Choose your reading route
eduKateSG · Why Mathematics? · Robust PCA for Sparse Sensor Corruption: Low-Rank Structure, Sparse Errors, Nuclear Norms and Recovery
Start with the question, audit the worked example, or jump to the assumptions and practice. Every chapter stays open for uninterrupted reading.
Understand the question
Define the quantity and follow the mechanism.Route 2 · Chapters 6–10
Reproduce the example
Audit the numbers one step at a time.Route 3 · Chapters 11–15
Check assumptions
See where design and calibration matter.Route 4 · Chapters 16–20
Explore applications
Connect the method to a concrete investigation.Route 5 · Chapters 21–24
Study and verify
Practise, inspect failure cases and read further.
Full chapter index · Practice and checking · Mathematics Learning Hub
Why mathematics is important becomes practical when most sensor relationships are structured but a small fraction of readings are grossly corrupted. Robust principal component analysis, in the principal component pursuit sense, writes an observed matrix D as D=L+S, where L is low rank and S is sparse. A convex programme replaces rank and the number of nonzero entries with the nuclear norm and entrywise L1 norm.
For a teaching example, let L=[[2,2],[2,2]], which has rank one, and S=[[0,5],[0,0]], with one nonzero entry. Then D=[[2,7],[2,2]]. The nuclear norm of L is 4 because its only nonzero singular value is 4, while the L1 norm of S is 5. With λ=1/√2, this candidate decomposition has objective 4+5/√2≈7.5355. The example makes the low-rank-plus-sparse idea visible; exact recovery claims require incoherence, sparsity and other assumptions stated in the research.
Did You Know? Ordinary PCA can rotate toward a gross outlier. Robust PCA instead asks whether a low-rank matrix and a sparse error matrix can explain the observations together.
This article answers one intent: The mechanism decomposes a data matrix into shared low-rank structure and isolated sparse corruption by balancing nuclear and L1 norms under explicit identifiability assumptions.
Robust PCA models an observed data matrix as the sum of a low-rank component and sparse corruption. The worked two-by-two example exposes rank, sparsity, the nuclear norm and the entrywise L1 norm used in principal component pursuit. Later chapters add identifiability conditions, missing entries, online use and sensor-fault interpretation. Students transfer matrix rank, norms and optimisation; practitioners transfer a disciplined separation of shared structure from isolated gross errors.
Full chapter index
Chapters 1–5 · Understand the question
Chapters 6–10 · Reproduce the example
Chapters 11–15 · Check assumptions
Chapters 16–20 · Explore applications
Chapters 21–24 · Study and verify
Begin with the object being measured, because a detector name is less useful than the statistic it changes. A few arbitrarily large entries can distort ordinary covariance and principal components. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
One sensor spike changes a data row. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Dense small noise is not sparse corruption. Model noise separately. Transfer the idea by changing one input at a time and explaining the direction before computing.
A useful audit begins by writing the claim in plain language, then identifying the observation, reference model and comparison rule. In this chapter, the central claim is: A few arbitrarily large entries can distort ordinary covariance and principal components. Test it first on a short sequence where every intermediate value can be inspected. Then change one condition and explain why the conclusion does or does not survive.
The mathematical value appears when a pattern is connected to one reproducible calculation. Correlated sensors or repeated patterns can place healthy data near a low-rank matrix. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
All rows or columns share a common pattern. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Healthy complexity may not be low rank. Check singular values. Keep raw evidence beside the transformed statistic so diagnosis remains possible.
For transfer, sketch a before-and-after case and label what is measured, what is predicted and what is inferred. The concrete case here is: All rows or columns share a common pattern. That example is illustrative rather than evidence about a particular school, learner or machine. A sound explanation keeps calculation, interpretation and action in separate sentences.
