Google Research has introduced a neural network that flags signs of hypertension using only the pulse and motion sensors built into ordinary smartwatches, with no cuff calibration required. Trained via self-supervised learning on nearly 500,000 hours of sensor data, the model matched traditional cuffs in a 196-person study across five US sites. The method is now folded into Google Health's blood pressure trends feature but remains a research project, not a diagnosis.
Contrastive Learning
对比学习
A method that learns representations by pulling similar samples closer and pushing dissimilar ones apart.
Curated Intelligence
Google has open-sourced EmbeddingGemma 2, a unified multimodal embedding model supporting text, code, images, audio, and video. With only 744M parameters, it outperforms models 2-12x larger while requiring just ~567MB RAM on a Pixel 11 Pro, enabling private on-device search.
Kakao Brain announced the release of new ViT and ALIGN models. Without specific parameters, performance metrics, or technical details, the actual impact cannot be assessed.
OpenAI released contrastive pre-trained embeddings for text and code to enhance semantic retrieval and understanding. The study leverages large-scale unsupervised training to optimize vector representations, supporting applications like RAG and code search.