Climate change research increasingly requires AI systems that reason across text, dynamic visual content, and scientific figures, yet existing climate QA benchmarks are small, mostly textual, and cover a narrow range of models. We introduce MMClima, a large-scale multimodal climate question answering framework with 104k+ expert-validated question-answer pairs spanning articles, video transcriptions, and figures across five core climate science domains. MMClima is constructed via automated claim extraction and QA synthesis with human-in-the-loop validation to ensure both scale and reliability. Using MMClima, we benchmark state-of-the-art multimodal language models on tasks requiring factual recall, visual interpretation, and cross-modal synthesis. We additionally fine-tune on the textual split to produce mmclima-70b-txt, a domain-adapted baseline that outperforms strong open- and closed-source models on textual QA. We release the dataset, evaluation pipeline, fine-tuned model weights, and data creation framework to support standardized multimodal evaluation for climate science.

Group Relative Policy Optimization (GRPO) relies on the diversity of $K$ rollouts within each group; otherwise, the group-mean advantage $A^{(k)} = r^{(k)} - μ_r$ collapses to zero. This presents a structural challenge for latent-reasoning models like Coconut, which feed continuous hidden states recurrently in place of discrete chain-of-thought tokens. Because the latent phase is inherently deterministic given the parameters and prompt, multiple rollouts produce identical trajectories, stalling GRPO's progress. Consequently, applying group-relative reinforcement learning to continuous latent reasoning has proven difficult. To address this, we propose sourcing the necessary stochasticity through structured dropout. By applying a single Bernoulli mask held constant across all latent recurrence steps for a given rollout, we generate essential trajectory variance. This shared mask effectively treats each rollout as a posterior sample from a variational distribution over parameters, allowing GRPO to optimize the expected reward of a Bayesian model-average policy. We provide both theoretical justification for this method -- including unbiasedness, variance reduction, and the well-definedness of the latent gradient -- and empirical validation. On GSM8K, dropout-GRPO improves a Coconut baseline from $27.29\%$ to $29.01\%$ pass@1, demonstrating the viability of GRPO learning for latent-reasoning models. Our work positions this as a practical, theoretically grounded approach for post-training latent-reasoning LLMs.

Modern Diffusion Transformers for video generation provide limited control over the progression of time and the editing of temporal dynamics. We propose a temporal-control methodology that extends a pretrained DiT with explicit time editing, allowing control over motion speed and temporal structure without redesigning the backbone. Its core implementation augments the pretrained model with a lightweight temporal module, preserving the original generative prior while expanding its controllable dynamic range.

We introduce flow control of vision-language-action (VLA) models, a simple and effective way to steer VLA actions in real-time through generic inputs, such as a keyboard. This method can be used out-of-the-box and does not require retraining or fine-tuning VLAs. It enables relatively crude user inputs to steer a VLA to align with user intent. The VLA transforms these inputs into action samples drawn from the VLA expert action distribution learned during training, so that the generated actions are high quality (conformity to the action expert distribution) and high fidelity (reflecting the user's intent). We demonstrate that flow control has many desirable properties: (1) flow control accurately and responsively steers robot actions with user inputs, (2) it is robust to suboptimal user inputs, (3) it enables users to steer VLAs to achieve significantly higher success rates and faster task completion, and (4) fine-tuning a VLA on flow control trajectories improves the autonomous policy. Together, these results provide a simple and intuitive way for users to help steer VLA actions, increasing task performance.

Wildfire detection and monitoring are critical for mitigating fire spread and reducing environmental and infrastructural damage. In this work, we introduce GWFP (Global Wildfire Prevention Dataset), a large-scale, open-source dataset of wildfire images and videos designed to support early fire and smoke detection research. GWFP contains geographically diverse wildfire scenes, including flames, smoke, Waterdog/Fog environmental conditions, Near Infrared (NIR) imagery, Ember, and challenging negative samples collected from real-world scenarios worldwide. To evaluate dataset robustness and cross-domain generalization, we benchmark multiple convolutional and transformer-based architectures across both in-domain and cross-dataset settings. Additionally, we explore lightweight frequency--spatial feature interaction using Hadamard-enhanced residual connections (HTE-ResNet) to analyze representation robustness under domain-shift conditions. Experimental results demonstrate strong cross-dataset generalization and practical utility for real-world wildfire monitoring applications. The dataset and source code will be publicly released upon acceptance.

