Skip to content

Compression

Prime #
158
Origin domain
Information Theory
Also from
Computer Science & Software Engineering, Statistics & Experimental Design, Cognitive Science
Aliases
Data Compression, Encoding Reduction, Redundancy Removal
Related primes
Entropy (Thermodynamic Sense), information, Redundancy, shannon limit, Abstraction

Core Idea

Compression is the process of reducing the amount of information, space, or effort required to represent or transmit something while preserving its essential meaning or function.

How would you explain it like I'm…

Making things smaller

Imagine writing 'AAAAA' instead as 'five A's.' That is shorter but says the same thing. Compression is squishing a message into fewer letters or bits by spotting parts that repeat. Computers do this so songs, pictures, and games fit on your phone and load fast.

Squishing information

Compression is encoding information using fewer symbols than the original, by spotting patterns and redundancy. If a letter shows up a lot, you can give it a shorter code; if pixels in a photo are nearly the same, you can describe a whole region at once. Lossless compression lets you rebuild the original exactly, like ZIP files. Lossy compression throws away tiny details you would not notice, like JPEGs and MP3s, so you can shrink things much more.

Shrinking data without losing it

Compression replaces a representation of information with a shorter one by exploiting redundancy: statistical regularity (some symbols are more common), structural predictability (patterns repeat), or perceptual unimportance (humans cannot detect some details). Lossless schemes let you reconstruct the original exactly and are bounded below by the source's Shannon entropy — you literally cannot beat that limit without losing information. Lossy schemes accept controlled errors in exchange for much smaller sizes, trading off distortion against rate. Every concrete method picks a source model, a loss discipline, an algorithm (like Huffman codes or LZ-style dictionaries), and a use context such as storage versus streaming.

 

Compression is the encoding of information in a representation shorter than the original, exploiting redundancy — statistical regularity, structural predictability, or perceptual unimportance — to reduce the symbols, bits, or physical resources needed to store or transmit it. It comes in two disciplines: lossless, which guarantees exact reconstruction and is bounded below by the source entropy (the Shannon limit, a hard floor no lossless code can beat), and lossy, which accepts controlled approximation in exchange for far greater reduction, governed by rate-distortion theory. Any concrete compressor is specified by four choices: a source model (text, image, audio, video, scientific data, code) with its statistical properties; a loss discipline; an algorithm family (entropy coding such as Huffman or arithmetic, dictionary methods like LZ77, transform coding like DCT or wavelets, predictive or neural coding); and a use context (one-shot vs streaming, latency-sensitive vs bandwidth-sensitive). The field rests on Shannon's 1948 information theory and the long sequence of algorithmic refinements since.

Broad Use

  • Computing: Data compression (JPEG, ZIP, MP3).

  • Cognitive Science: Chunking in memory, where humans store patterns instead of individual details.

  • Biology: Genetic encoding, where DNA stores massive biological information in a compact form.

  • Economics: Cost-cutting strategies (e.g., streamlining supply chains to minimize waste).

  • Education: Summarization techniques for textbooks and lecture materials.

Clarity

Identifies ways to simplify representations without losing meaning.

Manages Complexity

Provides strategies to reduce cognitive, computational, or logistical burden.

Abstract Reasoning

Encourages pattern recognition and efficient information encoding.

Knowledge Transfer

The principle of eliminating redundancy while keeping meaning intact applies in computing, communication, learning, and engineering.

Example

High-speed language interpreters mentally "compress" complex grammar rules into intuitive patterns to process speech in real time.

Relationships to Other Primes

One-hop neighborhood: parents above, mutual partners to the right, children below.Compressionsubsumption: AggregationAggregationsubsumption: OptimizationOptimizationsubsumption: AbstractionAbstractionsubsumption: ChunkingChunkingsubsumption: Dimensionality ReductionDimensionalityReductioncomposition: Predictive CodingPredictiveCoding

Parents (3) — more general patterns this builds on

  • Compression is a kind of Abstraction — Compression is a specialization of abstraction in which the retained structure is information-theoretic regularity and the discarded structure is the redundancy.
  • Compression is a kind of Aggregation — Compression is a kind of aggregation: it collapses redundant detail into a unified shorter representation while retaining chosen structure.
  • Compression is a kind of Optimization — Compression is a kind of optimization: it minimizes representation length subject to a reconstruction-fidelity constraint.

Children (3) — more specific cases that build on this

  • Chunking is a kind of Compression — Chunking is a specialization of compression in which a set of items is grouped into a single meaningful unit that working memory then tracks as one element.
  • Dimensionality Reduction is a kind of Compression — Dimensionality reduction is a specialization of compression in which redundancy in a high-dimensional representation is removed by projecting onto a lower-dimensional latent structure.
  • Predictive Coding presupposes Compression — Predictive coding presupposes compression because transmitting only the prediction error exploits the predictable signal's redundancy to shorten its representation.

Path to root: Compression → Aggregation

Not to Be Confused With

  • Compression is not Dimensionality Reduction because Compression reduces representation size while preserving all information (lossless case) or achieving controlled fidelity loss, while Dimensionality Reduction deliberately discards low-information dimensions to make high-dimensional data tractable, accepting information loss as intentional.
  • Compression is not Chunking because Compression is an information-theoretic encoding that reduces bit-length of a representation, while Chunking is a cognitive process that reduces the number of mental units tracked, operating in an entirely different substrate (cognition vs. information).
  • Compression is not Representation because Representation is the faithful mapping of a target system onto a medium preserving selected structure, while Compression is the reduction of representation size by exploiting redundancy, often accepting some information loss in the lossy case.
  • Compression is not Entropy (Thermodynamic Sense) because Entropy quantifies the number of accessible microstates consistent with a macrostate (a measure of possibility), while Compression exploits statistical regularity and structure in data to reduce encoding length (a measure of economy).