Johns Hopkins University

Standard galaxy quenching pipelines have been systematically discarding the very high-luminosity quasars ($L_{\text{Bol}} \gtrsim 10^{45}\text{ erg s}^{-1}$) responsible for shutting down star formation in the first place.

Routine clinical reports and historical scans can entirely replace manual voxel annotations, yielding dense tumor segmenters that outperform supervised public baselines at zero labeling cost.

Human hair priors can substitute for non-existent animal fur datasets, enabling dense, editable 3D animal grooming with an order-of-magnitude faster optimization.

Decades-old conjectures on the tight convergence of Nesterov's accelerated methods can now be solved automatically by swapping general-purpose formal provers for domain-structured symbolic synthesis.

Experiments show that VoiceTrace achieves state-of-the-art performance on established semantic speech retrieval benchmarks, while substantially outperforming cascade-based approaches on VoiceTrace-Bench, demonstrating its effectiveness for both conventional semantic retrieval and the new hybrid retrieval setting.

The Mind2Dialogue framework proposes a psychology-guided simulator that preserves personal characteristics while updating mental states through interaction to generate coherent conversations, and trains models on the Oracle's well-informed responses to assist users without direct access to their mental states at deployment.

Deformable registration barely outperforms simple affine baselines for neurovascular MRI-MRA fusion, frequently generating anatomically nonsensical deformations that easily trick standard image-similarity metrics.

Sentiment in social-media conversations is significantly influenced by discourse moves, with specific interventions capable of mitigating negativity or amplifying hostility.

Multi-stereo reconstruction outperforms feed-forward models by achieving 67% coverage of ground-truth surgical surfaces with just three viewpoints, highlighting the critical role of viewpoint diversity in surgical perception.

Loneliness in older adults can be detected through a powerful combination of speech patterns and vocal characteristics, revealing critical insights into their emotional states.

Cleaner speech datasets may boost in-domain performance but can undermine generalization, revealing a hidden cost in Alzheimer's detection models.

Transcript-based shortcuts in dialogue models lead to a staggering drop in accuracy, revealing a critical flaw in current evaluation methods.

Unbiased sampling of source prefixes in TLMs can reduce runtime by several orders of magnitude while maintaining accuracy in estimating target prefix probabilities.

AnaDiffusion allows for precise and controllable editing of 3D brain MRIs, achieving the lowest FID scores while preserving anatomical integrity.

Achieving a 50x speedup in Bayesian imaging sampling without sacrificing quality could revolutionize real-time imaging applications.

The curse of multilinguality may not be a fundamental barrier; instead, it’s a byproduct of data and training practices.

A modular deep learning framework that enhances neuroimaging task performance by leveraging optimal task sequences from over 49,000 MRIs.

Achieving \(O(ε^{-3})\) complexity for stochastic root-finding without the usual variance reduction techniques could revolutionize the efficiency of numerical methods in uncertain environments.

Programmatic skill learning can slash agent costs while enhancing performance, with SpeedRunner leading the charge in cost-efficient adaptation.

Exploiting the mismatch between implicit human context and LLM safety alignment can lead to unprecedented attack success rates, revealing a critical vulnerability in current AI systems.

MT models show a troubling bias towards US content, with specialized models proving less robust across diverse locales.

Emotion-sensitive neurons in LALMs are language-specific, but pooling cross-lingual evidence reveals powerful, transferable Multilingual Emotion Neurons that enhance affective control.

Achieving millimeter-scale depth accuracy in real-time laparoscopic surgery without the need for synchronization could revolutionize surgical guidance systems.

Myopic planners can fail spectacularly when faced with the need to acquire capabilities for future experiments, leading to unbounded approximation ratios in goal-directed discovery.

Training-free methods outperform their training-based counterparts in robustness against adversarial prompts while still achieving competitive accuracy on simpler tasks.

Ignoring the social dynamics of human-agent collaboration in science could stifle innovation and diversity in research outcomes.

Dropping echocardiogram data nearly doubles error rates in cardiac AI models, revealing critical modality dependencies that could impact clinical decisions.

A groundbreaking collection of 16 schemas that standardizes scientific process descriptions, paving the way for enhanced reproducibility and automation in research.

Proton reveals 23 zero-day vulnerabilities in Electron apps, with 22 enabling OS command execution, exposing critical security flaws in popular software frameworks.

