Skip to content

Curse Of Dimensionality

Prime #
770
Origin domain
Data Science & Analytics
Subdomain
high dimensional methods → Data Science & Analytics
Aliases
Bellmans Curse of Dimensionality

Core Idea

The curse of dimensionality is the structural pattern by which the volume of a space, the sparsity of any finite sample within it, and the distortion of common intuitions about distance, neighbourhood, and concentration all grow so fast with the number of dimensions that methods which work fluently in low dimensions break down qualitatively — not merely quantitatively — in high ones. The number of cells in a regular partition of a unit hypercube grows exponentially in the dimension. The fraction of a high-dimensional ball's volume lying near its surface approaches one. The ratio between the nearest and farthest distances among random points approaches one, so "nearest neighbour" loses meaning. The volume of an inscribed unit ball goes to zero. Any fixed-size sample becomes exponentially sparse with respect to a regular grid. These are not separate facts but consequences of one underlying geometry: exponential volume growth against non-exponential coverage.

Five structural commitments make the pattern portable. There is a space of candidate states indexed by some number of independent dimensions. There is a sample or search budget that does not grow exponentially with dimension. There are methods — estimation, optimisation, search, classification, governance, design — that depend implicitly on local density of sample points or on meaningful nearness. As dimension grows, the budget-to-space ratio collapses, and local density and meaningful nearness both go to zero. And the methods fail qualitatively at a threshold, not gracefully: a small further increase produces a regime where the old intuitions no longer apply at all, where distances cease to discriminate, sample coverage becomes asymptotically empty, and naive interpolation, search, or governance becomes infeasible. The curse is the prediction that this regime change is generic — it arrives whenever exponential space meets sub-exponential budget — and that the only escapes are structural assumptions that reduce the effective dimension below the nominal one.

How would you explain it like I'm…

Too Many Rooms

Imagine hide-and-seek in one hallway: easy to check every spot. Now imagine a house with more and more rooms, each opening into more rooms, until there are way more hiding spots than you could ever peek into. Suddenly everywhere feels equally far away and equally empty, and your few peeks barely cover anything.

Too Many Directions

When you describe something using just a couple of numbers, the space of possibilities is small and easy to fill with examples. But each new number you add multiplies how many possibilities there are, so the space balloons unbelievably fast. Your collection of examples stays the same size, so it gets stretched thinner and thinner until almost everywhere is empty. Weirder still, in these huge spaces everything ends up about equally far from everything else, so 'which point is nearest?' stops giving a useful answer. That's why tricks that work great with two or three numbers can totally break once you have hundreds.

When Nearness Stops Meaning

Each independent feature you measure adds a 'dimension,' and the size of the space of possibilities multiplies with every one you add — doubling the dimensions doesn't double the space, it squares or worse. Your sample of data points, however, grows only at a normal pace, so it gets spread impossibly thin: any fixed-size sample becomes a few specks scattered through an enormous void. Two strange things follow. Almost all of a high-dimensional ball's volume sits in a thin shell near its surface, and the distance to your nearest point and your farthest point become almost equal. So tools that rely on 'points near each other are similar' — nearest-neighbor search, density estimates, interpolation — quietly stop working. The failure is qualitative, a regime change, not just a slowdown.

 

The curse of dimensionality is the family of geometric facts that make low-dimensional intuitions fail catastrophically as the number of dimensions grows. The core driver is exponential volume growth: a regular grid over a unit hypercube has a number of cells that grows exponentially in the dimension, so any sample or search budget that grows sub-exponentially covers an asymptotically vanishing fraction of the space — the sample becomes exponentially sparse. Geometry distorts in tandem: the fraction of a hyperball's volume within a thin surface shell approaches one, the volume of a unit ball inscribed in the unit cube goes to zero, and the ratio of nearest to farthest pairwise distances among random points approaches one, so 'nearest neighbor' loses discriminating power. These are not independent curiosities but consequences of one structure — exponential space versus non-exponential coverage. Methods that implicitly assume local sample density or meaningful nearness (estimation, optimization, classification, search) therefore break down at a threshold rather than gracefully. The standard escape is to reduce the EFFECTIVE dimension — exploiting low-dimensional manifolds, sparsity, or structural assumptions — below the nominal one.