This paper extends the concept of Learning Entropy (LE) from temporal adaptive systems to spatial learning in multilayer perceptron networks (MLPs) applied to image data. Instead of evaluating image structure directly from gradients or covariance operators, as local neighborhood methods do, the proposed approach analyzes the learning process itself through Learning Entropy. An MLP is trained to predict the intensity of a center pixel from its surrounding spatial context, while LE is evaluated from the incremental adaptation of neural weights during learning across image-derived samples. The resulting Spatial Learning Entropy Maps (SLEM) identify unusual image points and regions that induce strong adaptation of the neural network and therefore have an important role in the learning process. The results indicate that spatial Learning Entropy provides a complementary perspective to conventional feature extraction and explainability methods by highlighting spatial locations that are particularly informative for network learning. Spatial Learning Entropy provides a complementary perspective to conventional feature extraction and explainability methods by identifying image points and regions according to their learning impact rather than their local structural properties. The proposed framework may open new directions for learning-driven image or scene analysis in computer vision, manufacturing, and robotics.

Recent learning-based path planners use neural networks to process visual map representations and approximate heuristics for classical search algorithms, yielding near-optimal paths with reduced search effort. However, these methods are tied to the shortest-path objective implicit in their supervision, which limits their flexibility to accommodate alternative criteria. We introduce FlexPath, a two-stage framework that decouples feasibility from preference. In Stage 1, we use imitation learning to acquire a task-independent spatial prior over feasible paths from visual map inputs. In Stage 2, differentiable Path Shape Objectives (PSOs) adapt this prior toward task-specific criteria without relearning path structure, requiring only efficient objective-level adaptation. A single pretrained model can be adapted to multiple objectives. For shortest-path planning, FlexPath reduces search effort on TMP by 14.3% compared to the state-of-the-art TransPath, while also finding lower-cost paths on average and demonstrating strong zero-shot generalization across three unseen domains. For obstacle clearance with minimum clearance distance 2, it achieves 96.8% full obstacle avoidance while maintaining low search cost. The framework further extends to semantic-aware avoidance and waypoint guidance via objective-level adaptation, and remains compatible with classical planners at inference time. Data and code are available at https://github.com/FraunhoferIVI/FlexPath.

Current cross-view localization methods predominantly rely on satellite imagery as the aerial modality. Although recent work explores planimetric maps (e.g., OpenStreetMap tiles), these approaches often lag in performance. Yet both modalities are widely available and possess complementary properties. Satellite images are closer to ground-level camera imagery, offering finer detail, whereas planimetric maps contain annotated objects (e.g., streetlamps) and remain informative in areas where the ground is occluded, such as by foliage. Despite this, only one prior work provides an end-to-end method to fuse the two modalities, and it does not demonstrate their potential within state-of-the-art methods. To combine the strengths of both modalities, we propose a new fusion module that augments standard encoders and demonstrates that integrating satellite imagery with planimetric maps improves state-of-the-art single-modality methods. The module comprises (i) cross-modal conditioning, which processes each modality's encoding with awareness of the other, and (ii) a patch-level fusion rule that controls the granularity of information exchange. We achieve state-of-the-art results, reducing the mean localization error by 30.13\%. Qualitatively, the fusion adaptively selects the more informative modality, improving overall accuracy.

