Papers

Long-horizon spacecraft trajectory forecasting suffers from error accumulation due to the absence of corrective observations in the forecast regime, making reliable uncertainty estimation crucial for safety-critical decision-making such as space domain awareness and conjunction assessment. While high-fidelity physics-based orbit propagators provide accurate deterministic forecasts, they typically lack calibrated uncertainty estimates over long horizons. We introduce a Predictor–Corrector framework in which a physics-based continuous-time $\textit{deterministic}$ forecaster is augmented with a learned continuous-time $\textit{probabilistic}$ Corrector that models forecast errors. The proposed Corrector can be wrapped around an existing deterministic propagator to improve forecast accuracy while producing sharp and calibrated full-covariance uncertainty estimates. The Corrector is based on Latent Neural Controlled Differential Equations (Latent NCDEs) and models the probabilistic temporal evolution of forecast errors in continuous time, naturally supporting irregular sampling and missing features. We further introduce a loss function that promotes calibration and sharpness in long-horizon uncertainty propagation. We evaluate the proposed framework on long-horizon spacecraft trajectory forecasting using real-world data from NASA's Crustal Dynamics Data Information System (CDDIS), wrapping the Corrector around NASA's General Mission Analysis Tool (GMAT). Across forecast horizons of 2–4 days without observations and six rolling test windows, the proposed approach consistently improves accuracy and uncertainty calibration compared to deterministic baselines and Latent ODE-based correctors, demonstrating the effectiveness of the continuous-time probabilistic Corrector for trajectory forecasting.

As artificial intelligence advances into the era of Embodied AI, live musical interaction urgently needs to break free from the limitations of offline, unidirectional generation, achieving a "virtual synergy" capable of low-latency, dynamic interplay. To address this, this technical report presents LK_Jam, a real-time, bidirectional human-computer interactive music generation system based on a lightweight Gated Recurrent Unit (GRU) and a high-performance audio host architecture. In the algorithmic representation layer, this system abandons the computationally expensive fixed time-grid. Instead, it constructs a multi-dimensional sparse event stream integrating time-shifts, continuous harmonic embeddings, and role-aware encoding, enabling the model to accurately capture turn-taking logic and micro-timing in a single-step inference. In the engineering implementation layer, this paper builds a strict multithreaded lock-free communication bridge using C++ and the JUCE framework, incorporating the RTNeural inference engine designed specifically for real-time audio. By utilizing compile-time network topology solidification and a zero-allocation (allocation-free) mechanism, the end-to-end overhead of autoregressive decoding is strictly locked at \(O(1)\) complexity, structurally mitigating the risk of audio thread dropouts in DAW plugin environments. Furthermore, this study designs a three-stage progressive training strategy, achieving a leap from basic chord harmonization to expert-level interaction. Preliminary observations and architectural analysis demonstrate that while ensuring musical coherence and interactive role-play, the proposed system successfully challenges extreme real-time engineering constraints, offering a highly robust and deployable technical paradigm for next-generation AI co-performers in live music.

Forecasting future events is a critical challenge for large language model (LLM) agents, spanning domains from elections and monetary policy to financial markets. However, evaluating progress on this task presents a fundamental trade-off between efficiency and environment fidelity. While live evaluation benchmarks suffer from an inherently slow feedback loop, existing retrospective replays typically restrict agents to static, pre-frozen databases that sacrifice the environmental realism of actual deployments. To tackle this issue, we introduce Agentic Time Machine (TM), an infrastructure that approximately reconstructs the web state at any chosen past time by filtering post-cutoff content. Leveraging this evaluation infrastructure, we further propose a planner-solver-aggregator multi-agent framework that breaks each question into diverse analytical angles, gathers evidence in parallel, and combines the results into a single forecast. Experiments show that offline scores under TM correlate strongly with live FutureX scores, validating that TM offers a fast and reliable sandbox for forecasting-agent evaluation. On FutureX-Past and Polymarket evaluated under TM, our framework achieves the highest score among strong closed-book, tool-augmented, and self-consistency baselines. On the official FutureX live leaderboard, our system achieves the best average rank over four consecutive weeks, including 1st place in May Week 1. As of June 17, it also ranks 1st on FutureX's official eight-week overall leaderboard.

