Уровень 0 · материалов: 3
В кластер входят документы о конкретных технологических достижениях Google в сфере глубокого обучения и нейросетей, но не общие обзоры рынка ИИ.
Общие признаки: Google, нейронные сети, масштабируемость ИИ, автоматизация проектирования моделей
Группа выше: Технологии Google: ИИ и робототехника
Смысл: The text describes Google's successful deployment of a massive self-learning neural network to improve practical services like speech recognition and image processing, marking a significant leap in AI scalability.
Google deployed a massive 1,000-computer neural network that significantly improved speech recognition and image identification, mirroring mammalian visual processing.
Смысл: The text describes the success of Google's AutoML project in creating NASNet, a neural network designed by another AI. This AI-designed model outperformed human-expert designs in image classification and object detection tasks across various scales, including mobile platforms.
Google's AutoML system designed a new neural network called NASNet that surpasses human-expert models in image recognition and mobile efficiency.
Смысл: The text explains how Google's transition to a fully neural machine translation system allows for a universal internal representation of meanings, enabling higher quality translations and the ability to translate between untrained language pairs (Zero-Shot Translation).
Google's GNMT uses a deep learning 'universal meaning base' to improve translation quality and enable translation between languages it wasn't specifically trained to pair.