AI is increasingly used to support scientific peer review, from manuscript screening, reviewer assistance to editorial triage. Although such systems promise to reduce reviewer burden and accelerate publication, their robustness to strategic manipulation remains poorly understood. Here we show that AI-mediated peer review is vulnerable to a simple, low-cost manipulation: superficial rephrasing of the manuscript abstract. Without changing the underlying scientific content and communication, and even without knowledge of the reviewing model, adversarially rewritten abstracts substantially improve AI review outcomes. We see this across disciplines and publication venues, for both human-written and AI-generated papers. Our strongest attack achieves an attack-success-rate of about 38%, increasing acceptance ratings by +1.31 for Gemini 3 Flash reviewers and by +0.88 for GPT 5.4 Mini reviewers on a 10-point scale. When the original AI review suggests 'reject', the success rate rises to more than 50%. This effect extends beyond overall score inflation, increasing review confidence and scores on core scientific criteria such as soundness, significance and perceived contribution. The attack is practical, requiring only about 5 minutes and $1 for a 10-page AI conference submission, and is hard to distinguish from ordinary scientific editing. Inflated AI reviews could bias downstream human decision-making, shifting editorial recommendations from rejection towards acceptance. These findings reveal a general vulnerability in AI-assisted scientific evaluation: when AI-generated review influence editorial decisions, authors may be incentivized to optimize manuscripts for AI judgment rather than scientific merit. Our results suggest that AI tools should not be treated as neutral evaluators in high-stakes peer review without systematic robustness testing, transparent safeguards and careful human oversight.

Quantized checkpoints are often screened first with quality metrics and only later, if at all, with direct safety tests. This paper audits that shortcut on a matched 51-row matrix spanning 6 models, 4 families, a 7-level GGUF ladder, and AWQ/GPTQ INT4 checkpoints. In this matrix the shortcut fails: all 36 quality-safety pairings split direction across models, and 9 hidden-danger rows plus 1 near-hidden-danger row show quality stable or improved while refusal falls by 12-68 percentage points. Seven of the 11 AWQ/GPTQ rows are hidden-danger. A four-probe mechanistic follow-up over the 17 Hugging Face-backed FP16/AWQ/GPTQ cells does not rescue it: entropy, refusal-direction, and calibration probes are weak or null separators of dangerous rows, and although probe-identified safety-associated neurons absorb 1.39$\times$ more quantization error overall ($p < 5 \times 10^{-7}$), the effect is not regime-specific. Claude Sonnet 4 relabels 11,470 items in a predefined stratified set, agrees with the primary gemma3:12b judge on 89.9\% of rows ($κ= 0.873$, 95\% CI [0.866, 0.881]), and changes 0/10 hidden-danger cells. A calibrated study-internal behavioral screen -- the Refusal Template Stability Index (RTSI), built from four refusal-template drift features and calibrated on this matrix -- routes 10/10 hidden- or near-hidden-danger rows to direct safety testing (Wilson 95\% CI lower bound 0.72) while leaving 23 of 45 non-baseline rows in a low-risk bucket under both in-sample scoring and row-level leave-one-out validation; on the same matrix, the best single-feature baselines (unique-prefix-rate-delta, raw refusal-rate delta) recover 9/10 and 8/10 respectively at matched bucket size, and cross-stack transfer requires recalibration. For the quantized checkpoints, model families, and safety outcomes studied here, retained quality cannot waive direct safety evaluation.

Learning the compositional nature of the physical world requires joint observation of interacting factors. However, because practical data is often decentralized, these factors are fragmented across isolated silos. Existing decentralized generative approaches focus only on modeling the union of siloed data, overlooking novel combinations implied by the collective whole. To bridge this gap, we introduce Decentralized Compositional Flow Matching (DCFM), a framework that enforces structural constraints across the global set of generative factors, without exchanging any raw data. DCFM enables novel combinations to emerge through peer interactions, even when no single data source can independently support the composition. Empirically, DCFM substantially outperforms federated learning and mixture-of-experts baselines across conditional image generation, robotic spatial planning, and medical attribute co-occurrence modeling.

