The process of encoding information using fewer bits than an original representation by removing redundancy or exploiting statistical or structural regularities, producing either recoverable (lossless) or approximate (lossy) representations.

Principle

Model the source or data structure to identify predictable or redundant parts, then map sequences to codewords whose length reflects surprisal (entropy). Lossless schemes preserve exact recoverability; lossy schemes trade fidelity for higher compression by discarding information deemed perceptually or task‑irrelevant.

Demonstration

A text file with repeated words is encoded with Huffman coding (or dictionary methods like LZW): commonly occurring tokens receive shorter codewords, producing a shorter bitstream that can be reconstructed exactly (lossless). An image is transformed and quantized in JPEG: small coefficients are dropped to reduce size with acceptable visual loss (lossy).

Misapplication

Attempting to compress data that is already near‑maximally entropic (e.g., encrypted or random data) yields little or no reduction and may increase size. Using lossy compression where bit‑exact reconstruction is required (e.g., legal records) causes unacceptable data loss.

Consequence

Reduces storage and transmission costs and can enable new uses (streaming, archival). Compression introduces trade‑offs: compute cost, latency, potential quality loss, and sensitivity to model mismatch; tight lower bounds exist (entropy) that prevent arbitrary compression.

Reversal

Data expansion or encryption that produces high‑entropy outputs: without exploiting redundancy, representations are longer or intentionally indistinguishable from random, preventing size reduction and complicating subsequent analysis.

Boundary

Applies to representations of information (text, images, audio, structured data). The effectiveness depends on source statistics and chosen distortion measure; fundamental limits are set by information theory (entropy) and specific fidelity requirements. Compression is distinct from encryption (aiming at confidentiality) though outputs can appear similar.

Semantic Tension

Tension between lossless and lossy objectives: lossless preserves every bit but achieves lower compression ratios; lossy attains higher ratios at the cost of fidelity. Also tension between universal compressors (no model) and model‑based compressors (higher efficiency when the model fits).

Synthesis

Data compression builds a compact representation by exploiting predictable or redundant structure in data—selecting either reversible encodings to preserve all information or irreversible approximations to trade some fidelity for reduced size—subject to theoretical entropy limits and practical fidelity constraints.