Publications

Evaluating Predictor-Corrector Dependence in Hard-Constrained Climate Emulators
Xinyu Wang
Zhenghan Tai
Xiao-Wen Chang
Climate emulators often enforce physical constraints by correcting each state predicted by a neural network, but evaluating only the correct… (see more)ed forecast conflates predictor skill with the effect of the correction. We introduce an evaluation framework that separates these contributions by comparing corrected forecasts, pre-correction predictions along corrected trajectories, and autonomous rollouts with correction disabled. We further cross correction during training and inference to distinguish its immediate effect from how it shapes the learned predictor, and use matched nonphysical controls to test whether improvements are specific to the physical correction. In ACE2, correction substantially improves physical consistency. In a three-seed ACE2 pilot, models trained with correction produce less accurate uncorrected precipitation forecasts at short horizons than models trained without correction, but more accurate forecasts at 10 and 30 days. This contradicts our preregistered hypothesis that the deficit would grow with forecast horizon. In a separate nine-seed experiment, directly supervising the uncorrected output reverses its precipitation forecast deficit at 72 hours while maintaining the accuracy of corrected forecasts at 24 hours. In ConCerNet's heat-equation system, projection-trained models have higher uncorrected trajectory-based vector-field RMSE than models trained without projection, while restoring the physical projection recovers accuracy far more effectively than a matched nonphysical control. Together, these results show that hard corrections can reshape the predictors trained alongside them, creating dependencies that are not visible from corrected forecasts alone. Hard-constrained emulators should therefore be evaluated by separately measuring predictor accuracy, correction effects, and their interaction during training.
GCAM-Gym: A Benchmark for Agent-Based Climate Policy Search
Zhenghan Tai
Xinyu Wang
Comparing climate-policy search methods requires accounting for differences in policy control, available observations, feedback timing, and … (see more)simulator budgets. We introduce GCAM-Gym, a Gymnasium-compatible environment that supports full-policy evaluation and stepwise control through a common interface and exposes the components of the policy objective. We compare seven search and control systems in three settings: target attainment and cost, cost-temperature trade-offs, and one-step delayed feedback. Each system is evaluated with ten seeds per setting, for a total of 3,210 attempted GCAM evaluations. Four systems share a budget of 24 attempts per seed; among these, CMA-ES achieves the highest interquartile mean (IQM) of the best returns from completed policy evaluations in all three settings. We release the evaluation archive and analysis tools to reconstruct the reported comparisons without running GCAM.
VERIDAH: solving enumeration anomaly–aware vertebra labeling across imaging sequences
Hendrik Möller
Hanna Schoen
Robert Graf
Matan Atad
Anjany Sekuboyina
Bettina K. Budai
Fabian Bamberg
Steffen Ringhof
Christopher Schlett
Tobias Pischon
Thoralf Niendorf
Josua A. Decker
Marc-André Weber
Bjoern Menze
Daniel Rueckert
Jan S. Kirschke
OBJECTIVES: Although the identification of enumeration anomalies has potential clinical implications for chronic back pain and operation pla… (see more)nning, the thoracolumbar junction is poorly assessed and rarely described in clinical reports. Additionally, even though multiple deep-learning-based vertebra labeling algorithms exist, there is a lack of methods to automatically label enumeration anomalies. MATERIALS AND METHODS: We introduce "Vertebra Identification with Anomaly Handling" (VERIDAH), a novel vertebra labeling algorithm based on multiple classification heads combined with a weighted vertebra sequence prediction algorithm. We use the NAKO dataset for MRI and a CT cohort to retrospectively evaluate the labeling performance by computing how often the model is correct in predicting the vertebra labels, including enumeration anomalies. Statistical significance was assessed using the Wilcoxon signed-rank test. RESULTS: On 1000 T2w sagittal (mean age 52 ± 11, 52% female) and 365 CT (mean age 74 ± 21, 44% female) images, we