Separate the mechanism from the interpretation so a number does not become unsupported certainty. Store isolated gross errors in S while most entries remain zero. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Only entry (1,2) equals 5. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
A persistent failed sensor forms an entire corrupted column and may violate random sparsity assumptions. Test patterns. Compare with a simple baseline and an alternative cause before deciding.
Reliability work needs a counterexample as well as a successful example. Here the important boundary is: A persistent failed sensor forms an entire corrupted column and may violate random sparsity assumptions. Ask what other mechanism could create the same observed pattern, what additional measurement would separate the explanations, and how uncertainty should be recorded. This turns a formula into a defensible decision process.
Keep timing, units and assumptions visible before applying a threshold. Require the observed matrix to equal the sum of structured and sparse parts in the noiseless formulation. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
[[2,7],[2,2]] splits into L and S. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Measurement noise usually requires a relaxed model. Declare the formulation. Explain the key equation aloud and point to the term carrying each meaning.
A student can practise this idea without specialised equipment. Start with a tiny table, reproduce the stated operation, and predict how the output changes when one entry is doubled, delayed or removed. The practical next step is: Declare the formulation. Explain the result aloud, including the assumption that would most change the answer.
Make the example auditable by predicting direction, calculating and checking a nearby case. Rank counts independent matrix directions; the teaching L has one nonzero singular value. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
The singular value is 4. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Rank is discontinuous and hard to optimise directly. Use a convex surrogate carefully. Report observation, assumption, calculation and authorised action as separate layers.
A useful audit begins by writing the claim in plain language, then identifying the observation, reference model and comparison rule. In this chapter, the central claim is: Rank counts independent matrix directions; the teaching L has one nonzero singular value. Test it first on a short sequence where every intermediate value can be inspected. Then change one condition and explain why the conclusion does or does not survive.
Begin with the object being measured, because a detector name is less useful than the statistic it changes. Subtract L from D to recover the illustrative S. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
The difference isolates a value 5 at one position. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
A chosen decomposition is not automatically the optimiser. Check the objective and assumptions. Transfer the idea by changing one input at a time and explaining the direction before computing.
For transfer, sketch a before-and-after case and label what is measured, what is predicted and what is inferred. The concrete case here is: The difference isolates a value 5 at one position. That example is illustrative rather than evidence about a particular school, learner or machine. A sound explanation keeps calculation, interpretation and action in separate sentences.
| Object | Value |
|---|---|
| Low-rank L | [[2,2],[2,2]] |
| Sparse S | [[0,5],[0,0]] |
| Observed D=L+S | [[2,7],[2,2]] |
| rank(L) | 1 |
| Nonzero sparse entries | 1 |
| ‖L‖* | 4 |
| ‖S‖₁ | 5 |
| λ=1/√2 objective | 4+5/√2≈7.5355 |
The mathematical value appears when a pattern is connected to one reproducible calculation. Sum singular values to form the convex nuclear-norm penalty. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
The teaching L has nuclear norm 4. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Nuclear norm shrinks singular values and can bias magnitudes. Evaluate reconstruction. Keep raw evidence beside the transformed statistic so diagnosis remains possible.
Reliability work needs a counterexample as well as a successful example. Here the important boundary is: Nuclear norm shrinks singular values and can bias magnitudes. Ask what other mechanism could create the same observed pattern, what additional measurement would separate the explanations, and how uncertainty should be recorded. This turns a formula into a defensible decision process.
Separate the mechanism from the interpretation so a number does not become unsupported certainty. Sum absolute entries to encourage a sparse S. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
The teaching S has L1 norm 5. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
L1 can spread or shrink errors under correlation. Inspect support stability. Compare with a simple baseline and an alternative cause before deciding.
A student can practise this idea without specialised equipment. Start with a tiny table, reproduce the stated operation, and predict how the output changes when one entry is doubled, delayed or removed. The practical next step is: Inspect support stability. Explain the result aloud, including the assumption that would most change the answer.