Multimodal Large Language Models (MLLMs) can listen and see, but how do audio and visual signals actually travel through the network to shape an answer? Despite their growing role in research and real-world applications, the internal pathways through which audio and visual tokens influence the final prediction remain poorly understood. In this study, we examine audio-visual information flow inside Audio-Visual Large Language Models (AVLLMs), tracing how AVLLMs route, utilize, and integrate audio and visual information across two input configurations, audio-visual video and multiple interleaved audio-visual items. We find that for audio-visual video, AVLLMs follow the sequential information flow pathway established for VLMs and VideoLLMs, with audio and visual contribution flowing along this pathway in proportion to the task's reliance on each modality. In settings with multiple interleaved audio-visual items, this routing shifts to different parallel streams. Furthermore, we demonstrate that audio-visual and other token types can be discarded once their information is transferred to LLM, with minimal impact on the model's prediction or even slight improvement, generalizing across multiple tasks and datasets, enabling more efficient inference. These findings hold across multiple models and scales, Qwen2.5-Omni and Video-SALMONN2 Plus at 3B and 7B scales, leading to hypotheses on why these flow structures emerge. Together, these results deliver the first coherent picture of how AVLLMs orchestrate sound and sight inside the network and lay the groundwork for the next wave of interpretability, design, and efficiency advances in audio-visual and broader MLLMs.

Recent advances in 3D generation have led to substantial improvements in realism, controllability, and efficiency, yet the evaluation of 3D assets remains underexplored. Existing evaluation paradigms, including human evaluation, learned metrics, and vision-language models (VLMs) as judges, suffer from limitations in cost, scalability, resolution handling, or task-specific alignment. In this work, we focus on 3D mesh evaluation and introduce DB-3DME, the Dataset and Benchmark for 3D Mesh Evaluation. DB-3DME contains 2,619 synthetic 3D meshes paired with human ratings on Geometry and Prompt Adherence. Using this dataset, we systematically benchmark state-of-the-art VLMs and identify visual encoding of 3D representations as a key factor for human-aligned evaluation performance. Motivated by this finding, we fine-tune an open-weight VLM, Qwen-2.5-VL-7B, for 3D mesh evaluation by adapting the visual encoder while freezing the language model. The fine-tuned model substantially outperforms existing pre-trained VLMs across multiple evaluation dimensions, establishing a new benchmark for automatic 3D mesh evaluation. We publicly release the benchmark dataset on GitHub and Hugging Face to facilitate future research.

A common assumption in strategic classification is that the classifier is public knowledge. However, it remains unclear whether, and why, a system would choose to commit to full disclosure. We study a setting in which regulation requires the system to disclose some, but not all, of the information. This induces a learning task in which the learner must jointly optimize the classifier and the uncertainty surrounding it. To this end, we adopt from robust mechanism design the notion of ambiguity, which in our setting allows the learner to reveal a set or range of possible classifiers, while privately choosing which of them to ultimately realize. We investigate how ambiguity affects the learning task, develop efficient algorithms for computing best-responses and training, and empirically explore strategic learning and its outcomes in this novel setting and using our approach.

While deep Reinforcement Learning (deep-RL) has been increasingly applied to parameter control in evolutionary algorithms, rigorous theoretical analysis of parameter control remains largely restricted to single-parameter settings, owing to the difficulty of deriving effective, interpretable multi-parameter policies amenable to formal study. We demonstrate how deep-RL can be leveraged to overcome this barrier, using the (1+($λ$,$λ$))-genetic algorithm optimizing OneMax, one of the few problems where a super-constant speedup of dynamic control has been formally proven, as a representative case study. We first show that standard approaches struggle to converge in this multi-parameter setting, and introduce algorithm-agnostic enhancements targeting action-space decomposition, reward shifting, and long-horizon discounting. With these in place, we compare common deep-RL methods and find that Double Deep Q-Networks uniquely avoid the policy collapse observed in Proximal Policy Optimization, yielding trajectories suitable for downstream analysis. Crucially, we move beyond the ``black-box'' nature of neural networks by distilling the learned behaviors into a transparent, symbolic control policy. This resulting policy does not only offer interpretability for future theoretical analysis but also yields exceptional performance, consistently outperforming existing baselines across a wide range of problem sizes.

