A new multi-agent motion prediction method using mode-world weighted regression loss and an iterative decoder has achieved first place in the Argoverse 2 benchmark, promising to improve autonomous veh…
Researchers developed an AI-driven MRI protocol that cuts scan time from 27 to 14 minutes using an explainable AI framework with XGBoost, SHAP, and Recursive Feature Elimination, reducing features to …
Researchers developed a Tri-Branch Modular Fusion Neural Network that integrates 3D spatial geometry, discrete topological grammar, and macroscopic physicochemical descriptors to predict molecular pro…
Researchers developed FedPPO-PG, a federated multi-agent reinforcement learning framework for smart grid transient stability control. In tests on the IEEE 39-bus system, it achieved 100% stabilization…
New research introduces 'reasoning consistency scanning' to detect inconsistencies between AI models' stated logic and their answers, without requiring internal model access. The method, tested on saf…
Researchers introduced HALE, a framework that combines agent-based modeling with large language models to predict human decision-making in real-time, demonstrated by simulating COVID-19 in Salt Lake C…
A new AI governance method combines participatory evaluation, expert cost analysis, and genetic algorithm-based simulations to identify effective policy combinations for mitigating AI harms. The appro…
A study of neural-interface models reveals a critical gap between formal robustness certificates and real-world safety, with EEGNet's classification accuracy dropping 25.7% under attack while certific…
Researchers introduced VTC, a tensor compilation framework that eliminates unnecessary data movement in deep neural networks by using virtual tensors and index mappings. VTC outperforms existing ML co…
Researchers combined SageMath with large language models in a ReAct-style setup, achieving a 9.7 percentage point average improvement on the RealMath benchmark. GPT-5.5 reached a 75.2% solve rate with…
Researchers have developed a new approach to overcome the quadratic cost of causal self-attention in long-context transformers, using state update design and structural interventions like sink tokens …
New advancements in Vision Language Action (VLA) models are achieving 39 times faster inference speeds than OpenVLA and 46 Hz throughput on edge platforms, reducing latency and costs through fewer act…
Compass, a new AI methodology for prostate cancer detection, analyzes 2D micro-ultrasound studies using a multi-view approach that mimics clinical practice by integrating rotational video sweeps, outp…
Palantir's forward deployed engineering (FDE) methodology, which embeds engineers directly with clients, is setting a standard for AI deployment success. Other companies are racing to adopt similar st…
Tencent is preparing to acquire a majority stake in AI agent startup Manus at a $2 billion valuation, after Beijing blocked Meta's planned acquisition. The move signals Tencent's ambition to enhance i…
Researchers have developed a new compression algorithm for decentralized AI training that achieves up to 99% compression without degrading convergence, enabling billion-parameter models to be trained …
Researchers introduced Refine Thought (RT), a method that enhances text embedding models by executing multiple forward passes to improve semantic reasoning. RT significantly boosted performance on spe…
Researchers studying neural networks on modular arithmetic found that embedding compression lags behind generalization by at least 10,000 steps, challenging traditional convergence metrics. The grokki…
DeepSeek V3.2 achieves a 67.25% pass rate on the ARC-AGI-1 challenge using a novel Explorer-Definer Pipeline and Reflective Orchestrator, without fine-tuning or heavy compute. The open-weight model's …
AgentLens, a new benchmark for code agents, evaluates user experience and trajectory rather than just pass/fail metrics. Released as an open-source tool on GitHub, it combines formal verification with…