We introduce a benchmark that is fully self-contained, needs no ground truth, and rises with the models it measures. Language models compete at making analogies and subjectively grade one another; nothing enters from outside. The benchmark reproduces GPQA Diamond, a keyed benchmark of expert-written questions, at r = 0.98, audited for a leak and found clean. We hypothesize that both benchmarks measure the same thing in different ways: a language model holds its knowledge as archetypal contexts, relationship patterns valid across many topic domains. GPQA instantiates the required knowledge in one domain; the Metanym Game instantiates one archetype into several domains, generating analogies, no reasoning required. The reasoning feature of an LLM hardly changes the game's ratings, while it lifts GPQA, which requires derivations. In the game, a player writes a context template whose slots, filled with a set of keywords from a topic domain, instantiate a factually true description of that domain; the instantiations are each other's metaphors, and keywords filling the same slot are metanyms, metaphorically synonymous. Correctness is settled sentence by sentence. Ground truth is replaced by the SVD of the factual rating matrix: its left and right singular vectors rate the players as judges and as generators, two ratings from one factorisation, to our knowledge a first for an LLM council of peers. On the subjective criteria, judges are weighted by their rating consistency under a swept calibration anchor. Generating and judging are different skills: on this roster the strongest generators were middling judges. A council of the five best issues the official ratings; its contestable seats keep it current, a candidate steering signal for self-improving AI. The paper is accompanied by a validating package that recomputes every number.

Understanding the generalization capabilities of visuomotor policies is essential in the development of capable robotic agents. Generalizable models learn structures that transfer across domains. However, in practice, visuomotor policies test performance by interpolation on known distributions using unstructured domain shifts (e.g. lighting, clutter, diverse objects). We argue that to measure generalization capabilities we must instead test the inductive capacity of policies on progressively harder, out-of-distribution task variants. We call this inductive generalization, drawing directly on how axis-based evaluation has revealed inherent generalization limitations in language models (e.g. sequence length, counting) arXiv:2502.00197 . We provide a reusable and formal evaluation protocol for measuring inductive generalization in any manipulation policy, and establish baselines showing that existing paradigms fail this test; e.g. SoTA Vision-Language-Action models and find that policies that appear to generalize to prior domain shifts (distractors, etc) fail inductive generalization tests. These results expose a class of learning challenges orthogonal to those addressed by data and model scaling in robot learning, yet are imperative to solve in order to realize general purpose robots.

Protein-peptide interactions are central to cellular regulation and peptide-based drug discovery, yet existing computational methods mainly focus on interaction classification, binding-site prediction, or peptide binder generation. These formulations provide limited insight into the physicochemical mechanisms that determine how a peptide binds to a protein. In this work, we introduce PepLLM, an instruction-tuned framework for structured protein-peptide interface understanding. Given protein-peptide sequences, PepLLM generates a machine-readable JSON annotation describing multiple interface properties, including peptide burial state, hydrogen-bond density, salt-bridge presence, hotspot residues, hydrophobicity, and electrostatic complementarity. To support this task, we construct a new protein-peptide interface dataset by integrating structural interface analysis, solvent-accessible surface area computation, hydrophobic burial estimation, electrostatic potential calculation, and redundancy-aware data splitting. PepLLM connects a pretrained ESM encoder with a LLaMA decoder through a nonlinear modality adapter. The adapted ESM residue embeddings are injected into the LLaMA prompt as continuous soft tokens via placeholder-token replacement, enabling the decoder to generate structured interface annotations under instruction tuning. By moving beyond single-label prediction toward multi-property and mechanism-aware generation, PepLLM establishes a new task and modeling paradigm for interpretable protein-peptide interface analysis.

Molecular docking requires reasoning jointly about ligand pose and protein flexibility. Most diffusion-based docking models predict torsional updates with generic Euclidean heads that ignore the periodic geometry of angular variables. This mismatch is especially limiting in flexible docking, where ligand conformations and pocket side chains co-adapt to form the bound complex. Here, we introduce Harmony, a harmonic torsional diffusion framework for flexible protein-ligand docking. Harmony parameterizes ligand and side-chain torsional score fields as derivatives of learned harmonic potentials on the circle, which are noised mode by mode in the Fourier basis owing to the properties of variance-exploding diffusion on the torus. This construction makes periodicity explicit and gives the model a frequency-aware inductive bias over rotameric motion. On the PDBBind benchmark, Harmony improves ligand pose accuracy and pocket all-atom reconstruction over recent flexible docking methods. On PoseBusters, it improves the physical validity of generated complexes. Case studies on EBNA1 and KRAS G12D illustrate the method's behavior on a polar and a shallow binding site, respectively. Together, these results indicate that aligning the score parameterization with the geometry of the diffusion process is a simple and effective lever for improving flexible docking.

