Titans: neural memory that changes during inference
A technical reading of Titans v1: associative memory, surprise, forgetting, MAC/MAG/MAL architectures and the limits of the reported long-context results.
Read articleAnalysis of IP practice, AI technology and international protection, alongside news from IPLEX. Explore the articles in English.
537 articles and updates
103 articles
A technical reading of Titans v1: associative memory, surprise, forgetting, MAC/MAG/MAL architectures and the limits of the reported long-context results.
Read articleA technical reading of US12361264B2: selecting layers, retaining intermediate features and retraining locally, with the claimed LLM configuration distinguished from an illustrative inspection camera.
Read articleHow the ICML 2023 method combines drafting, parallel target-model evaluation and probability correction, and what its T5-XXL experiments do and do not establish.
Read articleA technical reading of US12639823B2: semantic segmentation, region-specific trimap widths, face protection and per-pixel alpha estimation.
Read articleA technical reading of US12579565B1: two-stage language-model training, review ranking signals, product attributes and the context used to generate a review overview.
Read articleAn analysis of LSA’s learned clustering, Flat and Tree search, distilled and native training, and the conditions behind its accuracy, speed and memory results.
Read articleHow LG’s AI refrigerator patent connects capture timing, depth-based separation, food detection and identification, inventory updates and user corrections.
Read articleHow LoRA freezes a pretrained model and learns small update matrices, what the original GPT-2 experiment shows, and which training, deployment and patent-practice distinctions matter.
Read articleHow Alibaba’s published application US20250173522A1 combines domain vocabulary, feature vectors and corpus filtering, and how this differs from using Qwen-MT translation controls.
Read articleHow Switching Linear Attention combines linear regressors within a fixed recurrent state, what its recall experiments establish, and where capacity and computational limits remain.
Read articleUS12094447B2 explains how phoneme or character information can be combined with word embeddings and grammatical structure to generate speech.
Read articleA technical and patent-practice reading of gated bidirectional linear attention, its hybrid encoder and the limits of the reported speed and retrieval results.
Read article