A Max-Sum algorithm for training discrete neural networks. (13th August 2015)

• Record Type: Journal Article
• Authors: Baldassi, Carlo; Braunstein, Alfredo
• Is Part Of: Journal of statistical mechanics. (2015:Aug.)
• Publication Date: 2015-08-13
• Subjects: 7 -- 11; 7/010 -- 11/050 -- 11/016 -- 11/045; Statistical mechanics -- Periodicals; Mechanics -- Statistical methods -- Periodicals; 530.1305
• DOI: 10.1088/1742-5468/2015/08/P08008
• Languages: English
• ISSNs: 1742-5468

Abstract

Abstract: We present an efficient learning algorithm for the problem of training neural networks with discrete synapses, a well-known hard (NP-complete) discrete optimization problem. The algorithm is a variant of the so-called Max-Sum (MS) algorithm. In particular, we show how, for bounded integer weights with q distinct states and independent concave a priori distribution (e.g. l 1 regularization), the algorithm's time complexity can be made to scale as per node update, thus putting it on par with alternative schemes, such as Belief Propagation (BP), without resorting to approximations. Two special cases are of particular interest: binary synapses and ternary synapses with l 0 regularization. The algorithm we present performs as well as BP on binary perceptron learning problems, and may be better suited to address the problem on fully-connected two-layer networks, since inherent symmetries in two layer networks are naturally broken using the MS approach.