jacobians to preserve both spectral accuracy and sensitivity to geophysical parameters. Coupling the MLP-based RT surrogate with the NanoCarb instrumental response yields an efficient and precise forward model for NanoCarb measurements, which shows promising results for CO2 concentration retrieval.

Real-world LLM deployments increasingly rely on runtime-injected prohibitions–enterprise policies, PII redlines, tool boundaries–that vary per request and per tenant. Conventional post-training is structurally ill-suited: SFT hides the violation signal in compliant labels, and DPO's sequence-level preferences mismatch token-localized violations. We propose DUET, a token-selective on-policy distillation method for prohibition compliance. DUET pairs a teacher that sees the prohibition (positive) with an identical-weight teacher that does not (negative). Because the two teachers differ only in prohibition visibility, their per-token disagreement isolates the prohibition's causal effect–yielding a clean supervision signal uncontaminated by model capacity or mismatch. This disagreement drives two complementary mechanisms: signal cleaning, which discards agreement tokens as redundant or prefix-corrupted, and preference-directed learning, which pushes the student away from the negative teacher and toward the positive one at token granularity, embedding DPO-style optimization directly into OPD without offline preference data. We construct an industrial Prohibition-Compliance benchmark spanning five task families covering explicit-refusal, paraphrase robustness, and over-refusal. Across 1.5B-8B Qwen variants, DUET achieves 72.3-85.2% violation compliance while preserving 88-93% normal utility, dramatically outperforming teacher model and other distillation baselines. External evaluation on SysBench confirms improved safety alignment with minimal degradation on GSM8K and MATH-500.

Reinforcement Learning (RL) agents trained on a single reward signal exploit the gap between the designed reward and the intended behavior. This is particularly a problem when we are trying to imbue ethical behavior into RL agents. An agent can look ethical on average while concentrating its violations in a few bad episodes, and a creature in the environment harmed in one episode is not restored by good conduct in another. We compare four ways of training ethical behavior in Craftax, an open-ended survival benchmark. The four are: scalar penalties with termination, a linear multi-objective weight sweep, an adaptive Lagrangian constraint, and a non-compensatory utility optimized per episode under the Expected Scalarized Returns (ESR) criterion. All are evaluated under a single detector-based protocol that counts every violation in every episode without censoring. On the frontier of mean return against mean violation rate, the four methods are indistinguishable; per episode they separate sharply. At matched mean return, the ESR agent holds its stated budget of one violation in effectively every episode (worst-decile 1.04 +/- 0.07 violations), the Lagrangian leaks past the same budget (1.14 +/- 0.03), and the weight sweep's worst episodes double it (2.20 +/- 0.20). An observation-augmentation control attributes the separation to the training objective rather than to what the agent observes, and the per-episode guarantee costs nothing on the mean frontier. When ethical violations do not average away across episodes, we argue both training and evaluation must target the per-episode distribution rather than the mean.

Rigorous dataset partitioning is a foundational, yet frequently overlooked, prerequisite for reliable deep learning in longitudinal medical imaging. Naively shuffling small clinical cohorts routinely introduces covariate shifts and temporal sampling imbalances across training, validation, and test subsets, exposing downstream models to out-of-distribution evaluation. We address this vulnerability with an auditable Tripartite Dataset Analytics Framework that systematically characterizes spatial grid integrity, multi-parametric intensity fingerprints, and longitudinal temporal trajectories, quantifying the heavy-tailed feature dispersion and irregular, episodic sampling intervals typical of real-world clinical cohorts. Building on this characterization, we formalize an unsupervised spatio-temporal cohort-balancing standard operating procedure (SOP) that combines elbow-optimized K-means clustering over a standardized, six-dimensional joint intensity-temporal feature space with intra-cluster proportionate stratified sampling. On a longitudinal, contrast-enhanced $T1$-weighted brain MRI cohort (N=149), the protocol reduces the maximum cross-subset intensity bias from 34.1% under conventional random shuffling to under 2.1%, while aligning longitudinal follow-up intervals closely around the population mean. Monte Carlo stress testing across ten random seeds and three split configurations confirms that this alignment remains tightly bounded, in clear contrast to the substantial variability of random partitioning. The resulting protocol offers a reproducible, generalizable procedure for cohort engineering in variable-length longitudinal clinical imaging workflows.

As Earth Observation (EO) enters the Big Data era, the exponential volume of daily satellite imagery poses significant computational and storage challenges for classical Deep Learning (DL) models. Moreover, current approaches often struggle to generalize across heterogeneous sensors and volcanic environments while requiring large labeled datasets and substantial computational resources. These limitations are particularly critical for emerging On-Board Processing (OBP) applications, where memory, computational power, and annotated data are inherently limited. This work proposes a Hybrid Quantum AlexNet architecture for cross-sensor recognition of volcanic thermal activity at the global scale. The proposed model combines a classical convolutional backbone for high-level spatial features extraction with a parameterized quantum circuit (PQC) acting as a variational layer. By embedding high-level image representations into a high-dimensional Hilbert space, the quantum layer learns task-specific representations that enhance feature discrimination. Experimental results demonstrate that the proposed hybrid quantum model learns more discriminative feature representations, leading to improved cross-sensor transferability and robustness across heterogeneous volcanic environments using fewer trainable parameters and reduced training data than its classical counterpart.