We investigate whether harmful chain-of-thought (CoT) traces from compromised language models can transfer unsafe behaviour and be distilled into reusable jailbreak attacks. Using an emergent-misalignment organism and a refusal-ablated jailbroken organism, we transplant harmful CoTs into $29$ open-source and $5$ closed-source targets. Transferred traces raise harmful-response rates above $80\%$ on the most vulnerable open-source models, while semantically mismatched CoTs fail entirely. LLooM concept mining identifies four recurring components of harmful reasoning: proceduralisation, ethical decoupling, evasion, and target–vulnerability framing. Distilling these patterns into reusable system prompts produces effective black-box jailbreaks, outperforming direct CoT transplantation on strongly aligned models by up to an order of magnitude, including a $10\times$ improvement on GPT-4.1 AdvBench. Reasoning-enabled models are more than twice as vulnerable, and output-side safeguards such as Llama-Guard~3 frequently miss harmful generations. Our results show that harmful reasoning transfers at both the trace and pattern levels, motivating defences that evaluate reasoning context in addition to final outputs.

Despite the proliferation of Explainable AI (XAI) techniques – from feature attributions to sparse autoencoders – explanations rarely influence real-world workflows. In practice, they are often generated and discarded without guiding meaningful action. This gap reflects foundational shortcomings: research has not yet established methodologies for integrating explanations into end-to-end, human-in-the-loop systems. This position paper argues that the machine learning community must pivot from ad-hoc XAI methods toward addressing foundational & structural challenges, including unclear problem formulations, underspecified evaluation objectives, and the absence of pipelines for explanation-driven feedback. We support this claim through an analysis of recent ICML, NeurIPS, and ICLR papers and a survey of XAI practitioners, revealing recurring issues that limit cumulative progress. We conclude by outlining a practical checklist designed to shift XAI toward a more human-centered, action-oriented paradigm. By emphasizing foundational clarity over the development of ad-hoc methods, we hope to provide a roadmap for integrating explanations into actionable, feedback-driven AI systems.

Stage B heart failure is characterized by asymptomatic structural or functional cardiac abnormalities. Identifying individuals at this stage is clinically important, as early detection may enable targeted interventions to prevent progression to symptomatic disease. Explainable artificial intelligence (XAI) may support early detection, transparent risk stratification, and selection of clinically actionable interventions. This review examines the use of XAI in detecting and characterizing stage B heart failure. A literature search of Web of Science, Scopus, and PubMed was conducted on 27 March 2026. Studies were included if they applied AI with XAI techniques to stage B heart failure. After screening, 20 studies were included. Data on modalities, outcomes, demographic reporting, and XAI methods were extracted and synthesized. SHAP was the most commonly used method, followed by LIME, saliency maps, and Grad-CAM; however, XAI adoption was inconsistent, with some studies relying on limited or ad hoc interpretability approaches. Notably, none compared explanations across sex or ethnic subgroups, despite evidence of subgroup differences in disease burden. Evaluation of XAI outputs was often insufficient: some studies did not assess explanations, while others relied only on literature-based comparisons, introducing potential bias. These limitations suggest explainability was not systematically validated or leveraged to support robust and fair clinical inference. XAI shows promise for improving transparency in stage B heart failure identification, but current implementations remain limited. Key gaps include limited consideration of sex and ethnicity, absence of subgroup-specific analyses, inconsistent evaluation, and lack of external validation, all of which constrain generalisability and clinical adoption.

The problem of predicting links in complex networks appears in different disciplines and has led to a variety of ingenious human-designed methods. We use this rich program space to explore the performance and behavior of automated code-evolution systems tasked to obtain machine-designed methods for link prediction. Despite being trained on limited data, algorithms evolved through code evolution outperform human-designed methods (with an average AUC score of 0.915 vs. 0.783, computed over 580 networks) and show improved computational efficiency, allowing them to be applied to networks with millions of links. The discovered methods follow approaches that have been employed in human-designed methods, but contain key innovations in the selection and combination of node- and link-features. This illustrates the role modern large language models and genetic algorithms can play in algorithmic innovation and scientific discovery more generally.