New benchmarks reveal that even advanced LLMs struggle with cultural value alignment, but targeted fine-tuning using Sri Lankan values can significantly enhance their performance.

Achieving desired accuracy in optimization may require significantly less memory than previously thought, with a critical phase transition in online iteration budgets that can redefine efficiency benchmarks.

Exponential sample complexity in reachability analysis is intrinsic, revealing fundamental limits that challenge the scalability of sampling-based methods in high-dimensional systems.

Automating real-to-sim conversion with vision-language agents could revolutionize how we simulate robotic interactions, making it faster and cheaper than ever before.

IRT models can mislead AI evaluations, especially when benchmarks deviate from traditional testing conditions, risking inaccurate performance assessments.

NodeImport reveals that strategically filtering nodes based on importance can dramatically enhance GNN performance in imbalanced settings.

HTFM transforms the handling of heavy-tailed data, achieving superior sample quality and tail recovery while maintaining efficient sampling speeds.

Adversarial attacks on vision-language agents reveal critical vulnerabilities, with multi-view optimization strategies proving significantly more effective than isolated approaches.

Over 1,100 submissions reveal groundbreaking advancements in sports video understanding, with new methods pushing the boundaries of action prediction and localization.

SSL representations of satellite imagery can reveal hidden environmental associations that significantly impact downstream task performance.

Role-typed credit assignment can drastically improve reinforcement learning outcomes by accurately distinguishing between useful exploration and regression in agent actions.

FaceMoE achieves unprecedented performance in low-resolution face recognition by leveraging a dynamic Mixture of Experts architecture that promotes specialization and preserves pretrained knowledge.

Automated extraction of data from vector graphics achieves unprecedented precision, enabling researchers to recover and verify scientific data with confidence and efficiency.

Matched reference regimes for prosody evaluation reveal that traditional methods over-flag deviations, leading to misinterpretations in speech AI assessments.

Semantic embeddings falter in stylistic evaluations, revealing a critical gap in current embedding methodologies.

GaussDet achieves a remarkable 16.7% boost in referential grounding accuracy, redefining the capabilities of open-vocabulary 3D scene understanding.

GRAINS achieves up to 47.8x speedup in genome graph analysis by processing data directly within storage, revolutionizing efficiency in genomic research.

State-of-the-art models barely scrape 20% accuracy on a new benchmark designed to tackle the complexities of theory-scale auto-formalization.

Traditional measures of sentence processing fail to account for garden path difficulties, but a new framework reveals that syntactic belief updates are key to understanding these challenges.

Retaining the original question in multilingual reasoning cascades can dramatically enhance performance, revealing that context is key to effective translation and reasoning.

AI-generated liver MRI reports can achieve 76% case-level sensitivity and are rated clinically acceptable by radiologists, challenging the status quo of manual reporting.

ForceBand transforms human muscle signals into precise force data, enabling robots to learn manipulation tasks with unprecedented accuracy.

Current AI models miss critical tumor detections in underrepresented demographics, revealing a hidden bias that could compromise patient outcomes.

A novel electrochemical sensor can accurately distinguish between DNA intercalators and minor groove binders, streamlining drug candidate screening.

Prompt optimization can significantly enhance multi-agent LLM systems, but its effectiveness hinges on the specific configuration of the agents and their interactions.

Identifying and resolving parameter degeneracies can cut simulation costs by up to 10x while enhancing our understanding of complex models.

Memory systems struggle to adapt as user profiles evolve, with over 93% of failures linked to memory retrieval rather than response generation.

Gradient descent can reliably converge to stationary points in complex neural networks, challenging previous assumptions about initialization and architecture.

Runtime safety detection in coding agents can be achieved through a novel intervention in hidden representations, drastically reducing harmful actions during multi-turn interactions.

Transforming 3D image editing from ambiguous controls to precise geometry specifications, this method allows for unprecedented accuracy in real-world applications.

A single generalist model outperforms specialized systems, achieving over 35% improvement in real-world robotic task success.

OR3's innovative use of action-driven digital twins enables precise retrieval of critical OR events that traditional methods struggle to identify.

Geometry-guided adaptation can dramatically enhance the reliability of endoscopic navigation, overcoming traditional challenges in depth perception and feature alignment.