Federated learning (FL) is often subject to aggregation variance if clients do not consistently participate in training rounds. While reusing stale model updates from inactive clients is a common technique to reduce this variance, we find that with skewed client participation, the resulting update staleness can become severe enough to destabilize training. To remedy this, we propose FedSteer, a novel method that constructs a gradient subspace from a cache of recent client gradients to serve as a low-dimensional representation of the current optimization landscape. FedSteer projects an active client's true gradient onto this subspace to find a set of optimal coordinates. For an inactive client, FedSteer reuses these coordinates with the now-evolved subspace drifted by other active clients. This process effectively "steers" outdated gradients toward the current global objective. This is complemented by a selective caching strategy that identifies a representative client subset to form the subspace, reducing server memory. Experiments demonstrate that FedSteer significantly outperforms baselines, preventing performance collapse in challenging scenarios while delivering accuracy gains of over 7% in others.

Accurate lesion segmentation from whole-body Positron Emission Tomography (PET)/Computed Tomography (CT) scans is essential for cancer staging and treatment planning. PET provides functional metabolic information with different radiotracers, while CT offers anatomical localization. Lesion delineation from PET/CT imaging is clinically challenging due to subtle imaging features, confounders, and inter-reader variability. Existing deep learning approaches suffer from training-related stochasticity, inconsistent predictions, missed lesions in high tumor-burden cases, and lack uncertainty quantification, limiting their clinical reliability. Using nnU-Net as a baseline, we propose an uncertainty-aware framework for whole-body PET/CT lesion segmentation that integrates (1) Bayesian ensembling to reduce training stochasticity, (2) voxel-wise uncertainty quantification with epistemic and aleatoric decomposition, and (3) epistemic uncertainty-augmented training to improve lesion detection. Two public datasets, AutoPET-III (1,611 scans) and Deep-PSMA (200 scans), comprising FDG and PSMA studies across multiple cancer types, are used for training and evaluation. Bayesian ensembling improves robustness and performance over deterministic nnU-Net models on the unseen AutoPET-III test set. Uncertainty maps highlight regions of model disagreement and correlate with misclassifications, particularly false positives. Uncertainty-augmented training improves lesion recovery at the cost of increased FPVol, reflecting a precision-recall trade-off. A case-adaptive routing strategy further improves Dice by selecting between the base and augmented models. To our knowledge, this is the first study to systematically investigate uncertainty quantification in multi-tracer, pan-cancer PET/CT segmentation and to combine Bayesian ensembling with uncertainty-aware modeling for this task.

This paper investigates whether LLM translations exhibit identifiable emotional profiles and how post-editing reshapes them toward human-like norms. We compare LLM translations of Margaret Atwood's Oryx and Crake with their post-edited versions and a human translation, using a large-scale corpus of contemporary Italian science-fiction as a baseline. We examine emotion through lexicon-based and multilingual modeling, conducting a fine-grained analysis of emotional variation across systems. We find that MT systems introduce model-specific and statistically significant emotional fingerprints across translations, leading to a limited preservation of an author's voice.

This paper presents a nonlinear parameter estimator for Wiener-type state-space models obtained as a fixed-point architecture that couples two affine minimum mean-squared error (MMSE) estimators: one for the unknown parameters and one for latent variables. The architecture retains the functional structure of the optimal affine MMSE parameter estimator while incorporating Dynamic Basis Statistics (DBS) estimates that summarize nonlinear basis-function evaluations. Two DBS construction strategies are developed, leading to two nonlinear estimator frameworks. The dual basis-parameter estimator combines an affine basis estimator with the affine parameter estimator, whereas the dual state-parameter estimator first computes affine state estimates and their covariances, then maps these state-estimate statistics through a Gaussian DBS operator to obtain DBS estimates. Both dual estimators admit fixed-point characterizations that alternate between estimating each component using the updated prior of the other, obtained from that component's plug-in estimate statistics from the previous iteration. The efficacy of the proposed methods is examined via extensive Monte Carlo experiments, showing that the dual basis-parameter estimator attains parameter mean-squared errors comparable to those of the purely affine parameter estimator, while the dual state-parameter estimator achieves the lowest parameter mean-squared error, outperforming both the dual basis-parameter and purely affine parameter estimators, as well as sequential Monte Carlo variants of classical Particle Gibbs and Expectation-Maximization schemes.