Understanding and predicting microstructure evolution is central to materials design, yet purely data-driven spatiotemporal learning models often suffer from limited physical consistency and degraded long-term prediction accuracy. In this work, we introduce a physics-guided fully convolutional spatiotemporal learning framework for microstructure evolution prediction. Unlike prior self-supervised approaches, the proposed method explicitly incorporates governing physical equations into the training objective, thereby encouraging the learned dynamics to remain consistent with known thermodynamic and kinetic laws. This physics-guided formulation improves predictive accuracy, long-horizon stability, and robustness across spatial resolutions and temporal prediction settings. Extensive experiments for spinodal decomposition demonstrate that incorporating physics-guided residual regularization leads to more faithful reproduction of microstructural morphology, statistics, and evolution trends compared with purely data-driven baselines. The proposed framework preserves the scalability and computational efficiency of fully convolutional architectures while bridging the gap between high-fidelity physics-based simulations and data-driven surrogate modeling, offering a reliable and efficient surrogate-modeling step toward digital-twin-enabled microstructure evolution prediction.

Radio-frequency (RF) monitoring is essential for space domain awareness, but it often generates large, variable, and sparsely populated datasets with few labels. These observations can capture satellites, space debris, and the ionospheric background, yet interpreting them typically requires specialized subject-matter expertise. Supervised deep learning methods can perform well on labeled RF data, but they require many annotated examples and may need careful retraining as RF conditions change. Semi-supervised approaches offer a practical alternative for limited-data settings by using unlabeled observations to reveal latent patterns that experts can interpret. In this paper, we present a semi-supervised RF detection and classification workflow for satellite monitoring that combines Non-negative Matrix Factorization with automatic model determination (NMFk), expert-guided cluster interpretation, and classifier-based prediction. We first represent RF observations as a non-negative feature matrix and apply NMFk to estimate the number of clusters that best captures patterns in the unlabeled data. Subject-matter experts then assign physical meaning to the resulting clusters, including satellite detections, ionospheric environmental conditions, and other RF event categories. Finally, we train a classifier on these interpreted clusters to evaluate performance on a test set and categorize future observations. This pipeline reduces reliance on large pre-labeled datasets by pairing unsupervised factorization with expert interpretation, enabling an interpretable and transferable methodology for detecting, observing, and classifying behavior in RF data.

Generalization is crucial for modern control, but learning across tasks can be costly. Recent work shows that zero-shot transfer, in which a controller designed for a source task is directly deployed to a target task, can achieve strong performance. We formalize task-space complexity, which characterizes the difficulty of achieving sufficiently good performance across a collection of tasks, by casting source selection as a covering problem. We provide bounds on the complexity and demonstrate its applicability.

Continual adaptation is essential for multimodal large language models (MLLMs) deployed across evolving domains, but the state-of-the-art MR-LoRA method highly relies on the assumption that a MLLM-based router is necessary to process complex multimodal inputs. This paper revisits this claim on the MLLM-CL benchmark and argues for two claims. First, routing does not require an MLLM: a simple training-free, replay-free ptotypical routing method (RePRo), uses frozen pretrained features and task prototypes to match the MLLM-based router of MR-LoRA at far lower computational cost. Second, shared experts do not improve continual learning for MLLMs, despite their theoretical appeal. We show that these findings arise from two structural limitations of MLLM-CL: (1) its tasks are highly separable in representation space, and (2) its fixed task order makes conclusions sensitive to a single curriculum rather than robust across diverse continual-learning trajectories. As a result, the benchmark primarily rewards learning in isolation rather than genuine continual transfer. This motivates a new design for future benchmarks of continual MLLM learning, with overlapping task manifolds, multiple task orders, fine-grained domain shifts, and evaluation protocols that reward forward transfer as well as retention.

Nash equilibrium (NE) arises from selfish utility maximization, yet its social welfare can be arbitrarily far from optimal. Moreover, computing an NE is intractable in general. We study augmented game models in which players use budget-balanced internal transfers to improve incentives before play. We first introduce Self-Enforcing Transfer Equilibrium (SETE), where players commit to nonnegative peer-to-peer transfers that are paid only if the recipient does not deviate from a prescribed strategy. For polymatrix games, we show that every stationary point of the social welfare function, in particular any socially optimal strategy profile, can be sustained as a SETE. This induces a Nash equilibrium in the agent normal form of the corresponding augmented game. We further propose a polynomial-time algorithm and a decentralized learning dynamic to compute such product-form equilibria. We then introduce Mediated Self-Enforcing Transfer Equilibrium (M-SETE), where a mediator makes both the payment schedule and the prescribed strategies binding offers. This additional enforcement resolves the agent-normal-form limitation: an M-SETE is a Nash equilibrium of the augmented game itself, not merely of its agent normal form, and any socially optimal strategy profile can be supported as an M-SETE in any finite game while preserving budget balance. Thus, internal transfers improve welfare and computation while preserving independent play on the equilibrium path. When full sequential-game stability is required, binding mediation provides the corresponding implementation.

