AI's role in the workforce is shifting from automation to collaboration, with humans managing AI agents to maximize economic value. The integration requires algorithmic coordination and cooperation to…
Researchers have found that large language models (LLMs) can generate diverse software code, potentially reducing common-mode failures, but reliability improvements depend on programming language and …
Decentralized federated learning, promoted for privacy and efficiency, faces significant challenges from network inhomogeneities that slow model convergence, according to recent research. Real-world t…
PixCon, a new framework for semi-supervised semantic segmentation, guarantees contamination-free pseudo-labeling by using a per-class memory bank that only admits correctly classified labeled pixels. …
Researchers introduced Subspace-Aligned Rewiring (SAR), a post-training editing method for large language models that isolates reasoning-effective components in spectral space, preserving over 99% of …
Researchers have developed CONFLUX, a 3D medical imaging model combining a latent diffusion model with reinforcement learning to generate chest CT scans with unprecedented precision. The model outperf…
New research shows that 12-31% of weights in sparse transformers can be clearly interpreted, compared to dense models. An automated pipeline generates human-readable descriptions for each weight, test…
AI systems are increasingly being used in foreign policy decisions, raising urgent governance challenges due to the high stakes and complexity of statecraft. Researchers call for a new framework that …
Researchers introduced PRECEDE, a precedent-guided AI approach for drug redesign that mitigates side effects while preserving therapeutic benefits, using large language models coordinated with human o…
Researchers introduced PROMPTPET, an LLM-based agent that uses a reinforcement-learning-inspired rule optimizer to dynamically obfuscate sensitive user data in chatbot prompts, achieving the best priv…
Researchers developed VISTA, an auditing tool that detects biases in vision-language models by coupling semantic entropy with divergence analysis. In tests, VISTA identified 142 suspicious cases acros…
Researchers introduced Hierarchical Representation Regularization (HiR²), a method that improves large multimodal models' ability to understand visual hierarchies by enforcing taxonomic structures thr…
Researchers introduced HetDPT, a depth pruning method for Vision Transformers that speeds up DeiT-B by 1.58x and DeiT-S by 1.39x with minimal accuracy loss, and combined with width pruning (HetDPT+) a…
A new study finds that a plain, fully fine-tuned RoBERTa model matches or surpasses specialized AI-generated text detectors across benchmarks, challenging the industry's focus on architectural complex…
Researchers have introduced Threshold Gating, a new activation method for neural networks that unifies traditional functions like ReLU and Sigmoid under a single framework. The approach allows pre-tra…
VideoSearcher, a new framework for video understanding, extends Vision-Language Models with multi-tool reasoning to dynamically ground visual clues in video content. It introduces the Bi-branch Sequen…
AI agents are being evaluated for personal negotiations, but a new benchmark called SovereignNegotiation-Bench reveals that high agreement rates often come at the cost of user privacy and utility. The…
Researchers introduced Lagrangian Reward Augmentation (LARA), a framework that steers frozen language models using safety constraints during inference without repeated weight updates. LARA improves th…
Researchers demonstrated domain-specific safety adaptation in large language models, particularly in cybersecurity, using the Kimi K2 model. The study found that model architecture and safety training…
Researchers introduced K9-Bench, a benchmark evaluating AI's ability to understand canine behavior through 5,000 question-answer pairs across 907 videos. Current AI models struggle with subtle animal …