Research Feed Page 15

The paper introduces UMPIRE-Net, an accelerated MRI reconstruction method that independently regularizes magnitude and sign components to improve performance in partial Fourier imaging.

The paper introduces a framework to ensure safe robotic operation in complex urban environments by defining a dynamic safety envelope rather than using static constraints.

The researchers replaced black box Neural Architecture Search with a structured Linear Programming framework to improve both efficiency and accuracy.

MathForm improves mathematical autoformalization by integrating external retrieval and verification-guided feedback loops to generate verified Lean 4 code.

The paper introduces a framework to improve the auditability and calibration of LLM-based judges used for verifying compliance with principle-based financial regulations.

The paper introduces a method using synthetic data and reinforcement learning to help language model agents explore proactive action strategies instead of falling into rigid behavioral patterns.

The paper introduces a reader-centered evaluation framework called information satisfaction to determine if summaries meet the specific requirements of a target persona.

The paper explores the development of autonomous AI systems capable of performing scientific research by integrating neural learning, robotics, and formal reasoning.

The researchers developed a method called NAS-NGMA to automatically identify efficient and high-performing neural network architectures by using differentiable optimization for gating and activation functions.

CForce improves parallel text generation in diffusion large language models by using later decoding stages to stabilize predictions in earlier stages.

Researchers audited seven large language models to determine how they weigh physician reputation and demographic factors when recommending doctors to patients.

The paper introduces TripWire, a method that identifies and activates safety-specific neurons to trigger refusal behavior against jailbreak attacks while preserving model utility.

The paper introduces the ALD/E-ImageMiner benchmark to improve how AI models interpret and analyze complex scientific figures and tables.

CRAFT enables subject-driven image personalization using only reference-side supervision, completely eliminating the need for costly multi-stage curation pipelines that generate paired reference and composed target data.

The paper examines how to effectively repeat high-quality domain data during the pretraining phase as model sizes and training token budgets scale.

The paper introduces CalibRank, an objective function that prevents time-series foundation models from producing flat, ineffective stock return predictions.

The paper introduces ATLAS, a system that records and abstracts LLM agent interactions into probabilistic finite-state Markov chains to make their decision-making strategies human-interpretable and useful for downstream tasks.

ScienceFlow uses state management and evidence-aware execution to maintain stable workflows for complex machine learning research over long time horizons.

The paper introduces a feedback loop between the tracking controller and trajectory planner that adjusts spatial constraints to prevent sub-optimal performance caused by model mismatches.

ForgeWM is a progressive training framework that converts action-conditioned video generators into efficient few-step world models for interactive game environments.