Structural Signature

the candidate-state space indexed by independent dimensions — the sub-exponential sample/search budget — the methods depending on local density or meaningful nearness — the collapsing budget-to-volume ratio — the qualitative threshold failure — the effective-dimension-reduction escape

A situation exhibits the curse-of-dimensionality pattern when each of the following holds:

  • A dimensioned state space. A space of candidate states is indexed by some number of independent degrees of freedom; its volume grows exponentially in that count.
  • A sub-exponential budget. A finite sample or search budget is available that does not grow exponentially with dimension — typically fixed, linear, or polynomial.
  • Density- or nearness-dependent methods. The operative methods (estimation, search, optimisation, classification, design, governance) rely implicitly on local density of sample points or on distances that meaningfully discriminate near from far.
  • A collapsing coverage ratio. As dimension rises, the budget-to-volume ratio collapses: any finite sample becomes exponentially sparse, near and far distances converge, and edge volume dominates — so local density and meaningful nearness both tend to zero.
  • A qualitative threshold failure. The methods fail not gracefully but at a threshold: a small further increase in dimension produces a regime where the old low-dimensional intuitions no longer apply at all.
  • The effective-dimension escape. The only remedies are structural assumptions — sparsity, manifolds, factor models, embeddings, modularisation — that lower the effective dimension below the nominal one.

The components compose one geometric fact wearing many costumes: exponential volume against sub-exponential coverage. Diagnosing a dimensional collapse therefore means counting degrees of freedom against budget and reaching for an effective-dimension reduction rather than attributing the failure to data quality or tuning.

What It Is Not

  • Not dimensionality reduction. dimensionality_reduction (the nearest neighbour) is one of the escapes from the curse — a remedy that lowers effective dimension. The curse is the problem: the geometric pathology that makes the escape necessary. The two are diagnosis and treatment, not the same object.
  • Not dimensional analysis. dimensional_analysis reasons about physical units and their consistency to derive scaling laws; the curse concerns degrees of freedom of a state space and how volume grows with their count. The shared word "dimension" hides unrelated structures.
  • Not a scalability limit. scalability is about whether a system handles growing load gracefully; the curse is a specific geometric failure where added dimensions make low-dimensional methods break down qualitatively. Not every scaling wall is the curse.
  • Not commensurability. commensurability concerns whether two quantities share a common measure; the curse concerns the volume-versus-coverage mismatch of a many-dimensioned space. Adding dimensions is not the same as failing to compare across them.
  • Not generic time/space complexity. complexity_time_space measures an algorithm's resource growth; the curse is the deeper geometric reason that coverage of a space (statistical, not just computational) collapses — it has statistical and geometric faces, not only a computational one.
  • Common misclassification. Diagnosing a curse where the dimensions are heavily correlated. Exponential volume growth presumes independent degrees of freedom; twenty correlated features may span three effective directions, and counting raw columns as if independent manufactures a curse that barely exists.

Broad Use

In statistics and machine learning, the origin substrate, kernel density estimation, nearest-neighbour methods, decision trees, and uniform sampling all degrade exponentially with feature count; Bellman's recognition of the problem in dynamic programming motivated whole families of mitigation — regularisation, structured priors, low-dimensional embeddings. In numerical analysis, deterministic quadrature scales exponentially in dimension, which is precisely why Monte Carlo and quasi-Monte Carlo methods exist. In combinatorial optimisation and operations research, the search space of a multi-attribute decision grows exponentially in attribute count, defeating brute force. In drug discovery and molecular design, chemical space is so enormous that any finite library covers a vanishing fraction, motivating active learning and pharmacophore-restricted search. In governance and policy design, a policy with many parameters, target groups, and timing variables defines a high-dimensional design space that any finite consultation or sensitivity analysis covers exponentially less of as the parameter count rises — so "comprehensive review" becomes structurally infeasible past a point. In experimental design, factorial designs with many factors require exponentially many runs, which is why fractional factorial and response-surface methods exist. In engineering design, a product with many parameters cannot be exhaustively explored, motivating standard-component libraries and modularisation. And in high-throughput biology, the joint configuration space of many genes or interactions exceeds any feasible sample, forcing sparsity priors.

Clarity

Naming the curse forces a single sharpening question: how does the cost of my method scale with the number of independent dimensions I am reasoning over? That question surfaces the otherwise-hidden gap between exponential space growth and non-exponential budget growth, and it explains why a method that works fluently with two or three variables fails unexpectedly with twenty. The failure is not a tuning problem to be patched with more data or a better learning rate; it is a qualitative regime change.