show that our approach surpasses existing models on T2w TSE sagittal (98.30% vs. 94.24% of subjects with all vertebrae correctly labeled, p 0.01) and CT imaging (99.18% vs. 77.26% of subjects with all vertebrae correctly labeled, p 0.01) and works in arbitrary field-of-view images. VERIDAH correctly labeled the presence of thoracic enumeration anomalies in 87.80% and 96.30% of T2w and CT images, respectively, and lumbar enumeration anomalies in 94.48% and 97.22% for T2w and CT, respectively. CONCLUSION: VERIDAH offers enumeration anomaly robust vertebra labeling in CT and T2w sagittal imaging and is publicly released along with a spine segmentation model. KEY POINTS: QuestionAutomatically identifying enumeration anomalies in the spine is important in determining the correct vertebra labels, especially in limited field-of-view imaging. FindingsA carefully designed classifier with multiple classification heads and a sequence predictor provides superior results on large benchmarks, both in CT and MRI. Clinical relevanceThe proposed method identifies vertebra levels even in very limited field-of-view imaging, and correctly predicts enumeration anomalies, making downstream tasks such as surgery planning easier. Its public availability enables studies and applications to take these spinal anomalies into account.
Evaluating the Pedagogical Quality of LLM-Generated Feedback: A Criterion-Based and Comparative Study
Harvey Ngoe Kolle
Carrie Demmans Epp
Amna Liaqat
FeaTok: Feature-Loss Tokenization for Image Generation
Zhang Xiaofeng
Samuel Lavoie
Nicolas Beltran-Velez
Jianhao Yuan
Felix Friedrich
Jakob Verbeek
Michal Drozdzal
Adriana Romero-Soriano
Discrete image generators learn the distribution of tokens produced by an image tokenizer. These token representations should be easy for ge… (see more)nerators to model, not merely enable faithful image reconstruction. Yet these tokenizers are commonly trained to reconstruct pixels, which prioritizes low-level fidelity and may disrupt representation structure useful for generation. Motivated by feature-predictive self-supervised learning, we introduce feature-loss tokenization (FeaTok), which trains discrete tokens by reconstructing pretrained features rather than pixels. Ablations across SigLIP, SigLIP2, DINOv2, and DINOv3 show better generative performance with FeaTok. In class-conditional generation, FeaTok advances the quality-compute Pareto frontier, reducing gFDr6 at 256x256 by up to 29 percent over the strongest discrete baselines and outperforming the strongest unguided continuous baseline (RAEv2) with 38 times less sampling compute. FeaTok also achieves higher human-preference reward-model scores than prior art without reward-based training and surpasses continuous and discrete text-conditional baselines on three benchmarks with only 500 fine-tuning updates after pre-training.
Hard Way, The Only Way: Provable Merits of Training on Hardest Examples
Elvis Dohmatob
Despite a surge of empirical interest in the ``unreasonable effectiveness'' of drastic data pruning for AI systems, a unified theoretical un… (see more)derstanding of this phenomenon remains elusive. In this work, we provide a precise high-dimensional theory of learning under data pruning via weighted empirical risk minimization (ERM). As in previous work, we analyze a setting where pruning is performed along a one-dimensional subspace of the input space. Our analysis precisely characterizes the limiting test error curves in the regime where the input dimension N and sample size P tend to infinity with a fixed ratio N/P. The resulting theory recovers several recent results in the literature as special cases and provides a general lens to evaluate how pruning strategies interact with the loss function and label noise. Crucially, we characterize the optimal pruning strategy in the high-sample regime where P greatly exceeds N. We demonstrate that strategic pruning can accelerate the decay of classification test error from the classical sqrt(N/P) rate, typical of vanilla ERM with loss functions without exponential tail (e.g., quadratic loss), to the much faster Bayes-optimal scaling of order N/P. In this setup, we prove that, surprisingly, this optimal performance is achieved by a drastic pruning strategy: discarding almost all training data except for a small proportion, consisting strictly of the 2N hardest training examples.