Reliable evaluation of instance segmentation models requires metrics that accurately and consistently reflect segmentation quality. However, the metrics most widely used in biological imaging carry fundamental mathematical weaknesses: hard Intersection-over-Union (IoU) thresholds that produce discontinuous, low sensitivity scoring; per-object normalization that distorts scores under object size variation; and greedy or one-to-many matching procedures that yield non-optimal, order-dependent correspondences. Together, these properties produce unintuitive and unreliable model rankings under common failure modes such as split cells, merged cells, and cell boundary imprecision. We propose Maximum Matching Accuracy (MMA), a threshold-free continuous metric that finds a globally optimal one-to-one matching between predicted and ground truth objects and aggregates total overlap using per-pixel normalization. We evaluate MMA against AP@50, PQ, SEG, and AJI across three experiments: synthetic failure cases, progressive corruption tests, and a model ranking comparison. MMA produces scores that are more stable, more sensitive, and more interpretable than existing alternatives, providing a principled foundation for fair instance segmentation benchmarking in biological cell imaging.

The rapid development of large language models (LLMs) has raised concerns about misuse such as plagiarism, misinformation, and automated influence operations, motivating the need for robust detectors. Recent work has shown that neural representations of writing style are effective for detection and, crucially, robust to adversarial attacks that defeat most existing detectors. However, current style-based detectors rely on authorship labels for training, and are limited to few-shot inference for detection, requiring in-distribution samples that may not always be available. We learn discriminative style features without authorship labels by training a style encoder to reconstruct human-authored text from its machine-generated paraphrase; freezing a semantic encoder during training biases the style encoder to capture only the non-semantic features needed for reconstruction. We evaluate the learned representations via two detection strategies: a few-shot detector and a zero-shot DeepSVDD-based detector. Across benchmarks, our method matches or outperforms all baselines in the few-shot setting and, in the zero-shot regime, is competitive with fully supervised classifiers on in-distribution test data while generalizing better to unseen LLMs. Beyond detection, the learned representations generalize to unseen tasks, achieving competitive performance on authorship verification and fine-grained style discrimination despite never being trained on either objective.

Classical theories of cognition describe problem solving as exploratory search through structured problem spaces in which repeated interaction gradually compresses search into efficient representational structures. Predictive artificial intelligence systems introduce a distinct regime in which stabilization may occur before exploratory diversification unfolds, supplying solutions and decision trajectories prior to internally generated search. This paper develops a geometric dynamical framework in which attention evolves over a landscape of strategies shaped by stabilizing drift, endogenous exploratory perturbation, and responsiveness-gated learning. Predictive assistance is modeled as a process of exogenous exploratory compression that stabilizes trajectories before self-generated exploration broadens the accessible regions of strategy space. The framework yields three main results. First, sustained predictive stabilization reduces exploratory responsiveness by attenuating the effective influence of intrinsic perturbations even when exploratory variability remains present. Second, curvature accumulates and relaxes asymmetrically, producing hysteresis and delayed recovery of exploratory mobility after assistance withdrawal. Third, developmental outcomes depend critically on the timing of stabilization, with early intervention narrowing future exploratory traversal before broad representational diversification has occurred. The framework generates empirically testable predictions concerning exploratory entropy, premature convergence, and delayed recovery following predictive stabilization. More broadly, the results suggest that predictive systems may reshape the geometry of exploratory cognition itself.