Language models may encode both outdated facts and their newer replacements. We introduce Parametric Temporal Conflict (PTC), where the newer fact is present and recoverable, but the default forward pass prefers the outdated one. We release a deterministically verified benchmark of 8,746 Wikidata position-holder transitions and evaluate four open-weight language models across three families. A date-prefix prompt recovers the newer fact in 61-81% of PTC cases. Activation patching flips predictions in 72-85% of cases and localizes the preference to model-specific upper-layer regions. Residual-stream steering outperforms norm-matched random directions, indicating direction-specific representations. These results show that PTC reflects a localized representational preference rather than missing knowledge. Recovery is measured on oracle-identified conflicts because automatic detection remains unreliable. We release the benchmark, code, and statistics.

Dynamic average estimation is a critical problem in multi-agent systems, enabling agents to collaboratively estimate time-varying signals using only local information exchange. Traditional model-based approaches often face challenges related to convergence speed and sensitivity to network topology changes. This paper introduces a novel learning-based solution leveraging Gated Graph Neural Networks (GGNNs) for fast-convergent dynamic average estimation in a fully distributed manner. Taking advantage of the inherent structure of GGNNs, the proposed method models the estimation process as a distributed autoregressor, ensuring rapid convergence while maintaining stability. We incorporate a regularization term during training to enforce convergence guarantees and introduce an encoding-decoding mechanism to reduce communication overhead without sacrificing accuracy compared to standard GGNNs. Extensive numerical experiments demonstrate that our approach significantly outperforms conventional model-based estimators in terms of both convergence speed and precision, making it a promising alternative for multi-agent applications that require dynamic average estimation.

Self-attention is central to Transformer performance and is often the most expensive part of the Transformer at long context lengths because its pairwise token interactions scale quadratically with sequence length. Standard dense attention also applies the same set of attention heads to every token regardless of token difficulty or information content. This uniform activation can waste compute, especially as sequences grow longer and attention cost increases rapidly. We propose Grouped Query Experts (GQE), a mixture-of-experts layer on top of grouped-query attention (GQA). Within each GQA group, a router selects k query-head experts per token while all key-value (KV) heads remain dense and unchanged. Thus, GQE keeps the KV cache benefits of GQA and reduces only the active query-head computation. On a fixed 30B token budget at the 250M parameter scale, GQE matches the all-active GQA baseline in downstream accuracy while activating half the query heads per token.

We study internalization processes, by which neural-network-based systems absorb an explicit computational procedure into their own weights, and how they facilitate learning. We investigate how transformers internalize the simulation of semiautomata by internalizing chain-of-thought (CoT) tokens, which classes of semiautomata are harder to internalize, and expose the flip side of internalization, that is, a progressive degradation of out-of-distribution performance. We then provide the first provable analysis of successful internalization: for the task of learning parities, we show that a simplified one-layer transformer provably first learns the target with explicit CoT supervision and then internalizes the autoregressive generation as CoT tokens are progressively removed, learning to directly compute the parity. This task is computationally hard to learn from data without CoT supervision. Finally, we discuss how learning through internalization relates to the Positive Distribution Shift phenomenon recently introduced by~\citet{Med+26}.

The choice of loss function and optimizer is an important decision, that shapes further model training. Yet automated architecture search pipelines (AutoML) benefits significantly more from the optimal pairing selection and vice versa. This paper investigates whether a single recipe is sufficient for heterogeneous architecture pools, or whether the optimal pairing varies across structurally diverse models. We conduct a systematic empirical study of all $3 \times 6 = 18$ combinations of six optimizers (SGD+Momentum, Adam, AdamW, RMSprop, Adagrad, Adadelta), paired with three loss functions: Cross-Entropy (CEL), Negative Log-Likelihood (NLL), and the recently introduced genetically evolved NGL loss across the base models presented in LEMUR heterogeneous architecture pool on six image classification datasets (CelebA-Gender, CIFAR-10, CIFAR-100, ImageNette, MNIST, SVHN). The 18 loss-optimizer configurations are applied to each of the 33 compatible base architectures taken from the LEMUR pool, resulting in 594 variants that were generated fully automatically by a source-level injection pipeline and evaluated under fixed hyperparameters, ensuring that observed accuracy differences are attributable solely to the loss-optimizer pairing. Our results confirm that no single pairing is universally optimal. Cross-Entropy with Adam or AdamW is the most robust choice across architecture families and datasets. NGL is a competitive alternative to CEL on standard convolutional classifiers, but only when paired with adaptive optimizers; it degrades substantially with SGD or accumulation-based methods. Adagrad and Adadelta consistently underperform under fixed hyperparameters regardless of loss function, highlighting their sensitivity to learning rate tuning. These findings provide actionable guidance for loss-optimizer selection within NNGPT Framework.

