DR^{3}-Eval: Towards Realistic and Reproducible Deep Research Evaluation

▲ 29 🎓 Nanjing University

• DR$^{3}$-Eval is a benchmark for evaluating deep research agents on multimodal, multi-file report generation, featuring a realistic simulation of web

RAD-2: Scaling Reinforcement Learning in a Generator-Discriminator Framework

▲ 25 🎓 Huazhong University of Science & Technology

• A unified generator-discriminator framework for autonomous driving motion planning that improves stability and performance through diffusion-based tra

How to Fine-Tune a Reasoning Model? A Teacher-Student Cooperation Framework to Synthesize

▲ 24 🏢 Shanghai AI Laboratory

• Teacher-student cooperation data synthesis framework addresses stylistic divergence in synthetic data for improved model fine-tuning performance.

GlobalSplat: Efficient Feed-Forward 3D Gaussian Splatting via Global Scene Tokens

▲ 20 🎓 The Hebrew University of Jerusalem

• GlobalSplat introduces a global scene representation framework that achieves compact, consistent 3D Gaussian splatting with reduced computational over

ASGuard: Activation-Scaling Guard to Mitigate Targeted Jailbreaking Attack

▲ 18 🎓 Korea University

• Activation-Scaling Guard (ASGuard) mitigates brittle refusal behaviors in large language models by identifying and recalibrating specific attention he

Dive into Claude Code: The Design Space of Today's and Future AI Agent Systems

▲ 16 🎓 Mohammed Bin Zayed University of Artificial Intelligence

• The study analyzes Claude Code's architecture, identifying five motivating human values and tracing them through thirteen design principles to specifi

Vision-Based Safe Human-Robot Collaboration with Uncertainty Guarantees

🏛️ Stanford University AI & Machine Learning

• Ensuring safety in human-robot collaboration is challenging due to uncertainties in human motion prediction.
• A framework for vision-based human pose estimation and motion prediction with conformal prediction guarantees.

TokenGS: Decoupling 3D Gaussian Prediction from Pixels with Learnable Tokens

🏛️ NVIDIA AI & Machine Learning

• Current methods for 3D Gaussian prediction often struggle with pixel dependency and lack flexibility.
• A novel approach that decouples 3D Gaussian prediction from pixels using learnable tokens.

RadAgent: A tool-using AI agent for stepwise interpretation of chest computed tomography

🏛️ ETH Zurich AI & Machine Learning

• The interpretation of complex medical imaging, such as CT scans, often lacks systematic and stepwise methodologies.
• A tool-using AI agent designed for stepwise interpretation of chest computed tomography.

LeapAlign: Post-Training Flow Matching Models at Any Generation Step by Building Two-Step

🏛️ ByteDance AI & Machine Learning

• Aligning flow matching models with human preferences in generative tasks is often inefficient and complex.
• A method for post-training flow matching that builds two-step trajectories for better alignment.

Expanding into Reality: Random Graphs for Datacenter Networks

🏛️ Amazon Web Services Systems & Infrastructure

• Traditional datacenter network topologies often struggle with cost and fault tolerance issues.
• The paper presents a novel datacenter fabric design based on random graphs, optimizing for both cost and fault tolerance

Combinatorial Contracts Through Demand Types

🏛️ University of Oxford Theory & Algorithms

• Designing effective contracts in combinatorial settings is complex due to the diverse actions and incentives involved.
• A combinatorial action model that utilizes demand types to create linear contracts incentivizing specific actions.

Efficient Retrieval Scaling with Hierarchical Indexing for Large Scale Recommendation

🏛️ Meta Applications

• The increasing volume of data and complexity in recommendation systems makes efficient retrieval challenging.
• A hierarchical indexing approach to scale retrieval for large-scale recommendation tasks.

Robust Optimal Experimental Design Accounting for Sensor Failure

🏛️ Duke University Other CS

• Sensors often fail during experiments, complicating optimal experimental design for vibration analysis.
• A robust optimal experimental design method that accounts for sensor failures.

Tweedie Calculus

🏛️ Massachusetts Institute of Technology Economics

• Measurement-error analysis and empirical Bayes methods often struggle with recovering posterior means from observed marg
• The paper introduces a novel application of Tweedie's formula to directly recover posterior means in the Gaussian locati