The Tilted Sampling Problem: Optimism, Posterior Sampling, and Logarithmic Regret
Many modern learning systems are used not to select a single best action, but to sample diverse high-quality outputs. Examples include entro… (see more)py-regularized reinforcement learning, RLHF, and reward-guided generative modeling. Motivated by these settings, we study tilted sampling: an online learning problem in which the goal is to sample from the Gibbs distribution induced by an unknown reward function. We define sampling regret as the temperature-normalized cumulative KL divergence between the learner's distribution and the Gibbs target, corresponding to the suboptimality gap in entropy-regularized reward maximization. Our framework generalizes recent sampling-based and KL-regularized formulations in bandits and reinforcement learning. Our main contribution is a general regret framework that captures the different uncertainty structure of tilted sampling and classical regret minimization. While standard regret scales as
Understanding Stability of Outer Nesterov Momentum on Stochastic Pseudo-Gradients
Laura Gomezjurado Gonzalez
Ganesh Talluri
Large neural networks are commonly trained across many workers, and communication between them is a major cost. Workers therefore take sever… (see more)al local optimization steps before sending their net parameter displacements, or pseudo-gradients, to an outer optimizer, which often applies Nesterov momentum. We study how the outer momentum, the number of local steps and the averaging over workers control the training dynamics, and in particular their stability. For random quadratics, we derive an exact mean-square stability criterion in any dimension, which reveals that the inner batch size determines whether increasing the number of local steps eventually stabilizes or destabilizes the outer iteration. Moreover, while averaging the pseudo-gradients of independent workers can make larger outer momentum values admissible, the number of workers required for stability can grow exponentially with the number of local steps. Finally, we show that the mean-square dynamics governing stability can be probed directly at intermediate neural-network checkpoints using shared-minibatch perturbations. We test our theoretical findings across networks, and measure their consequences for training loss.
Building capacity for cervical cancer: a mixed-method evaluation of the training program for nursing professionals
Tavleen Kaur Malik
Anisha Mathew
Bharti Singh
Anila Goswami
Sandeep Bhalla
Apoorva Karan Rai
Nurses play a critical role in patient care and health promotion. Enhancing their knowledge and capacity in cervical cancer prevention, scre… (see more)ening, and early detection can improve patient outcomes and community awareness. This study provides the results of the assessment of an ECHO (Extension for Community Healthcare Outcomes) tele-mentoring program designed to build the capacities of nursing professionals in these domains. A mixed-methods study was conducted. Seven hubs delivered seven virtual sessions each, totalling 49 sessions and training 447 nursing professionals. The study consisted of three components: first, pre- and post-surveys ( n = 191) assessed changes in knowledge and attitudes; second, post-session surveys ( n = 1168) gathered participant perceptions after each session. Qualitative insights were obtained through the Most Significant Change (MSC) technique, capturing in-depth stories from the hub members who delivered the training. The pre- and post-assessments and the post-session survey were conducted with the training participants and included both quantitative and open-ended questions. The Wilcoxon signed-rank test was used to analyse the statistical significance of knowledge gains, while thematic analysis was applied to the qualitative data. The Institutional Review Board of the International Institute of Health Management Research (IIHMR), Delhi, granted ethical approval for the study. Informed consent was obtained separately for each tool from the study participants. The average knowledge score improved from 6.6 to 7.5 ( p 0.001) post-intervention. A large proportion of participants reported an increased willingness to undergo cervical cancer screening (83.6%) and a commitment to promote screening and awareness among peers and communities (98%). Post-session responses found the content relevant to their clinical roles (77%) and expressed intentions to apply the learnings (78%). Feedback highlighted the need for improved session timing, internet connectivity, and inclusion of additional health topics. Qualitative stories through the Most Significant Change technique hinted that some participants had already begun fostering positive work environments and enhancing communication with patients and families. The ECHO training has strengthened the knowledge and attitudes of nursing professionals in cervical cancer prevention, detection and screening. The qualitative interviews from the stakeholders also suggest interest and motivation to continue this initiative. The findings demonstrated potential for scale-up and adaptation across other oncology-related topics, positioning nurses as catalysts for change in community health promotion.
