Google secures grants for reinforcement learning and quantum system evaluation while publishing on sparse attention and model architecture.
The filing
On November 10, 2026, Google LLC received US grants for methods performing reinforcement learning with centralized inference and training, deterministic training of machine learning models, multi-modal neural network architecture search, privacy maintenance during condition attribution, knowledge graph completion and multi-hop reasoning at scale, systems for evaluating the scalability of quantum computing systems, and learning token importance using multi-model stochastic sparsity inducing regularization (USPTO 12,792,905 (opens in a new tab), USPTO 12,792,919 (opens in a new tab), USPTO 12,792,920 (opens in a new tab), USPTO 12,792,926 (opens in a new tab), USPTO 12,792,932 (opens in a new tab), USPTO 12,792,943 (opens in a new tab), USPTO 12,792,960 (opens in a new tab)). On November 11, 2026, the company published European applications for handling component failures in surface code circuits and variational quantization models for efficient evaluation and content creation that use Gaussian Mixture models to compute predicted quality metrics for asset and layout combinations (EP4837205A2 (opens in a new tab), EP4838447A1 (opens in a new tab)). On November 12, 2026, Google published US applications for full attention with sparse computation cost, adaptive large language model architecture based on system-on-chip temperature and power, and decentralized learning of large machine learning models, alongside a PCT application for managing performance monitoring and reporting for measurement prediction in wireless networks (USPTO 20260359561A1 (opens in a new tab), USPTO 20260359663A1 (opens in a new tab), USPTO 20260360353A1 (opens in a new tab), WO2026232385A1 (opens in a new tab)).
This filing’s full text wasn’t read for this issue, so its claims and the applicant’s background aren’t shown. Both are in the record, under Sources.