Keep timing, units and assumptions visible before applying a threshold. Minimise ‖L‖*+λ‖S‖1 subject to D=L+S. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
λ balances low rank and sparsity. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Default λ is theory-guided, not universally optimal. Validate sensitivity. Explain the key equation aloud and point to the term carrying each meaning.
A useful audit begins by writing the claim in plain language, then identifying the observation, reference model and comparison rule. In this chapter, the central claim is: Minimise ‖L‖*+λ‖S‖1 subject to D=L+S. Test it first on a short sequence where every intermediate value can be inspected. Then change one condition and explain why the conclusion does or does not survive.
Make the example auditable by predicting direction, calculating and checking a nearby case. With λ=1/√2, compute 4+5/√2≈7.5355. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Every term is auditable. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
A lower objective from another decomposition would win. Compare feasible alternatives. Report observation, assumption, calculation and authorised action as separate layers.
For transfer, sketch a before-and-after case and label what is measured, what is predicted and what is inferred. The concrete case here is: Every term is auditable. That example is illustrative rather than evidence about a particular school, learner or machine. A sound explanation keeps calculation, interpretation and action in separate sentences.
Begin with the object being measured, because a detector name is less useful than the statistic it changes. Recovery theory prevents low-rank singular vectors from being too concentrated in a few coordinates. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
A diffuse common pattern differs from a single-sensor spike. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Highly localised low-rank structure is confused with sparsity. Check leverage. Transfer the idea by changing one input at a time and explaining the direction before computing.
Reliability work needs a counterexample as well as a successful example. Here the important boundary is: Highly localised low-rank structure is confused with sparsity. Ask what other mechanism could create the same observed pattern, what additional measurement would separate the explanations, and how uncertainty should be recorded. This turns a formula into a defensible decision process.
The mathematical value appears when a pattern is connected to one reproducible calculation. Sparse errors must not occupy too many strategically aligned entries. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Random isolated corruption is favourable. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
A whole failed row or column is harder. Use structured-sparse extensions when needed. Keep raw evidence beside the transformed statistic so diagnosis remains possible.
A student can practise this idea without specialised equipment. Start with a tiny table, reproduce the stated operation, and predict how the output changes when one entry is doubled, delayed or removed. The practical next step is: Use structured-sparse extensions when needed. Explain the result aloud, including the assumption that would most change the answer.
Separate the mechanism from the interpretation so a number does not become unsupported certainty. Allow a dense small-noise matrix in addition to L and S. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
D=L+S+N with bounded N. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Three-way separation needs more assumptions. Report tolerances. Compare with a simple baseline and an alternative cause before deciding.
A useful audit begins by writing the claim in plain language, then identifying the observation, reference model and comparison rule. In this chapter, the central claim is: Allow a dense small-noise matrix in addition to L and S. Test it first on a short sequence where every intermediate value can be inspected. Then change one condition and explain why the conclusion does or does not survive.
Keep timing, units and assumptions visible before applying a threshold. Optimise equality or loss only over observed matrix entries. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
A validity mask excludes missing cells. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Missingness can be informative or concentrated. Track the mask. Explain the key equation aloud and point to the term carrying each meaning.
For transfer, sketch a before-and-after case and label what is measured, what is predicted and what is inferred. The concrete case here is: A validity mask excludes missing cells. That example is illustrative rather than evidence about a particular school, learner or machine. A sound explanation keeps calculation, interpretation and action in separate sentences.
Make the example auditable by predicting direction, calculating and checking a nearby case. Monitor the magnitude or support of newly estimated sparse corruption and changes in low-rank scores. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
A sensor spike appears in S. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Batch decomposition can use future data. Separate offline and online claims. Report observation, assumption, calculation and authorised action as separate layers.
Reliability work needs a counterexample as well as a successful example. Here the important boundary is: Batch decomposition can use future data. Ask what other mechanism could create the same observed pattern, what additional measurement would separate the explanations, and how uncertainty should be recorded. This turns a formula into a defensible decision process.