The frame also separates three failure types that are easy to lump together. There is a statistical problem — sparsity and inadequate sample coverage. There is a computational problem — enumeration and search cost. And there is a geometric problem — distances, neighbourhoods, and densities becoming uninformative. All three are downstream of the same exponential volume growth, but they call for different responses, and recognising which one is biting in a given case is the diagnostic the prime supplies. Clarity here means refusing to attribute a dimensional collapse to data quality or model choice when the true cause is the geometry of the space itself.

Manages Complexity

The pattern compresses a wide family of failure modes — sparsity, edge concentration, neighbourhood degeneration, integration intractability, search intractability — into a single explanation: the budget grows polynomially or linearly while the space grows exponentially. That one explanation subsumes phenomena that look unrelated on the surface, from the breakdown of quadrature to the meaninglessness of nearest-neighbour retrieval in raw embeddings.

The corresponding intervention vocabulary is equally unified, which is the second compression. To escape the curse one must reduce effective dimension (low-dimensional manifold assumptions, factor models, learned embeddings), impose structure (sparsity, priors, regularisation), search smartly (random sampling, active learning, hierarchical decomposition), or restrict scope (constrain attention to a low-dimensional sub-design). The practitioner does not need a separate playbook for each pathology; the same four moves apply because they all attack the same root cause — the mismatch between exponential space and finite budget. This is what makes the prime a complexity-management tool rather than merely a cautionary tale: it tells you both why high-dimensional reasoning fails and what the small, fixed set of escapes is.

Abstract Reasoning

The curse supports inference about qualitative regime change with dimension: at low dimension naive methods work, at moderate dimension they degrade, at high dimension they fail qualitatively and require structural assumptions. It supports a diagnostic move — when a method's performance collapses suddenly as you add features, ask whether the failure is exponential-budget-versus-space rather than tuning or data quality. And it supports a design move: before designing a policy, model, or search in a space of dimension d, estimate the smallest informative sample size and check it against your budget; if that size grows exponentially in d, you already know you will need a sparsity or structure assumption, and you can build it in from the start rather than discovering the need after failure.

The reasoning generalises because it is stated in terms of degrees of freedom and coverage budget, not in terms of any particular substrate's variables. A geneticist counting interacting loci, a regulator counting policy parameters, and an engineer counting design knobs are all reasoning about the same quantity — the dimension of a candidate-state space — and the same exponential-versus-budget logic governs all three.

Knowledge Transfer

The transfer is exceptionally clean because the curse is a piece of pure geometry wearing different domain costumes. From statistics to governance: a high-dimensional policy space inherits the curse directly, so exhaustive consultation, full sensitivity testing, and uniform stakeholder coverage all become exponentially expensive; the imported intervention is to identify the low-dimensional manifold of policy variation that actually matters (the analogue of feature selection), prefer modular policies that decompose into nearly independent sub-policies, and accept that comprehensive review is infeasible beyond a point. From numerical analysis to engineering design: the failure of deterministic grids transfers to the failure of exhaustive design exploration, and the same mitigations — random or quasi-random sampling, active learning, surrogate models — transfer with it. From optimisation to drug discovery: the impossibility of enumerating chemical space motivates structured generation, active learning, and low-dimensional embeddings. From geometry to recommendation and search: nearest-neighbour retrieval degrades in raw high-dimensional embeddings, motivating learned low-dimensional representations and approximate-nearest-neighbour methods.

What ports is not a metaphor but a theorem-shaped fact, and a fixed set of responses to it. The portable moves — feature selection, sparsity priors, embeddings, active learning, modularisation, surrogate models — are themselves substrate-neutral, because each is a way of asserting that the true dimension is lower than the nominal one. A practitioner who has internalised the curse in one field can walk into another, count its degrees of freedom, compare them to its coverage budget, and immediately know both whether the curse applies and which escape to reach for. The transfer's main subtlety is distinguishing the curse from its dual, the "blessing of dimensionality" — concentration of measure, Johnson-Lindenstrauss embeddings, double descent — in which high dimension creates new regularities exploitable by the right methods. The blessing does not challenge the curse's structural status; it is the complementary fact that some high-dimensional structure is benign, and a mature reasoner carries both.