Decision-making under risk is typically studied through single-shot lottery choices. Yet many real decisions involve combinatorial risk, where risk arises from multiple risky components, so the lottery over outcomes is induced rather than given outright and can be costly to evaluate exactly. We introduce an investment-allocation task to study decision under combinatorial risk, where investing in a component raises its success probability and thereby reshapes the outcome distribution. Participants favor the option with the larger probability increment, and, when increments are equal, the option with the higher initial success probability. Revealing the induced probability mass function (PMF) substantially changes behavior, making participants less responsive to combinatorial-risk features and reducing choice variance. To explain these patterns, we move beyond standard benchmarks and hand-crafted hypotheses with symbolic regression to discover compact descriptive models. The discovered models rely mainly on combinatorial-risk features, such as the after-investment success probability, rather than exact evaluation of the full induced distribution. Behavior under the displayed PMF is then well explained by augmenting this model with a prospect-theoretic residual model. The results show that people navigate combinatorial risk primarily through its core features, shifting toward lottery valuation only when the induced PMF is displayed.

In this work, we develop theoretical foundation for flow matching with neural-network-parameterized conditional velocity fields. We establish convergence guarantees for gradient descent in the over-parameterized 2-layered ReLU neural network regime. We derive generalization bounds for the conditional velocity-field matching objective. Building on these results, we provide Wasserstein-distance guarantees for the samples generated by the induced flow. Our analysis is based on generalization bound for multi-task representation learning with unbounded losses, which may be of independent interest beyond flow-based generative modeling. These theoretical results are validated through extensive experiments on both synthetic and real-world image benchmarks.

Reduced facial expressivity is a common motor manifestation of Parkinson's disease (PD), often described as hypomimia or facial bradykinesia. This paper examines whether temporal motion descriptors extracted from facial-region keypoints can support in-the-wild PD-related video classification on the YouTubePD benchmark. Each video is represented using geometric descriptors from 14 predefined facial regions. Static geometry, normalized geometry, velocity-based descriptors, relative-velocity descriptors, and a GRU sequence baseline are compared under the same binary classification protocol. To assess stability and interpretability, the study includes seed-robustness analysis, region-level ablation, and permutation importance. The best result is obtained with normalized velocity descriptors and a Random Forest classifier, reaching a balanced accuracy of 0.826 and an AUROC of 0.855 on the held-out test split. Across 10 random seeds, this representation remains stable, with balanced accuracy of 0.810 +/- 0.018 and AUROC of 0.855 +/- 0.005. Overall, the results suggest that normalized facial-region motion is a lightweight and interpretable representation for YouTubePD video classification. The study is framed as a benchmark-level analysis and does not claim clinical severity assessment or MDS-UPDRS facial-expression scoring.

Pre-training on raw code teaches syntax but provides sparse signal for diverse real-world task formats. While synthetic data has proven transformative for language models, code remains largely unexplored beyond limited quality improvements. We present CodeAlchemy, a synthetic data generation framework that transforms publicly sourced code into semantically-rich training data through 5 strategies: CodeEnhance (quality-aware rewriting), CodeQA (template-based problems), CodeDev (developer tasks), CodeDialogue (multi-turn conversations), and CodeTrace (execution traces). We process 3 corpora across 15 languages to generate 500B+ tokens of synthetic data plus 350B reasoning tokens, orders of magnitude more than prior efforts. CodeTrace instruments and executes 1.3M+ files across 14 languages and 5K libraries, capturing control flow, state tracking, and library knowledge. We introduce DevEval (developer tasks) and TraceEval (execution prediction) benchmarks; frontier models like Claude Sonnet 4.5 achieve only 5.6% exact match on TraceEval, revealing critical gaps in semantic understanding. Our 3B models achieve 83.5% on HumanEval, 63.2% on MBPP, 8.09% win rate on DevEval, and 15.36 ROUGE-2 on TraceEval, outperforming frontier models 10x the size including 27B Gemma-3 and 32B Granite-4.0.

