Improved time bounds for near-optimal sparse Fourier representations

A. C. Gilbert, Shanmugavelayutham Muthukrishnan, M. Strauss

    Research output: Contribution to journalConference article

    Abstract

    We study the problem of finding a Fourier representation R of m terms for a given discrete signal A of length N. The Fast Fourier Transform (FFT) can find the optimal N-term representation in time O(N log N), but our goal is to get sublinear time algorithms when m " N. Suppose ∥A∥ 2 ≤ M ∥A - R opt2, where R opt is the optimal output. The previously best known algorithms output R such that ∥ A - R∥ 2 2 ≤ (1 + ε)∥ A - R opt2 2 with probability at least 1 - δ in time* polym,(m, log(1/δ), log N, log M, 1/ε). Although this is sublinear in the input size, the dominating expression is the polynomial factor in m which, for published algorithms, is greater than or equal to the bottleneck at m 2 that we identify below. Our experience with these algorithms shows that this is serious limitation in theory and in practice. Our algorithm beats this m 2 bottleneck. Our main result is a significantly improved algorithm for this problem and the d-dimensional analog. Our algorithm outputs an R with the same approximation guarantees but it runs in time m · poly(log(1/δ), log N, log M, 1/ε). A version of the algorithm holds for all N, though the details differ slightly according to the factorization of N. For the d-dimensional problem of size N 1 × N 2 × ⋯× N d, the linear-in-m algorithm extends efficiently to higher dimensions for certain factorizations of the N i's; we give a quadratic-in-m algorithm that works for any values of N i's. This article replaces several earlier, unpublished drafts.

    Original languageEnglish (US)
    Article number59141A
    Pages (from-to)1-15
    Number of pages15
    JournalProceedings of SPIE - The International Society for Optical Engineering
    Volume5914
    DOIs
    StatePublished - Dec 1 2005
    EventWavelets XI - San Diego, CA, United States
    Duration: Jul 31 2005Aug 3 2005

    Fingerprint

    Factorization
    factorization
    output
    Output
    draft
    Beat
    Fast Fourier transform
    Term
    Fast Fourier transforms
    Higher Dimensions
    synchronism
    polynomials
    Polynomials
    analogs
    Analogue
    Polynomial
    Approximation
    approximation
    Experience

    Keywords

    • Fourier analysis
    • Randomized approximation algorithms
    • Sampling
    • Sparse analysis

    ASJC Scopus subject areas

    • Electronic, Optical and Magnetic Materials
    • Condensed Matter Physics
    • Computer Science Applications
    • Applied Mathematics
    • Electrical and Electronic Engineering

    Cite this

    Improved time bounds for near-optimal sparse Fourier representations. / Gilbert, A. C.; Muthukrishnan, Shanmugavelayutham; Strauss, M.

    In: Proceedings of SPIE - The International Society for Optical Engineering, Vol. 5914, 59141A, 01.12.2005, p. 1-15.

    Research output: Contribution to journalConference article

    Gilbert, A. C. ; Muthukrishnan, Shanmugavelayutham ; Strauss, M. / Improved time bounds for near-optimal sparse Fourier representations. In: Proceedings of SPIE - The International Society for Optical Engineering. 2005 ; Vol. 5914. pp. 1-15.
    @article{79578a4b29ec468d957c504b8021fcc8,
    title = "Improved time bounds for near-optimal sparse Fourier representations",
    abstract = "We study the problem of finding a Fourier representation R of m terms for a given discrete signal A of length N. The Fast Fourier Transform (FFT) can find the optimal N-term representation in time O(N log N), but our goal is to get sublinear time algorithms when m {"} N. Suppose ∥A∥ 2 ≤ M ∥A - R opt∥ 2, where R opt is the optimal output. The previously best known algorithms output R such that ∥ A - R∥ 2 2 ≤ (1 + ε)∥ A - R opt ∥ 2 2 with probability at least 1 - δ in time* polym,(m, log(1/δ), log N, log M, 1/ε). Although this is sublinear in the input size, the dominating expression is the polynomial factor in m which, for published algorithms, is greater than or equal to the bottleneck at m 2 that we identify below. Our experience with these algorithms shows that this is serious limitation in theory and in practice. Our algorithm beats this m 2 bottleneck. Our main result is a significantly improved algorithm for this problem and the d-dimensional analog. Our algorithm outputs an R with the same approximation guarantees but it runs in time m · poly(log(1/δ), log N, log M, 1/ε). A version of the algorithm holds for all N, though the details differ slightly according to the factorization of N. For the d-dimensional problem of size N 1 × N 2 × ⋯× N d, the linear-in-m algorithm extends efficiently to higher dimensions for certain factorizations of the N i's; we give a quadratic-in-m algorithm that works for any values of N i's. This article replaces several earlier, unpublished drafts.",
    keywords = "Fourier analysis, Randomized approximation algorithms, Sampling, Sparse analysis",
    author = "Gilbert, {A. C.} and Shanmugavelayutham Muthukrishnan and M. Strauss",
    year = "2005",
    month = "12",
    day = "1",
    doi = "10.1117/12.615931",
    language = "English (US)",
    volume = "5914",
    pages = "1--15",
    journal = "Proceedings of SPIE - The International Society for Optical Engineering",
    issn = "0277-786X",
    publisher = "SPIE",