Begin with the object being measured, because a detector name is less useful than the statistic it changes. Map nonzero S entries to sensors and times as diagnostic candidates. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Column two at time one is highlighted. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Process faults can also be sparse in the data table. Confirm physical cause. Transfer the idea by changing one input at a time and explaining the direction before computing.
A student can practise this idea without specialised equipment. Start with a tiny table, reproduce the stated operation, and predict how the output changes when one entry is doubled, delayed or removed. The practical next step is: Confirm physical cause. Explain the result aloud, including the assumption that would most change the answer.
The mathematical value appears when a pattern is connected to one reproducible calculation. Update low-rank subspaces and sparse estimates as observations arrive. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
A new vector is decomposed against the current subspace. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Adaptive updates can absorb persistent faults. Use guarded learning. Keep raw evidence beside the transformed statistic so diagnosis remains possible.
A useful audit begins by writing the claim in plain language, then identifying the observation, reference model and comparison rule. In this chapter, the central claim is: Update low-rank subspaces and sparse estimates as observations arrive. Test it first on a short sequence where every intermediate value can be inspected. Then change one condition and explain why the conclusion does or does not survive.
Separate the mechanism from the interpretation so a number does not become unsupported certainty. Algorithms alternate singular-value thresholding and soft-thresholding operations. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Each iteration shrinks rank and sparse entries. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Large matrices make repeated SVD expensive. Use convergence and resource checks. Compare with a simple baseline and an alternative cause before deciding.
For transfer, sketch a before-and-after case and label what is measured, what is predicted and what is inferred. The concrete case here is: Each iteration shrinks rank and sparse entries. That example is illustrative rather than evidence about a particular school, learner or machine. A sound explanation keeps calculation, interpretation and action in separate sentences.
Keep timing, units and assumptions visible before applying a threshold. Sparse-corruption evidence can trigger raw-signal review, redundancy checks or sensor maintenance. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
One cell receives a large S estimate. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
It does not prove the sensor alone caused the anomaly. Seek independent evidence. Explain the key equation aloud and point to the term carrying each meaning.
Reliability work needs a counterexample as well as a successful example. Here the important boundary is: It does not prove the sensor alone caused the anomaly. Ask what other mechanism could create the same observed pattern, what additional measurement would separate the explanations, and how uncertainty should be recorded. This turns a formula into a defensible decision process.
Make the example auditable by predicting direction, calculating and checking a nearby case. The topic connects matrices, rank, absolute value, optimisation and modelling assumptions. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
A learner can add L and S and compute both norms. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Robust does not mean assumption-free. Ask what corruption pattern is allowed. Report observation, assumption, calculation and authorised action as separate layers.
A student can practise this idea without specialised equipment. Start with a tiny table, reproduce the stated operation, and predict how the output changes when one entry is doubled, delayed or removed. The practical next step is: Ask what corruption pattern is allowed. Explain the result aloud, including the assumption that would most change the answer.
Begin with the object being measured, because a detector name is less useful than the statistic it changes. Specify matrix orientation, scaling, λ, noise, missingness, solver tolerance, update policy and actions. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Tests include isolated spikes, failed columns, dense drift and regime changes. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Unlogged scaling changes the penalty balance. Version the model. Transfer the idea by changing one input at a time and explaining the direction before computing.
A useful audit begins by writing the claim in plain language, then identifying the observation, reference model and comparison rule. In this chapter, the central claim is: Specify matrix orientation, scaling, λ, noise, missingness, solver tolerance, update policy and actions. Test it first on a short sequence where every intermediate value can be inspected. Then change one condition and explain why the conclusion does or does not survive.
The mathematical value appears when a pattern is connected to one reproducible calculation. Ordinary PCA fits one low-rank approximation under squared loss; robust PCA explicitly allocates sparse gross errors. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
A spike can move to S instead of rotating L. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
RPCA costs more and needs sparsity assumptions. Use the simplest adequate method. Keep raw evidence beside the transformed statistic so diagnosis remains possible.