Grounded explanations for speech deepfake detection can boost accuracy by over 45%, transforming how we interpret AI decisions in this critical domain.

Confidence-based remasking in dLLMs may not deliver the expected improvements and can actually worsen diversity issues in certain decoding settings.

Every team in the generation task produced at least one report deemed the best by human annotators, highlighting a leap in multimodal generation quality.

Achieving 95% accuracy, MSUE's innovative multi-expert architecture redefines how we approach visual question answering in complex domains.

Outdated laws are stifling the potential of AI agents, which could revolutionize user interaction with the internet.

Delegating prediction tasks in human-AI teams may preserve calibration but imposes a daunting challenge on the rejector model to accurately assess expertise.

A groundbreaking dataset of 313 hours of real-world code-switched speech reveals rich patterns and frequencies previously overlooked in multilingual research.

Top systems in the ESDD2 challenge achieved a staggering Macro-F1 score of 0.8775, revealing the power of modular design and self-supervised learning in audio deepfake detection.

Adversaries can achieve complete control over robotic policies in real-time by exploiting visual conditioning vulnerabilities, turning them into remotely piloted instruments.

Synthetic eye movement data can reliably identify neurophysiological abnormalities, paving the way for accessible and efficient diagnostic tools.

The effective dimensionality $d_{eff}$ reveals how well physics constraints absorb network degrees of freedom, transforming our understanding of PINN performance and adaptation.

Document LoRA can recover up to 21 ROUGE-L points when context is scarce, redefining its role in memory architecture for question answering.

A strategic messaging shift on Google Search reduced CSAM-related queries by 3.8%, effectively redirecting some users towards therapeutic resources.

MPC-RL achieves superior humanoid locomotion and manipulation performance by integrating efficient MPC guidance, challenging the traditional RL training paradigms.

ColBERTSaR shrinks the index size by up to 70% without sacrificing retrieval performance, revolutionizing the efficiency of neural retrieval systems.

Agentic harnesses can significantly enhance LLM performance on deontic reasoning tasks, but not without introducing risks of degradation in numerical accuracy.

Cosmos 3 sets a new benchmark for omnimodal models, outperforming existing state-of-the-art in both Text-to-Image and Image-to-Video tasks.

Exploiting noise inseparability allows for a breakthrough in weakly-supervised speech denoising, enhancing domain adaptability and performance without relying on clean targets.

AI can autonomously generate superior healthcare messaging interventions by learning from past experimental data, outperforming traditional methods by a notable margin.

Certifying belief-space safety filters can lead to significantly less conservative safety measures in autonomous robotics, enhancing interaction efficiency without compromising safety.

Agents struggle to match human performance in long-horizon tasks, revealing critical gaps in their learning capabilities during deployment.

Forget just finding relevant documents – CoveR boosts the *diversity* of retrieved information by 10%, a critical step for comprehensive long-form RAG.

Diffusion models can get stuck in local modes, with high-level semantic features relaxing much slower than low-level details, hindering efficient sampling.

Energy-guided diffusion models can bridge the gap between source and target domains in off-dynamics offline RL, enabling effective trajectory generation without retraining the diffusion model.

Unlock the long tail of autonomous driving scenarios: Sensor2Sensor turns readily available dashcam footage into high-fidelity, multi-modal sensor data, bridging the gap between data scarcity and the need for robust AV training.

Achieve sharper MRI reconstructions from extremely sparse measurements by recasting the problem as discrete autoregressive prediction, sidestepping the blurring inherent in continuous pixel-domain approaches.

Open-sourcing a VLA model that beats closed-source giants on embodied reasoning tasks could finally make real-world robot deployment practical.

Adversarial attacks on speech models leave tell-tale geometric fingerprints in early representation layers that can be detected without transcripts.

Surprisal theory's reliance on arbitrary tokenization schemes undermines its validity, but this framework offers a way to fix it.

Finally, voice anonymization offers a smooth, tunable knob to balance privacy and prosody, instead of forcing you to pick just one.

Stop hand-tuning kernels for Koopman operator approximation: this dictionary learning approach automatically discovers optimal kernel parameters, simplifying kEDMD.

SER's noble aspirations of voice-activated healthcare are undermined by datasets that bear little resemblance to real-world emotional expression.