Testing noisy linear functions for sparsity
Testing noisy linear functions for sparsity
复制标题
测试噪声线性函数的稀疏性
DOI:
10.1145/3357713.3384239
复制
发表时间:
2020
期刊:
影响因子:
--
通讯作者:
Servedio, Rocco A.
中科院分区:
文献类型:
--
作者:
Chen, Xue;De, Anindya;Servedio, Rocco A.
We consider the following basic inference problem: there is an unknown high-dimensional vectorw∈ ℝn, and an algorithm is given access to labeled pairs (x,y) wherex∈ ℝnis a measurement andy=w·x+noise. What is the complexity of deciding whether the target vectorwis (approximately)k-sparse? The recovery analogue of this problem — given the promise thatwis sparse, find or approximate the vectorw— is the famoussparse recoveryproblem, with a rich body of work in signal processing, statistics, and computer science.We study the decision version of this problem (i.e. deciding whether the unknownwisk-sparse) from the vantage point ofproperty testing. Our focus is on answering the following high-level question: when is it possible to efficientlytestwhether the unknown target vectorwis sparse versus far-from-sparse using a number of samples which iscompletely independentof the dimensionn? We consider the natural setting in whichxis drawn from an i.i.d. product distributionDover ℝnand thenoiseprocess is independent of the inputx. As our main result, we give a general algorithm which solves the above-described testing problem using a number of samples which is completely independent of the ambient dimensionn, as long asDis not a Gaussian. In fact, our algorithm isfully noise tolerant, in the sense that for an arbitraryw, it approximately computes the distance ofwto the closestk-sparse vector. To complement this algorithmic result, we show that weakening any of our conditions makes it information-theoretically impossible foranyalgorithm to solve the testing problem with fewer than essentially lognsamples. Thus our conditions essentially characterize when it is possible to test noisy linear functions for sparsity with constant sample complexity.Our algorithmic approach is based on relating the cumulants of the output distribution (i.e. ofy) with elementary power sum symmetric polynomials inwand using the latter to measure the sparsity ofw. This approach crucially relies on a theorem of Marcinkiewicz from probability theory. In fact, to obtain effective sample complexity bounds with our approach, we prove a new finitary version of Marcinkiewicz’s theorem. This involves extending the complex analytic arguments used in the original proof with results about the distribution of zeros of entire functions.
登录
查看更多内容
DOI:
--
发表时间:
2017
期刊:
International Workshop and International Workshop on Approximation, Randomization, and Combinatorial Optimization. Algorithms and Techniques
影响因子:
--
作者:
Xi Chen;Adam Freilich;R. Servedio;Timothy Sun
通讯作者:
Timothy Sun
DOI:
10.1137/1.9781611973105.97
发表时间:
2012
期刊:
Electron. Colloquium Comput. Complex.
影响因子:
--
作者:
Arnab Bhattacharyya;E. Fischer;Shachar Lovett
通讯作者:
Shachar Lovett
DOI:
10.1002/rsa.20507
发表时间:
2010
期刊:
2010 IEEE 51st Annual Symposium on Foundations of Computer Science
影响因子:
--
作者:
Arnab Bhattacharyya;Elena Grigorescu;A. Shapira
通讯作者:
A. Shapira
DOI:
--
发表时间:
1939
期刊:
影响因子:
--
作者:
J. Marcinkiewicz
通讯作者:
J. Marcinkiewicz
DOI:
10.1109/isit.2012.6283954
发表时间:
2012
期刊:
2012 IEEE International Symposium on Information Theory Proceedings
影响因子:
--
作者:
Eric Price;David P. Woodruff
通讯作者:
David P. Woodruff