Examples

Formal/abstract

The breakdown of nearest-neighbour retrieval is the cleanest formal worked instance. Draw \(n\) points uniformly at random from the unit hypercube \([0,1]^d\), and pick a query point. The candidate-state space is the cube, indexed by \(d\) independent dimensions; the sample budget is the fixed \(n\) points; and the method — find the query's nearest neighbour — depends on distances meaningfully discriminating near from far. As \(d\) grows, the ratio of the distance to the farthest point over the distance to the nearest point converges to $1$: every point becomes almost equidistant from the query, so "nearest neighbour" loses meaning. This is not a tuning artefact but a consequence of the underlying geometry — the budget-to-volume ratio \(n / 2^d\) (relative to a regular grid) collapses exponentially, so any finite sample becomes asymptotically empty, and the volume of the unit ball inscribed in the cube goes to zero while volume concentrates near the cube's corners and surface. The qualitative-threshold property is visible: at \(d = 2\) or $3$ nearest-neighbour search works fluently, by \(d = 20\) it has degraded, and by \(d = 1000\) raw-distance retrieval is meaningless — a regime change, not a gradual decline. The only escape is the effective-dimension reduction: if the \(n\) points actually lie on a low-dimensional manifold embedded in the high-dimensional cube, a learned embedding recovers the discriminating distances, which is exactly why modern retrieval systems learn low-dimensional representations rather than searching raw high-dimensional vectors. The intervention follows from the diagnosis — count the true degrees of freedom, compare to the coverage budget, and reach for an effective-dimension reduction rather than blaming the data.

Mapped back: Uniform points in a hypercube instantiate every commitment — dimensioned space, sub-exponential budget, distance-dependent method, collapsing coverage, threshold failure, effective-dimension escape — and the nearest-far convergence is the geometric fact that all the surface phenomena wear as costumes.

Applied/industry

The same theorem-shaped fact, not a metaphor, governs drug discovery and high-dimensional policy design. In molecular design, the candidate-state space is chemical space — the astronomically large set of synthesisable small molecules, indexed by many structural degrees of freedom; the search budget is any feasible library of compounds that can actually be synthesised and assayed, which grows nowhere near exponentially; and the method (find active molecules by similarity to known actives) depends on meaningful nearness in chemical space. Because the budget covers a vanishing fraction of the space, brute-force screening is structurally hopeless, and the curse predicts this before any screen is run. The escapes are exactly the prime's effective-dimension moves: pharmacophore-restricted search (assert that activity lives on a low-dimensional manifold of structural features), active learning (spend the budget where it is most informative rather than uniformly), and learned low-dimensional molecular embeddings. Policy design is the same structure in a governance substrate: a policy with many parameters, target groups, and timing variables defines a high-dimensional design space, and "comprehensive review" — exhaustive consultation, full sensitivity analysis, uniform stakeholder coverage — becomes exponentially expensive as the parameter count rises, so past a point it is not merely costly but structurally infeasible. The imported intervention is to identify the low-dimensional manifold of policy variation that actually matters (the analogue of feature selection), prefer modular policies that decompose into nearly independent sub-policies (lowering the effective dimension), and accept that exhaustive review cannot be done — reaching for structured sampling and surrogate models instead. A practitioner who has internalised the curse in statistics can walk into either domain, count its degrees of freedom against its coverage budget, and immediately know both that the curse applies and which escape to reach for.

Mapped back: Chemical-space screening and high-dimensional policy review are the curse in genuine non-statistical domains: exponential space against sub-exponential budget, threshold infeasibility, and the same fixed set of effective-dimension escapes — feature selection, sparsity, embeddings, modularisation, active sampling.

Structural Tensions

T1 — Nominal versus Effective Dimension (scopal). The curse counts nominal degrees of freedom, but the escapes all assert the effective dimension is lower — data on a manifold, sparse interactions, a few latent factors. The whole drama turns on which number is real. The failure mode runs both ways: assuming the curse bites at nominal dimension when the data actually lives on a low-dimensional manifold (needless pessimism, over-aggressive reduction that discards signal), or assuming a convenient low effective dimension that is not there (the reduction fabricates structure). Diagnostic: estimate intrinsic dimension empirically before reaching for an escape; the curse's force is real only at the effective count, and that count is itself an inference, not a given.