This paper develops a theory of exploratory adaptation under AI-assisted optimization. The central argument is that the long-run adaptive effects of AI systems depend critically on how predictive assistance interacts with exploratory responsiveness itself. We formalize this mechanism using a dynamical framework in which cognitive, institutional, and technological systems evolve over rugged epistemic landscapes characterized by multiple locally reinforced configurations. A central state variable in the model is adaptive responsiveness, which measures the capacity of a system to traverse unfamiliar conceptual and institutional trajectories under changing conditions. Under convergent predictive regimes, AI systems substitute for exploratory engagement, reducing adaptive responsiveness and generating metastable trapping, hysteresis, premature convergence, and exploration-collapse dynamics in which systems become locally efficient but globally rigid. The framework also identifies contrasting exploration-enhancing regimes in which AI systems amplify exploratory search, conceptual traversal, and adaptive mobility. The effective substitution parameter is therefore responsiveness-dependent: systems possessing weak exploratory routines are more vulnerable to exploratory substitution, whereas systems already possessing high adaptive responsiveness may use AI assistance to expand exploratory mobility across rugged landscapes. The long-run adaptive effects of AI consequently depend not only on AI capability itself, but also on institutional structure, developmental context, and the architecture of human-machine interaction.

This work presents a divide-and-conquer modeling strategy for the CTF-4-Science Lorenz benchmark, which evaluates chaotic-system prediction across twelve hidden scores and five scenario families: clean forecasting, noisy reconstruction, noisy-input forecasting, few-shot learning, and parametric generalization. Rather than forcing one model class to handle all regimes, the final system matched each prediction block to the evaluation behavior of its task group. The main contributions are: smoothing-based reconstruction for noisy full-trajectory denoising; NG-RC/NVAR models tuned for noisy long-time attractor forecasting; a fitted Lorenz transition correction restricted to the sensitive clean short-time prefix; and a parametric prefix blend for the interpolation task. The resulting system with final public score of 79.63 shows that bounded, scenario-specific updates can outperform broad model replacement on mixed chaotic forecasting benchmarks.

Generative models have shown remarkable progress in a variety of domains such as protein design, but such power enables the opaque generation of hazardous proteins. In this work, we introduce VFUSE (Virulent Feature Understanding with Sparse autoEncoders), a mechanistic interpretability approach that trains SAEs on diffusion-transformer activations to audit protein models for hazard-aware features. We apply VFUSE to RoseTTAFold3 and RFDiffusion3, popular open-weight models for protein folding and synthesis. We find that for certain blocks, linear probes detect hazardous designs significantly better when fit in the SAE latent space over the original model's representations: improving interpretability without sacrificing model performance. Furthermore, we identify monosemantic features from the SAE that fire only on hazardous designs at up to AUROC $0.84$ ($q < 10^{-13}$). To our knowledge this is the first SAE trained on an all-atom diffusion model and the first feature-level virulence audit of a protein design model, paving the way towards safe and interpretable protein design.

We introduce Temporal Sheaf Neural Networks (TSNN), a temporal link prediction framework that equips each node with a time-varying orthogonal frame and compares node states only after explicit transport between local coordinate systems. In contrast to existing continuous-time graph models that operate in a shared global embedding space, TSNN models node-specific and evolving interaction semantics through dynamic local frames. The model parameterizes per-node frames via efficient low-rank Householder products, preserves stored hidden states exactly under frame updates, and uses a geometric-residual decoder that anchors predictions on transported distances while learning residual corrections. All computations are strictly causal and use only the pre-event history. We show that the symmetric degree-normalized sheaf Laplacian is orthogonally similar to the symmetric normalized graph Laplacian, with the random-walk normalized form similar in the corresponding degree metric; the full-active, feature-scaled diffusion used by TSNN is exactly a metric-gradient step on the combinatorial sheaf Dirichlet energy, with a degree-free monotone-descent and non-expansiveness guarantee. Frame drift perturbs updates only linearly. Across TGB v2 link-prediction and temporal-heterogeneous leaderboards, together with the DGB benchmark suite, TSNN matches or surpasses the strongest prior methods on most benchmarks, with the largest improvements on graphs exhibiting strong node-role heterogeneity. Ablations confirm the distinct benefit of dynamic frames, orthogonal transport, and geometric-residual decoding.