Sheaf Neural Networks (SNNs) generalize Graph Neural Networks (GNNs) by replacing scalar node signals with stalk-valued signals and by using restriction maps to measure compatibility across edges. Unlike standard graph diffusion, which encourages neighboring node features to become similar, sheaf diffusion promotes consistency through the restriction maps and can therefore model more general relationships between neighboring nodes. However, existing sheaf neural architectures mainly operate at a fixed graph resolution and do not provide a principled pooling mechanism for building hierarchical representations. In this paper, we introduce Hierarchical Sheaf Pool (HiSP), a sheaf-aware pooling framework based on local spectral coarsening. Given a partition of the graph, HiSP constructs each coarse stalk by projecting fine stalk-valued features onto the low-frequency eigenmodes of the cluster-internal sheaf Laplacian. These local modes define a cochain-level prolongation map, which allows the fine sheaf energy to be represented on the coarse space through a Galerkin operator. We further analyze the approximation induced by coarsening by separating truncation loss, due to discarded local modes, from realization loss, due to representing the projected operator as a coarse sheaf. Finally, we implement HiSP as a GNN pooling layer compatible with SNNs and provide a PyG implementation supporting batching, lifted sheaf Laplacians, and hierarchical architectures.

Characterizing non-Gaussian posterior distributions in partially observed high-dimensional nonlinear systems remains a fundamental challenge in data assimilation. Ensemble Kalman filters rely on Gaussian approximations that can be inaccurate for strongly non-Gaussian posteriors, whereas particle filters suffer from severe scalability limitations. Recent score-based generative approaches improve posterior characterization but typically require supervised training with ground-truth posterior samples, which are unavailable in most practical applications. We introduce $\Omega$ (Operator-based Mixture Ensemble for Generative Assimilation), a scalable framework that integrates conditional Gaussian surrogate modeling, unsupervised score learning, and generative sampling. The conditional Gaussian surrogate provides a nonlinear non-Gaussian baseline approximation while admitting closed-form conditional posterior distributions for the unresolved variables. First, $\Omega$ exploits these closed-form conditional distributions to analytically recover the high-dimensional unobserved component, reducing computational cost and mitigating the curse of dimensionality. Second, $\Omega$ learns only the residual discrepancy beyond an analytical baseline through denoising score matching using ensemble trajectories alone, eliminating the need for ground-truth posterior samples and substantially reducing the learning burden. Third, $\Omega$ reconstructs the full non-Gaussian posterior distribution of both observed and unobserved variables via a Gaussian mixture representation, capturing multimodal, skewed, and heavy-tailed statistics. Finally, $\Omega$ employs annealed Langevin sampling to iteratively refine ensemble members from the baseline toward the target posterior. $\Omega$ is validated on several turbulent models with intermittency and extreme events, consistently improving posterior accuracy.

