SGD Algorithms based on Incomplete U -statistics: Large-Scale Minimization of Empirical Risk

Abstract : In many learning problems, ranging from clustering to ranking through metric learning, empirical estimates of the risk functional consist of an average over tu-ples (e.g., pairs or triplets) of observations, rather than over individual observations. In this paper, we focus on how to best implement a stochastic approximation approach to solve such risk minimization problems. We argue that in the large-scale setting, gradient estimates should be obtained by sampling tuples of data points with replacement (incomplete U-statistics) instead of sampling data points without replacement (complete U-statistics based on subsamples). We develop a theoretical framework accounting for the substantial impact of this strategy on the generalization ability of the prediction model returned by the Stochastic Gradient Descent (SGD) algorithm. It reveals that the method we promote achieves a much better trade-off between statistical accuracy and computational cost. Beyond the rate bound analysis, experiments on AUC maximization and metric learning provide strong empirical evidence of the superiority of the proposed approach.
Complete list of metadatas

Cited literature [28 references]  Display  Hide  Download

https://hal.telecom-paristech.fr/hal-02107492
Contributor : Stephan Clémençon <>
Submitted on : Tuesday, April 23, 2019 - 4:37:57 PM
Last modification on : Wednesday, September 11, 2019 - 1:28:17 AM

File

5819-sgd-algorithms-based-on-i...
Files produced by the author(s)

Identifiers

  • HAL Id : hal-02107492, version 1

Citation

Guillaume Papa, Stéphan Clémençon, Aurélien Bellet. SGD Algorithms based on Incomplete U -statistics: Large-Scale Minimization of Empirical Risk. SGD Algorithms based on Incomplete U -statistics: Large-Scale Minimization of Empirical Risk, 2015, Advances in Neural Information Processing Systems 28 (NIPS 2015). ⟨hal-02107492⟩

Share

Metrics

Record views

20

Files downloads

17