Unified Line and Paragraph Detection by Graph Convolutional Networks (2022) (arxiv.org)

A new open, living textbook titled "The Beginner's Textbook for Fully Homomorphic Encryption" (arXiv, Ronny Ko) provides a practitioner-friendly introduction to fully homomorphic encryption (FHE), the class of schemes that allow computation directly on ciphertexts so decrypted results match plaintext operations. The draft walks through core mechanics—support for encrypted addition and multiplication and how those primitives compose to build subtraction, division, Boolean logic (AND/OR/XOR/NAND/MUX) and higher-level functions such as ReLU, sigmoid and trigonometric functions—implemented either exactly or via approximations to trade accuracy for efficiency. The submission is actively maintained with demos, code and supplementary material, and the authors invite collaborators to expand the draft. For the AI/ML community this is a practical resource that lowers the barrier to building privacy-preserving inference and analytics: FHE lets servers run models on encrypted inputs without learning either features or outputs, enabling confidential ML-as-a-service, encrypted database queries, private search, confidential smart contracts and secure outsourcing. By focusing on both the algebraic building blocks and the approximation/efficiency trade-offs relevant to ML (e.g., implementing activation functions under noise and circuit-depth constraints), the book is useful for researchers and engineers designing FHE-aware models and production pipelines.