For transfer, sketch a before-and-after case and label what is measured, what is predicted and what is inferred. The concrete case here is: A spike can move to S instead of rotating L. That example is illustrative rather than evidence about a particular school, learner or machine. A sound explanation keeps calculation, interpretation and action in separate sentences.
Separate the mechanism from the interpretation so a number does not become unsupported certainty. RPCA does not guarantee every nonzero S entry is a sensor fault. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Add a legitimate rare event and inspect the decomposition. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Sparse means uncommon in the matrix representation, not necessarily faulty. Interpret context. Compare with a simple baseline and an alternative cause before deciding.
Reliability work needs a counterexample as well as a successful example. Here the important boundary is: Sparse means uncommon in the matrix representation, not necessarily faulty. Ask what other mechanism could create the same observed pattern, what additional measurement would separate the explanations, and how uncertainty should be recorded. This turns a formula into a defensible decision process.
Keep timing, units and assumptions visible before applying a threshold. Principal component pursuit research states exact-recovery conditions and convex optimisation; monitoring studies adapt it to fault detection. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Recent eduKate articles cover PCA residuals and set-membership diagnosis. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Theoretical recovery is conditional. Stress-test violations. Explain the key equation aloud and point to the term carrying each meaning.
A student can practise this idea without specialised equipment. Start with a tiny table, reproduce the stated operation, and predict how the output changes when one entry is doubled, delayed or removed. The practical next step is: Stress-test violations. Explain the result aloud, including the assumption that would most change the answer.
The nearest published companions are PCA T-squared and SPE monitoring, dynamic PCA, canonical variate analysis and set-membership fault diagnosis. The eduKateSG Mathematics Learning Hub, eduKateSingapore Mathematics World, eduKate Sengkang Mathematics Hub, Bukit Timah Tutor Mathematics Hub and eduKate Punggol Mathematics Pathway provide wider learning routes. Primary sources: https://doi.org/10.1145/1970392.1970395, https://www.microsoft.com/en-us/research/publication/robust-principal-component-analysis-2/, https://doi.org/10.1088/1742-6596/659/1/012035.
Frequently asked questions
What does robust PCA decompose?
It separates an observed matrix into low-rank structure and sparse corruption, sometimes with an added dense-noise term.
What is the nuclear norm?
It is the sum of singular values and serves as a convex surrogate that encourages low rank.
What does the L1 norm do?
The entrywise L1 penalty encourages many sparse-error entries to become zero.
Does exact recovery always hold?
No. Recovery guarantees depend on incoherence, sparsity pattern, sample support and the stated model.
Is every nonzero sparse entry a sensor fault?
No. It is a diagnostic candidate that may also represent a rare process event or model mismatch.
Extended practice laboratory
Practice case 1: Add matrices
Setup: Use L and S.
Checking target: Confirm D=[[2,7],[2,2]].
Practice case 2: Rank L
Setup: Inspect repeated rows.
Checking target: Confirm rank one.
Practice case 3: Singular value
Setup: Compute the nonzero singular value.
Checking target: Confirm 4.
Practice case 4: L1 norm S
Setup: Sum absolute entries.
Checking target: Confirm 5.
Practice case 5: Objective
Setup: Use λ=1/√2.
Checking target: Confirm about 7.5355.
Practice case 6: Move corruption
Setup: Place 5 in another cell.
Checking target: Track support.
Practice case 7: Dense noise
Setup: Add .1 to every cell.
Checking target: Explain why it is not sparse.
Practice case 8: Failed column
Setup: Corrupt both entries of column two.
Checking target: Discuss assumptions.
Practice case 9: Missing cell
Setup: Mask D(1,2).
Checking target: State lost evidence.
Practice case 10: Decision memo
Setup: Report sparse corruption.
Checking target: List alternative explanations.