Statistical Mechanics of Optimal Convex Inference in High Dimensions

被引:37
|
作者
Advani, Madhu [1 ]
Ganguli, Surya [1 ]
机构
[1] Stanford Univ, Dept Appl Phys, Stanford, CA 94305 USA
来源
PHYSICAL REVIEW X | 2016年 / 6卷 / 03期
关键词
BIG DATA; NETWORKS;
D O I
10.1103/PhysRevX.6.031034
中图分类号
O4 [物理学];
学科分类号
0702 ;
摘要
A fundamental problem in modern high-dimensional data analysis involves efficiently inferring a set of P unknown model parameters governing the relationship between the inputs and outputs of N noisy measurements. Various methods have been proposed to regress the outputs against the inputs to recover the P parameters. What are fundamental limits on the accuracy of regression, given finite signal-to-noise ratios, limited measurements, prior information, and computational tractability requirements? How can we optimally combine prior information with measurements to achieve these limits? Classical statistics gives incisive answers to these questions as the measurement density alpha = (N/P) -> infinity. However, these classical results are not relevant to modern high-dimensional inference problems, which instead occur at finite a. We employ replica theory to answer these questions for a class of inference algorithms, known in the statistics literature as M-estimators. These algorithms attempt to recover the P model parameters by solving an optimization problem involving minimizing the sum of a loss function that penalizes deviations between the data and model predictions, and a regularizer that leverages prior information about model parameters. Widely cherished algorithms like maximum likelihood (ML) and maximum-a posteriori (MAP) inference arise as special cases of M-estimators. Our analysis uncovers fundamental limits on the inference accuracy of a subclass of M-estimators corresponding to computationally tractable convex optimization problems. These limits generalize classical statistical theorems like the Cramer-Rao bound to the high-dimensional setting with prior information. We further discover the optimal M-estimator for log-concave signal and noise distributions; we demonstrate that it can achieve our high-dimensional limits on inference accuracy, while ML and MAP cannot. Intriguingly, in high dimensions, these optimal algorithms become computationally simpler than ML and MAP while still outperforming them. For example, such optimal M-estimation algorithms can lead to as much as a 20% reduction in the amount of data to achieve the same performance relative to MAP. Moreover, we demonstrate a prediction of replica theory that no inference procedure whatsoever can outperform our optimal M-estimation procedure when signal and noise distributions are log-concave, by uncovering an equivalence between optimal M-estimation and optimal Bayesian inference in this setting. Our analysis also reveals insights into the nature of generalization and predictive power in high dimensions, information theoretic limits on compressed sensing, phase transitions in quadratic inference, and connections to central mathematical objects in convex optimization theory and random matrix theory.
引用
收藏
页数:16
相关论文
共 50 条
  • [41] Convex and Non-Convex Approaches for Statistical Inference with Class-Conditional Noisy Labels
    Song, Hyebin
    Dai, Ran
    Raskutti, Garvesh
    Barber, Rina Foygel
    JOURNAL OF MACHINE LEARNING RESEARCH, 2020, 21
  • [42] On the Foundations of Statistical Mechanics: Ergodicity, Many Degrees of Freedom and Inference
    Chibbaro, Sergio
    Rondoni, Lamberto
    Vulpiani, Angelo
    COMMUNICATIONS IN THEORETICAL PHYSICS, 2014, 62 (04) : 469 - 475
  • [43] Convex and non-convex approaches for statistical inference with class-conditional noisy labels
    Song, Hyebin
    Dai, Ran
    Raskutti, Garvesh
    Barber, Rina Foygel
    Journal of Machine Learning Research, 2020, 21
  • [44] STATISTICAL-MECHANICS OF CLASSICAL AMORPHOUS MAGNETS IN 3 DIMENSIONS
    AUSLOOS, M
    CLIPPE, P
    JOURNAL OF PHYSICS C-SOLID STATE PHYSICS, 1976, 9 (13): : L351 - L353
  • [45] Tensor Graphical Model: Non-Convex Optimization and Statistical Inference
    Lyu, Xiang
    Sun, Will Wei
    Wang, Zhaoran
    Liu, Han
    Yang, Jian
    Cheng, Guang
    IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, 2020, 42 (08) : 2024 - 2037
  • [46] Statistical Mechanics of Low Angle Grain Boundaries in Two Dimensions
    Zhang, Grace H.
    Nelson, David R.
    PHYSICAL REVIEW LETTERS, 2020, 125 (21)
  • [47] STATISTICAL-MECHANICS OF KINKS IN 1+1 DIMENSIONS
    ALEXANDER, FJ
    HABIB, S
    PHYSICAL REVIEW LETTERS, 1993, 71 (07) : 955 - 958
  • [48] Cluster-level statistical inference in fMRI datasets: The unexpected behavior of random fields in high dimensions
    Bansal, Ravi
    Peterson, Bradley S.
    MAGNETIC RESONANCE IMAGING, 2022, 87 : 19 - 31
  • [49] Cluster-level statistical inference in fMRI datasets: The unexpected behavior of random fields in high dimensions
    Bansal, Ravi
    Peterson, Bradley S.
    MAGNETIC RESONANCE IMAGING, 2018, 49 : 101 - 115
  • [50] Human trimodal perception follows optimal statistical inference
    Wozny, David R.
    Beierholm, Ulrik R.
    Shams, Ladan
    JOURNAL OF VISION, 2008, 8 (03):