Entropy Coding

Entropy coding turns probability estimates for symbols into compact bitstreams, making it a core component of lossless compression and learned image, video, audio, and 3D codecs. Methods range from Huffman, arithmetic, and range coding to rANS and learned context models that predict symbols from nearby data, prior frames, or neural representations. Current work combines these coders with neural predictors, quantization, and hierarchical or progressive representations to compress text, model weights, and visual data. A central tension is balancing compression efficiency against decoding speed, memory access, parallelism, and exact reproducibility across hardware.