Tight Bounds on Low-Degree Spectral Concentration of Submodular and XOS Functions

Submodular and fractionally subadditive (or equivalently XOS) functions play a fundamental role in combinatorial optimization, algorithmic game theory and machine learning. Motivated by learnability of these classes of functions from random examples, we consider the question of how well such functions can be approximated by low-degree polynomials in ℓ<sub>2</sub> norm over the uniform distribution. This question is equivalent to understanding the concentration of Fourier weight on low-degree coefficients, a central concept in Fourier analysis. Denoting the smallest degree sufficient to approximate f in ℓ<sub>2</sub> norm within ∈ by deg<sub>∈</sub>(ℓ<sub>2</sub>)(f), we show that : For any submodular function f : {0, 1}<sup>n</sup> → [0, 1], deg<sub>∈</sub>(ℓ<sub>2</sub>)(f) = O(log(1/∈)/∈<sup>4/5</sup>) and there is a submodular function that requires degree Ω(1/∈<sup>4/5</sup>). : For any XOS function f : {0, 1} → [0, 1], deg<sub>∈</sub>(ℓ<sub>2</sub>) (f) = O(1/∈) and there exists an XOS function that requires degree Ω(1/∈). This improves on previous approaches that all showed an upper bound of O(1/∈<sup>2</sup>) for submodular [CKKL12], [FKV13], [FV13] and XOS [FV13] functions. The best previous lower bound was Ω(1/∈<sup>2/3</sup>) for monotone submodular functions [FKV13]. Our techniques reveal new structural properties of submodular and XOS functions and the upper bounds lead to nearly optimal PAC learning algorithms for these classes of functions.

[1]  Pravesh Kothari,et al.  Nearly Tight Bounds on ℓ1 Approximation of Self-Bounding Functions , 2014, ArXiv.

[2]  Dana Ron,et al.  Testing Monotonicity , 2000, Comb..

[3]  Rocco A. Servedio,et al.  Optimal Cryptographic Hardness of Learning Monotone Functions , 2009, Theory Comput..

[4]  Li-Yang Tan,et al.  Approximate resilience, monotonicity, and the complexity of agnostic learning , 2014, SODA.

[5]  Ehud Friedgut,et al.  Boolean Functions With Low Average Sensitivity Depend On Few Coordinates , 1998, Comb..

[6]  Jan Vondrák,et al.  Is Submodularity Testable? , 2010, Algorithmica.

[7]  Sofya Raskhodnikova,et al.  Learning pseudo-Boolean k-DNF and submodular functions , 2013, SODA.

[8]  Peter L. Bartlett,et al.  Rademacher and Gaussian Complexities: Risk Bounds and Structural Results , 2003, J. Mach. Learn. Res..

[9]  F. Dunstan MATROIDS AND SUBMODULAR FUNCTIONS , 1976 .

[10]  Rocco A. Servedio,et al.  Agnostically learning halfspaces , 2005, 46th Annual IEEE Symposium on Foundations of Computer Science (FOCS'05).

[11]  Maurice Queyranne,et al.  A combinatorial algorithm for minimizing symmetric submodular functions , 1995, SODA '95.

[12]  Andreas Krause,et al.  Near-optimal sensor placements in Gaussian processes , 2005, ICML.

[13]  Leslie G. Valiant,et al.  A theory of the learnable , 1984, STOC '84.

[14]  R. Schapire,et al.  Toward efficient agnostic learning , 1992, COLT '92.

[15]  Daniel M. Kane,et al.  Learning Halfspaces Under Log-Concave Densities: Polynomial Approximations and Moment Matching , 2013, COLT.

[16]  Elchanan Mossel,et al.  On the noise sensitivity of monotone functions , 2003, Random Struct. Algorithms.

[17]  S. Boucheron,et al.  A sharp concentration inequality with applications , 1999, Random Struct. Algorithms.

[18]  Pravesh Kothari,et al.  Representation, Approximation and Learning of Submodular Functions Using Low-rank Decision Trees , 2013, COLT.

[19]  Noam Nisan,et al.  Constant depth circuits, Fourier transform, and learnability , 1989, 30th Annual Symposium on Foundations of Computer Science.

[20]  T. Sanders,et al.  Analysis of Boolean Functions , 2012, ArXiv.

[21]  Vahab S. Mirrokni,et al.  Approximating submodular functions everywhere , 2009, SODA.

[22]  Satoru Iwata,et al.  A combinatorial strongly polynomial algorithm for minimizing submodular functions , 2001, JACM.

[23]  C. Guestrin,et al.  Near-optimal sensor placements: maximizing information while minimizing communication cost , 2006, 2006 5th International Conference on Information Processing in Sensor Networks.

[24]  Nathan Linial,et al.  The influence of variables on Boolean functions , 1988, [Proceedings 1988] 29th Annual Symposium on Foundations of Computer Science.

