Explore our latest peer-reviewed papers, technical reports, and open-source contributions advancing the field of artificial intelligence.
We propose a lightweight transformer variant optimized for edge deployment that achieves 99.2% F1-score in detecting sensor anomalies across manufacturing pipelines. Our approach reduces latency by 60% compared to baseline RNN models.
A comprehensive study on tensor and pipeline parallelism strategies for training 70B+ parameter models. We introduce a novel communication-aware sharding technique that improves GPU utilization by 22%.
We present a governance-ready framework for auditing model bias, ensuring transparency, and maintaining compliance across regulated industries. Includes standardized evaluation metrics and automated bias detection pipelines.
Leveraging vision-language models for rare disease detection without labeled training data. Achieves diagnostic accuracy comparable to supervised baselines on 15 rare pathology categories.
An empirical comparison of 4-bit quantization and knowledge distillation techniques for deploying large language models on consumer hardware. Includes benchmark results across latency, memory, and accuracy trade-offs.
We introduce a multi-agent RL framework that dynamically optimizes logistics routing under uncertainty. Tested across global shipping networks, reducing fuel consumption by 18% while maintaining SLA compliance.
Try adjusting your search terms or filters to discover more research.