Defining and Generating Compositional Structure
Compositional structure is pervasive in nature and across scales: atoms combine into molecules, molecules into cells, and cells into organis… (see more)ms, while further compositional hierarchies appear within each level. More broadly, compositionality, the reuse and recombination of parts into novel structures, is a fundamental source of interesting complexity in both natural and artificial systems. Yet despite its importance, and despite our apparent ease in recognizing it, mathematical accounts capturing what we intuitively mean by compositional structure remain limited. Here, we fill this gap by formalizing a simple idea: compositional structure emerges when parts, considered together, reveal shared structure that is not captured when those parts are modeled separately. We express this idea using algorithmic information theory, the mathematics of information and compression, which provides a natural way to identify structure in arbitrary data without specifying in advance what that structure must look like. Applied across scales, our definition informs where and how compositional structure emerges within an object. Using tractable approximations, we then invert the framework to generate compositional objects: we specify the desired form of their compositional structure without prescribing their particular content, and optimize directly over strings and images. The resulting objects exhibit readily recognizable hierarchical reuse and part-whole organization, suggesting that the definition captures compositional structure rather than merely complexity, regularity, or shared information. Finally, we discuss implications for machine learning, including a new perspective on depth in neural architectures and principles for constructing compositional data curricula capable of supporting open-ended learning.
Dungeon and Visibility World Generator
Antonio Loría
Robotic exploration, coverage, and multi-agent deployment algorithms require reproducible environments with controllable geometry and visibi… (see more)lity. This paper presents the Dungeon and Visibility World Generator, an open-source framework for constructing two- and three-dimensional environments and evaluating geometric line-of-sight. The framework supports orthogonal, polygonal or polyhedral, and curvilinear geometries, with six construction families covering scattered rooms, clusters, winding passages, caverns, branches, and loops. Room dimensions, passage structures, obstacles, and spatial placement can be controlled through configurable parameters and probability distributions. Each generated environment provides continuous free-space geometry and a corresponding discrete occupancy representation. A Rust-based visibility engine supports line-of-sight queries in both representations, enabling the evaluation of observable regions from arbitrary agent positions. The framework records configurations, generated geometries, occupancy grids, and visibility results to support reproducible experiments. We demonstrate the generator across all six geometry classes, examine how geometry and spatial resolution affect visibility, and evaluate generation time and reproducibility. The resulting framework provides a configurable foundation for evaluating robotic exploration, coverage, and multi-agent deployment algorithms in diverse environments.
ELARSpeech: A 30K-Hour Dataset Across 675 Underrepresented Languages for Long-Tail Multilingual Speech Technologies
Massively multilingual speech modeling has made remarkable progress in recent years, yet existing training resources and benchmarks still un… (see more)derrepresent much of the linguistic long tail. To address this gap, we introduce ELARSpeech, a large-scale multilingual speech resource comprising 30,202 hours of speech across 675 underrepresented languages, curated from the Endangered Languages Archive. Combining human-annotated and unlabeled recordings, the corpus spans diverse recording conditions, domains, and language families. Beyond corpus construction, our experiments show how ELARSpeech supports extending ASR and TTS capabilities to diverse underrepresented languages, with further benefits from pseudo-labeling unlabeled data. At the same time, our speech representation analysis reveals remaining limitations in transferring existing multilingual models to the linguistic long tail. Together, ELARSpeech provides both a resource and a testbed for extending multilingual speech technology further into the linguistic long tail. We publicly release the dataset, all trained ASR and TTS models, and the complete data construction pipeline to facilitate research on long-tail multilingual speech technologies.