[25]  V. Koltchinskii,et al.  Rademacher Processes and Bounding the Risk of Function Learning , 2004, math/0405338.

[26]  Akira Maruoka,et al.  On learning monotone Boolean functions under the uniform distribution , 2006, Theor. Comput. Sci..

[27]  Jan Vondrák,et al.  Covering minimum spanning trees of random subgraphs , 2004, SODA '04.

[28]  Daniel Lehmann,et al.  Combinatorial auctions with decreasing marginal utilities , 2001, EC '01.

[29]  M. Talagrand On Russo's Approximate Zero-One Law , 1994 .

[30]  Gábor Lugosi,et al.  Concentration Inequalities , 2008, COLT.

[31]  Nathan Linial,et al.  Collective coin flipping, robust voting schemes and minima of Banzhaf values , 1985, 26th Annual Symposium on Foundations of Computer Science (sfcs 1985).

[32]  Ryan O'Donnell,et al.  On the Fourier tails of bounded functions over the discrete cube , 2006, STOC '06.

[33]  Sofya Raskhodnikova,et al.  Lp-testing , 2014, STOC.

[34]  Michel Talagrand,et al.  How much are increasing sets positively correlated? , 1996, Comb..

[35]  Colin McDiarmid,et al.  Concentration for self-bounding functions and an inequality of Talagrand , 2006 .

[36]  Jan Vondrák,et al.  Optimal approximation for the submodular welfare problem in the value oracle model , 2008, STOC.

[37]  John Langford,et al.  On learning monotone Boolean functions , 1998, Proceedings 39th Annual Symposium on Foundations of Computer Science (Cat. No.98CB36280).

[38]  Maria-Florina Balcan,et al.  Learning Valuation Functions , 2011, COLT.

[39]  Tim Roughgarden,et al.  Sketching valuation functions , 2012, SODA.

[40]  László Lovász,et al.  Submodular functions and convexity , 1982, ISMP.

[41]  Shahar Dobzinski,et al.  An improved approximation algorithm for combinatorial auctions with submodular bidders , 2006, SODA '06.

[42]  Andreas Krause,et al.  Near-Optimal Sensor Placements in Gaussian Processes: Theory, Efficient Algorithms and Empirical Studies , 2008, J. Mach. Learn. Res..

[43]  Andris Ambainis,et al.  The Need for Structure in Quantum Speedups , 2009, Theory Comput..

[44]  R. O'Donnell,et al.  Computational applications of noise sensitivity , 2003 .

[45]  Leslie G. Valiant,et al.  Learning Boolean formulas , 1994, JACM.

[46]  Aaron Roth,et al.  Privately releasing conjunctions and the statistical query barrier , 2010, STOC '11.

[47]  Vahab S. Mirrokni,et al.  Maximizing Non-Monotone Submodular Functions , 2011, 48th Annual IEEE Symposium on Foundations of Computer Science (FOCS'07).

[48]  Vitaly Feldman,et al.  Optimal bounds on approximation of submodular and XOS functions by juntas , 2014, ITA.

[49]  Rocco A. Servedio,et al.  Learning intersections of halfspaces with a margin , 2004, J. Comput. Syst. Sci..

[50]  Pravesh Kothari,et al.  Submodular functions are noise stable , 2012, SODA.

[51]  Bruce A. Reed,et al.  Concentration for self‐bounding functions and an inequality of Talagrand , 2006, Random Struct. Algorithms.

[52]  Uriel Feige,et al.  On maximizing welfare when utility functions are subadditive , 2006, STOC '06.

[53]  Mohammad Bavarian,et al.  On the Sum of L1 Influences , 2013, 2014 IEEE 29th Conference on Computational Complexity (CCC).

[54]  Jeff Kahn,et al.  Thresholds and Expectation Thresholds , 2007, Comb. Probab. Comput..

[55]  Ambuj Tewari,et al.  On the Complexity of Linear Prediction: Risk Bounds, Margin Bounds, and Regularization , 2008, NIPS.

[56]  Ryan O'Donnell,et al.  KKL, Kruskal-Katona, and Monotone Nets , 2009, 2009 50th Annual IEEE Symposium on Foundations of Computer Science.

[57]  Jan Vondrák,et al.  Optimal Bounds on Approximation of Submodular and XOS Functions by Juntas , 2013, 2013 IEEE 54th Annual Symposium on Foundations of Computer Science.

[58]  Satoru Iwata,et al.  A combinatorial, strongly polynomial-time algorithm for minimizing submodular functions , 2000, STOC '00.

[59]  Nader H. Bshouty,et al.  On the Fourier spectrum of monotone functions , 1996, JACM.

[60]  Pravesh Kothari,et al.  Learning Coverage Functions and Private Release of Marginals , 2014, COLT.

[61]  Vahab Mirrokni,et al.  Maximizing Non-Monotone Submodular Functions , 2007, FOCS 2007.