Accurate short-term electricity demand forecasting is critical for reliable power system operation, energy market planning, and infrastructure optimization. This paper benchmarks ten machine learning models for daily electricity demand forecasting across the New England ISO (February 2020 - March 2023). The models span four families: tabular gradient-boosted trees (Random Forest, LightGBM, CatBoost, XGBoost), standalone neural architectures (LSTM, Transformer encoder), and hybrid Transformer+tree variants (Hybrid XGBoost, Hybrid LightGBM, Hybrid CatBoost, Hybrid RF). All models use meteorological data from six cities, calendar and holiday effects, autoregressive demand lags, and COVID-19 epidemiological variables. Hyperparameter optimization uses Optuna (300 trials, multivariate TPE, seed=42) under a leakage-free 70/15/15 chronological split. CatBoost achieves the best test performance: RMSE 8316 MWh, MAPE 1.87%, R-squared 0.917, followed by XGBoost (9066 MWh, R-squared 0.901), Hybrid CatBoost (9068 MWh, R-squared 0.901), and Hybrid XGBoost (9208 MWh, R-squared 0.898). Standalone neural architectures perform substantially worse (Transformer: 21294 MWh; LSTM: 22808 MWh), confirming the Transformer's role as a feature extractor rather than an end-to-end forecaster. An ablation on CatBoost shows that demand lags are the dominant predictor: removal degrades RMSE from 8316 to 11310 MWh (+36%), while weather and calendar features alone achieve an R-squared of 0.864. Removing COVID-19 features improves test RMSE by 1.7% while reducing training RMSE by 17.3%, a signature of temporal validity decay. SHAP analysis confirms this: 3 of 8 COVID features rank higher on the post-acute test set than during pandemic-active training, indicating the model over-applies stale pandemic patterns after behavioral adaptation was complete by August 2022.

Iterative design and optimization of large, complex structures require fast and accurate prediction of stress, displacement, and other fields. Finite element analysis (FEA) is computationally expensive for this task. Existing neural network surrogates often struggle with varying topologies and complex boundary conditions. This study proposes the novel Dual-Stream Heterogeneous Graph Neural Network (DS-HGNN) for full-field stress and displacement prediction in thin-walled structures, demonstrated on box beams made of stiffened panels. DS-HGNN operates on panel-level heterogeneous graph representations and introduces physics-guided edge states initialized from edge types, spatial information, and boundary kinematics. These states are updated through dual-stream message passing that separates longitudinal and transverse structural information while allowing cross-stream exchange. Geometry and loading effects are incorporated through Feature-wise Linear Modulation (FiLM)-conditioned 1-D spectral convolutions, and physical fields are reconstructed using a spectral-bypass low-rank readout. The model is evaluated on stiffened panel datasets with different geometries, boundary kinematics, loading conditions, and material nonlinear responses. DS-HGNN achieves the lowest stress and displacement RMSE compared with six benchmark heterogeneous graph neural network models. It also reaches comparable accuracy to the strongest benchmark models using 19%-38% fewer training samples. A targeted evaluation further shows that DS-HGNN captures yield and post-yield stress features.

Medical vision-language models (VLMs) enable zero-shot clinical image classification, yet reliably detecting out-of-distribution (OOD) inputs at deployment remains an open problem. No static scoring method works across all shift types: Maximum Concept Matching (MCM) on FLAIR achieves 76.4% AUROC for far-OOD but only 42.4% for covariate shifts such as ultra-wide-field fundus images, effectively random. We trace this to a structural mismatch: covariate-shifted inputs are indistinguishable from in-distribution samples in softmax space, yet occupy distinct regions in the VLM embedding space. To exploit this untapped signal, we propose PROTON (PROtotype-based Test-time ONline OOD detection), a lightweight post-hoc module that maintains an online prototype bank from high-confidence test predictions and adaptively fuses prototype distance with MCM scoring via stream-level variance statistics, requiring no model modification, training data, or prompt engineering. On the ophthalmology benchmark FLAIR + FIVES, PROTON improves MCM by +23.9 AUROC on covariate shift, +8.8 on semantic shift, and +8.1 on far-OOD, making it the only zero-shot method to improve all three without hierarchical prompts or labeled data. Code is available at https://github.com/GenMI-Lab/PROTON, and the project page is available at https://genmi-lab.github.io/PROTON.

Molecular message-passing neural networks commonly propagate chemically diverse interactions through a single graph, which may mix interaction-specific signals and require deep propagation to capture long-range effects. We introduce the Multi-level, Multi-color Graph Neural Network (MMGNN), a hierarchical framework that decomposes a molecular graph into overlapping atom-type-pair-specific subgraphs while preserving atom-level resolution. MMGNN-2D constructs chemical-colored subgraphs from covalent connectivity, whereas MMGNN-3D constructs geometric-colored subgraphs from spatial proximity and augments their edges with distance, angular, and torsional descriptors. Both variants apply a shared communicative message-passing backbone to each subgraph and combine the resulting representations through atom-wise aggregation and molecular readout. We evaluated MMGNN on five classification and three regression benchmarks from MoleculeNet using common scaffold splits and five independent runs. MMGNN-2D achieved the highest macro-average AUC-ROC of 0.838 across the classification datasets and the lowest RMSE on ESOL (0.803). MMGNN-3D obtained the highest mean AUC-ROC on BBBP (0.956) and the lowest RMSE on FreeSolv (1.793), indicating complementary strengths of topological and geometric representations. Structural and leave-one-out analyses further illustrate how the subgraph decomposition affects learned representations and atom-type-pair sensitivities. These results support overlapping interaction-specific graph decomposition as a competitive strategy for molecular property prediction.