T2 — Curse versus Blessing of Dimensionality (sign/direction). High dimension does not only destroy — concentration of measure, Johnson-Lindenstrauss embeddings, and double descent make some high-dimensional structure exploitable. The same geometry that ruins nearest-neighbour also makes random projections nearly isometric. The failure mode is one-sided pessimism: invoking the curse to abandon a high-dimensional approach that the blessing would actually rescue, or its dual, trusting a benign concentration where the curse genuinely bites. Diagnostic: ask whether the method depends on local density and discriminating distances (curse territory) or on aggregate/projected structure (blessing territory); the sign of dimension's effect flips with what the method actually uses.

T3 — Qualitative Threshold versus Gradual Degradation (temporal/scalar). The prime stresses regime change — failure at a threshold, not graceful decline — but in practice many methods degrade smoothly enough that there is no clean cliff, and the "threshold" depends on sample size, noise, and the method. The failure mode is waiting for a dramatic break that never visibly arrives while performance quietly erodes, or conversely over-dramatising a smooth slope into a phantom cliff and discarding a method still inside its usable range. Diagnostic: plot performance against dimension at fixed budget rather than assuming a knee; whether the failure is a cliff or a slope is itself empirical and determines whether incremental data helps.

T4 — Geometric versus Statistical versus Computational Failure (scopal). Three failure types — sparse coverage (statistical), search/enumeration cost (computational), and uninformative distances (geometric) — are all downstream of exponential volume but call for different remedies. Lumping them hides which is biting. The failure mode is applying the wrong escape: throwing compute at a geometric collapse where distances have ceased to discriminate, or gathering more data against a search-cost wall that more data only worsens. Diagnostic: ask which resource the method actually exhausts — sample, cycles, or distance-information — because more data fixes sparsity, smarter search fixes enumeration, and only effective-dimension reduction fixes degenerate geometry.

T5 — Effective-Dimension Escape versus Assumption Risk (coupling). Every escape — sparsity, manifold, factor model, modularisation — buys tractability by importing a structural assumption, and that assumption can be wrong. The reduction that saves the computation can also encode a falsehood about the domain. The failure mode is laundering an unverified convenience assumption through a successful-looking pipeline: the embedding "works" on training data because it baked in the very structure it claims to discover, then fails out of distribution. Diagnostic: ask what the escape assumes about effective dimension and whether that assumption is independently testable; an escape is only as safe as the structural claim it smuggles in.

T6 — Independent Dimensions versus Correlated Structure (measurement). The curse's exponential volume growth presumes the dimensions are independent degrees of freedom — but real high-dimensional spaces are often heavily correlated, so the nominal count overstates the true volume the sample must cover. Counting raw variables as if independent is the measurement error at the root of many false curse-diagnoses. The failure mode is treating twenty correlated features as a twenty-dimensional curse when they span three effective directions, prescribing aggressive reduction for a problem that barely exists. Diagnostic: examine the covariance or interaction structure before counting dimensions; the load-bearing quantity is the number of independent directions of variation, not the number of columns.

Structural–Framed Character

Curse of dimensionality sits at the structural pole of the structural–framed spectrum, with an aggregate of 0.1 — effectively structural, the single non-zero criterion a minor concession to Bellman's naming. The prime is a piece of pure geometry: exponential volume growth against sub-exponential coverage, wearing many costumes. The breakdown of nearest-neighbour retrieval, the emptiness of any finite sample relative to a grid, the concentration of volume near a hypercube's surface — these are theorem-shaped facts, not interpretive frames.

Four diagnostics read 0.0. Vocabulary travels freely: the slots — dimensioned state space, coverage budget, density-or-nearness-dependent methods, the collapsing budget-to-volume ratio, the effective-dimension escape — need no home lexicon, because a geneticist counting interacting loci, a regulator counting policy parameters, and a chemist counting molecular degrees of freedom are all reasoning about the same quantity in their own words. Evaluative weight is zero: the curse is neither good nor bad — its dual, the "blessing of dimensionality" (concentration of measure, Johnson–Lindenstrauss embeddings), shows the same geometry can be benign, so high dimension carries no inherent valence. Human-practice binding is nil: the pathology runs in any space with independent degrees of freedom and a finite budget, with no human role required; it is a fact about geometry, not about practice. And import-versus-recognize falls on recognize: to diagnose the curse is to count degrees of freedom against budget and see a geometric collapse already present, not to add a frame. The only non-zero criterion is institutional origin at 0.5 — Bellman named the problem in dynamic programming, and that naming carries a faint disciplinary residue — but the structure he named is medium-neutral, which is why the aggregate is a near-structural 0.1. The prose label of "structural" matches the frontmatter; the entry's own Knowledge Transfer calls the transfer "exceptionally clean because the curse is a piece of pure geometry."