    }

    TY - JOUR

    T1 - Improved time bounds for near-optimal sparse Fourier representations

    AU - Gilbert, A. C.

    AU - Muthukrishnan, Shanmugavelayutham

    AU - Strauss, M.

    PY - 2005/12/1

    Y1 - 2005/12/1

    N2 - We study the problem of finding a Fourier representation R of m terms for a given discrete signal A of length N. The Fast Fourier Transform (FFT) can find the optimal N-term representation in time O(N log N), but our goal is to get sublinear time algorithms when m " N. Suppose ∥A∥ 2 ≤ M ∥A - R opt∥ 2, where R opt is the optimal output. The previously best known algorithms output R such that ∥ A - R∥ 2 2 ≤ (1 + ε)∥ A - R opt ∥ 2 2 with probability at least 1 - δ in time* polym,(m, log(1/δ), log N, log M, 1/ε). Although this is sublinear in the input size, the dominating expression is the polynomial factor in m which, for published algorithms, is greater than or equal to the bottleneck at m 2 that we identify below. Our experience with these algorithms shows that this is serious limitation in theory and in practice. Our algorithm beats this m 2 bottleneck. Our main result is a significantly improved algorithm for this problem and the d-dimensional analog. Our algorithm outputs an R with the same approximation guarantees but it runs in time m · poly(log(1/δ), log N, log M, 1/ε). A version of the algorithm holds for all N, though the details differ slightly according to the factorization of N. For the d-dimensional problem of size N 1 × N 2 × ⋯× N d, the linear-in-m algorithm extends efficiently to higher dimensions for certain factorizations of the N i's; we give a quadratic-in-m algorithm that works for any values of N i's. This article replaces several earlier, unpublished drafts.

    AB - We study the problem of finding a Fourier representation R of m terms for a given discrete signal A of length N. The Fast Fourier Transform (FFT) can find the optimal N-term representation in time O(N log N), but our goal is to get sublinear time algorithms when m " N. Suppose ∥A∥ 2 ≤ M ∥A - R opt∥ 2, where R opt is the optimal output. The previously best known algorithms output R such that ∥ A - R∥ 2 2 ≤ (1 + ε)∥ A - R opt ∥ 2 2 with probability at least 1 - δ in time* polym,(m, log(1/δ), log N, log M, 1/ε). Although this is sublinear in the input size, the dominating expression is the polynomial factor in m which, for published algorithms, is greater than or equal to the bottleneck at m 2 that we identify below. Our experience with these algorithms shows that this is serious limitation in theory and in practice. Our algorithm beats this m 2 bottleneck. Our main result is a significantly improved algorithm for this problem and the d-dimensional analog. Our algorithm outputs an R with the same approximation guarantees but it runs in time m · poly(log(1/δ), log N, log M, 1/ε). A version of the algorithm holds for all N, though the details differ slightly according to the factorization of N. For the d-dimensional problem of size N 1 × N 2 × ⋯× N d, the linear-in-m algorithm extends efficiently to higher dimensions for certain factorizations of the N i's; we give a quadratic-in-m algorithm that works for any values of N i's. This article replaces several earlier, unpublished drafts.

    KW - Fourier analysis

    KW - Randomized approximation algorithms

    KW - Sampling

    KW - Sparse analysis

    UR - http://www.scopus.com/inward/record.url?scp=30844452004&partnerID=8YFLogxK

    UR - http://www.scopus.com/inward/citedby.url?scp=30844452004&partnerID=8YFLogxK

    U2 - 10.1117/12.615931

    DO - 10.1117/12.615931

    M3 - Conference article

    AN - SCOPUS:30844452004

    VL - 5914

    SP - 1

    EP - 15

    JO - Proceedings of SPIE - The International Society for Optical Engineering

    JF - Proceedings of SPIE - The International Society for Optical Engineering

    SN - 0277-786X

    M1 - 59141A

    ER -