Filmmaking demands precise motion control and reference image compositing – capabilities that existing methods treat separately. Point-track-conditioned image-to-video models restrict content insertion to the first frame, while reference-to-video models lack fine-grained spatial-temporal control over how reference content integrates across frames. We present Go-with-the-Track, which unifies both capabilities by jointly conditioning on multiple reference images and reference-anchored point-tracks – extending conventional point-tracks to explicitly establish correspondences between generated frames and reference images, thus enabling precise compositing and motion control throughout the video. To achieve this, we introduce spatially-aware point-track embeddings that encode the full sequence of point-track coordinates using a coordinate-wise MLP followed by temporal pooling. This representation captures the spatial characteristics of each point-track (serving as a unique identifier), while the embedding similarity correlates directly with spatial proximity, enhancing the model's ability to distinguish and associate point-tracks. We inject these point-track embeddings into a video diffusion transformer via a lightweight adapter, resolving the pixel-to-patch resolution mismatch while avoiding the substantial motion detail loss inherent in naive point-track subsampling. We use a hybrid training strategy to train jointly on dynamic, static, and synthetic scene video datasets to boost motion controllability. Experiments demonstrate that Go-with-the-Track achieves superior motion and reference control in a single model and enables new capabilities: multi-reference conditioned video generation with point-track driven compositing, as well as camera control for both static and dynamic scenes. Project Page: https://eyeline-labs.github.io/Go-with-the-Track/

Estimating causal effects in industrial time series requires handling temporal dynamics, time-varying treatments, and unobserved confounders. Existing causal foundation models (CausalPFN, CausalFM) operate only on static cross-sectional data; neural temporal methods (CRN, G-Net) require per-dataset training; and concurrent temporal-PFN proposals have not been demonstrated at industrial scale. None output explicit per-pair reliability signals alongside their CATE estimates. We introduce Temporal Causal Prior-Data Fitted Networks (TCPFN), a foundation model for zero-shot temporal causal discovery with learned reliability signals. TCPFN makes four contributions: (1) a Causal Judgment Head that jointly predicts null-effect probability, confounding strength, identifiability, mediation fraction, and causal regime; (2) a mixed training prior covering six causal regimes (independent, direct, confounded, mediated, time-varying confounded, feedback) plus CausalFM-style front-door and instrumental-variable priors; (3) a discrete-token panel-data architecture with cross-attention masking that prevents inter-horizon leakage; (4) zero-shot inference at industrial scale via FAISS-based context selection and one-step posterior correction. On 19 benchmark datasets across five domains, TCPFN achieves competitive zero-shot causal discovery: AUROC 0.96 on Tennessee Eastman, 0.93 on SWaT, 0.98 on Causal Rivers, 0.97 on CAUSRCA. The null detector reaches NullF1 0.94, AUROC 0.99. TCPFN scales to V=1,275 on a proprietary Kraft pulp-and-paper dataset in 6 hours on a single GPU; PCMCI, a CPU-only library, on a V=666 sub-panel of the same data took 81.5 hours, extrapolating by O(V^2) to ~12.5 days at V=1,275. TCPFN's top edges identify cross-subsystem causal relationships while PCMCI's surface within-instrument controller-measurement coupling – a scalability case study.

Decision-making under partial or adversarial observability requires accurate inference of the environment's latent state and its associated uncertainty. This work analyses adversarial attacks on linear state-space models, where the attacker alters observations subject to likelihood constraints that ensure that the perturbations remain statistically consistent with the observation model. We analyse how such adversarial yet plausible observations shift inference about latent states and affect downstream decision-making and the performance of reinforcement learning agents. In addition, we introduce an online Bayesian defence based on directional covariance adaptation, which selectively reduces the influence of observations by comparing their estimated impact with that of the computed most disruptive direction, while preserving information in the remaining orthogonal observation subspace. The proposed framework provides a principled approach to constructing robust inference and decision-making systems, with direct relevance to safety-critical applications such as robotics, where reliable operation under sensor noise, partial failures, and adversarial conditions is essential.