Substrate Independence

Curse of Dimensionality is a strongly substrate-independent prime — composite 4 / 5 on the substrate-independence scale. Its signature is pure geometry and probability: as the number of dimensions grows, volume explodes, data points become sparse and nearly equidistant, and intuitions tuned to low dimensions break down — a medium-neutral structural fact with no domain commitments whatsoever (structural abstraction 5). It bites identically in statistics and machine learning, high-dimensional optimisation, drug-discovery search over chemical space, multi-criteria governance and design problems, and biology's high-feature measurement spaces (domain breadth 4). The transfer is concrete: the same volume-and-sparsity mathematics that defeats nearest-neighbour methods also defeats grid search and exhaustive design exploration, and the formal result carries unchanged across fields (transfer evidence 4). What holds it just below the top breadth band is that its named home cases cluster in computational and statistical settings, even though the underlying geometry is universal.

  • Composite substrate independence — 4 / 5
  • Domain breadth — 4 / 5
  • Structural abstraction — 5 / 5
  • Transfer evidence — 4 / 5

Neighborhood in Abstraction Space

Curse Of Dimensionality sits in a moderately populated region (54th percentile for distinctiveness): it has near-neighbors but no dense thicket of synonyms.

Family — Staged Processes & Drift (32 primes)

Nearest neighbors

Computed from structural-signature embeddings · 2026-06-14

Not to Be Confused With

The closest and most natural confusion is with dimensionality_reduction, the prime's nearest embedding neighbour, because the two are almost always discussed together — yet they stand in a problem–solution relation, not an equivalence. The curse of dimensionality is the geometric pathology: as independent dimensions accumulate, volume grows exponentially while any finite sample or search budget does not, so coverage collapses, distances cease to discriminate, and methods that depend on local density or meaningful nearness fail qualitatively. Dimensionality reduction is one family of escapes from that pathology — learned embeddings, factor models, feature selection, manifold methods — each a way of asserting that the effective dimension is lower than the nominal one and thereby restoring tractability. The distinction matters because the reduction is not free: every escape imports a structural assumption (the data lives on a manifold, the interactions are sparse, a few latent factors suffice) that can be false, and a reduction applied where the curse does not actually bite (because the dimensions were correlated and the effective count was already low) discards real signal for no gain. Conflating the two leads to reflexively reducing whenever dimensions are many, rather than first asking whether the curse genuinely bites at the effective count. The curse tells you whether and why high-dimensional reasoning fails; dimensionality reduction is one of several responses — and the practitioner who treats them as the same thing reaches for the remedy before confirming the disease.

A second genuine confusion is with scalability, because both surface as a method "breaking down as things get bigger." But they break down for different reasons and respond to different fixes. Scalability concerns whether a system or algorithm degrades gracefully as load grows — more users, more data, more requests — and its failures are typically about throughput, contention, and resource growth that better engineering or more hardware can often relieve. The curse is specifically geometric: the failure is driven by added dimensions (independent degrees of freedom), and crucially more data or more compute does not relieve a geometric collapse where distances have ceased to discriminate — only an effective-dimension reduction does. The prime's own decomposition separates three failure types the scalability frame tends to blur: a statistical failure (sparse coverage, fixed by more data), a computational failure (enumeration cost, a scalability-flavoured wall), and a geometric failure (uninformative distances, fixed by neither more data nor more compute). Reading a geometric curse as a scalability problem prescribes the wrong remedy — throwing hardware or data at a wall that only structural assumptions about effective dimension can move.

For the practitioner the distinctions are a triage. First, is the failure geometric (added dimensions destroying distance and coverage — the curse) or merely a load-scaling wall (scalability)? If the curse, second, does it bite at the effective dimension or only the nominal one (count independent directions before reducing)? Only then does dimensionality reduction become the right tool — and only with awareness of the structural assumption it smuggles in. Skipping this triage means reducing dimensions that did not need reducing, or buying hardware against a geometry that hardware cannot fix.

Solution Archetypes

No catalogued solution archetypes reference this prime yet.