Exact matrix completion via convex optimization
read more
Citations
A Singular Value Thresholding Algorithm for Matrix Completion
Finding Structure with Randomness: Probabilistic Algorithms for Constructing Approximate Matrix Decompositions
Geometric Deep Learning: Going beyond Euclidean data
Finding structure with randomness: Probabilistic algorithms for constructing approximate matrix decompositions
The Power of Convex Relaxation: Near-Optimal Matrix Completion
References
Compressed sensing
Robust uncertainty principles: exact signal reconstruction from highly incomplete frequency information
Matrix Factorization Techniques for Recommender Systems
Using SeDuMi 1.02, a MATLAB toolbox for optimization over symmetric cones
Related Papers (5)
A Singular Value Thresholding Algorithm for Matrix Completion
Guaranteed Minimum-Rank Solutions of Linear Matrix Equations via Nuclear Norm Minimization
Frequently Asked Questions (14)
Q2. What is the likely candidate for a few samples?
Matrices whose column and row spaces have low coherence are likely not to vanish in too many entries and are their most likely candidates for matrices that are recoverable from a few samples.
Q3. How can the authors recover a sparse signal from a small set of measurements?
For instance, if x is k-sparse in the Fourier domain, that is, x is a superposition of k sinusoids, then it can be perfectly recovered with high probability—by 1 minimization—from the knowledge of about k log n of its entries sampled uniformly at random.
Q4. How many sparse signals can be recovered from a small set of measurements?
If F is chosen randomly from a suitable distribution, then with very high probability, all sparse signals with about k nonzero entries can be recovered from on the order of k log n measurements.
Q5. What is the number of degrees of freedom of a rank r matrix?
Since U and v each satisfy r(r + 1)/2 orthogonality constraints, the total number of degrees of freedom is r + 2nr − r (r + 1) = r(2n − r).
Q6. What is the way to recover a low-rank matrix?
If the number of measurements is sufficiently large, and if the entries are close to uniformly distributed, one might hope that there is only one low-rank matrix with these entries.
Q7. What is the significance of the results of Recht et al.30?
While it was known that the nuclear norm problem could be efficiently solved by semidefinite programming, the results of Recht et al.30 and the full version of this paper have inspired the development of many special purpose algorithms to rapidly minimize the nuclear norm.
Q8. What is the definition of the nuclear norm minimization problem?
The nuclear norm minimization problem (2.3) can be interpreted as inflating the unit ball until it just touches the affine space Xij = Mij.
Q9. What is the meaning of simplicity beyond rank and sparsity?
There are likely notions of simplicity beyond rank and sparsity that can also be leveraged in highdimensional data analysis to open new frontiers in low-rate sampling.
Q10. How many users are given the opportunity to rate movies?
24 Users (rows of the data matrix) are given the opportunity to rate movies (columns of the data matrix), but users typically rate only very few movies so that there are very few scattered observed entries of this data matrix.
Q11. What is the way to solve the NP-hard optimization problem?
This is unfortunately of little practical use, because not only is this optimization problem NP-hard but also all known algorithms which provide exact solutions require time doubly exponential in the dimension n of the matrix in both theory and practice.
Q12. What is the x-axis of the rank r matrix?
The x-axis again corresponds to the fraction of the entries of the matrix that are revealed to the solver, but, in this case, the number of measurements is divided by Dn = n(n + 1)/2, the number of unique entries in a positive-semidefinite matrix, and the dimension of the rank r matrices is dr = nr − r(r − 1)/2.
Q13. What are the algorithms that are projected gradients?
These algorithms are projected gradient algorithms which operate by alternately correcting the predictions on the observed entries and soft-thresholding the singular values of the iterate.
Q14. What is the recovery region for positive semidefinite matrices?
the recovery region is much larger for positive semidefinite matrices, and future work is needed to investigate if the theoretical scaling is also more favorable in this scenario of low-rank matrix completion.