Local minima in training of deep networks

There has been a lot of recent interest in trying to characterize the error surface of deep models. This stems from a long standing question. Given that deep networks are highly nonlinear systems optimized by local gradient methods, why do they not seem to be affected by bad local minima? It is widely believed that training of deep models using gradient methods works so well because the error surface either has no local minima, or if they exist they need to be close in value to the global minimum. It is known that such results hold under strong assumptions which are not satisfied by real models. In this paper we present examples showing that for such theorem to be true additional assumptions on the data, initialization schemes and/or the model classes have to be made. We look at the particular case of finite size datasets. We demonstrate that in this scenario one can construct counter-examples (datasets or initialization schemes) when the network does become susceptible to bad local minima over the weight space.

[1]  E. Wigner.  On the Distribution of the Roots of Certain Symmetric Matrices , 1958 .

[2]  Kurt Hornik,et al.  Neural networks and principal component analysis: Learning from examples without local minima , 1989, Neural Networks.

[3]  J. van Leeuwen,et al.  Neural Networks: Tricks of the Trade , 2002, Lecture Notes in Computer Science.

[4]  Yan V Fyodorov,et al.  Replica Symmetry Breaking Condition Exposed by Random Matrix Calculation of Landscape Complexity , 2007, cond-mat/0702601.

[5]  A. Bray,et al.  Statistics of critical points of Gaussian fields on large-dimensional spaces. , 2006, Physical review letters.

[6]  Yoshua Bengio,et al.  Understanding the difficulty of training deep feedforward neural networks , 2010, AISTATS.

[7]  Klaus-Robert Müller,et al.  Efficient BackProp , 2012, Neural Networks: Tricks of the Trade.

[8]  James L. McClelland,et al.  Learning hierarchical category structure in deep neural networks , 2013 .

[9]  Razvan Pascanu,et al.  On the difficulty of training recurrent neural networks , 2012, ICML.

[10]  Geoffrey E. Hinton,et al.  On the importance of initialization and momentum in deep learning , 2013, ICML.

[11]  Surya Ganguli,et al.  Identifying and attacking the saddle point problem in high-dimensional non-convex optimization , 2014, NIPS.

[12]  Surya Ganguli,et al.  Exact solutions to the nonlinear dynamics of learning in deep linear neural networks , 2013, ICLR.

[13]  Jürgen Schmidhuber,et al.  Deep learning in neural networks: An overview , 2014, Neural Networks.

[14]  Oriol Vinyals,et al.  Qualitatively characterizing neural network optimization problems , 2014, ICLR.

[15]  Yann LeCun,et al.  Explorations on high dimensional landscapes , 2014, ICLR.

[16]  Jimmy Ba,et al.  Adam: A Method for Stochastic Optimization , 2014, ICLR.

[17]  Yann LeCun,et al.  The Loss Surfaces of Multilayer Networks , 2014, AISTATS.

[18]  Kenji Kawaguchi,et al.  Deep Learning without Poor Local Minima , 2016, NIPS.

[19]  Ohad Shamir,et al.  On the Quality of the Initial Basin in Overspecified Neural Networks , 2015, ICML.

[20]  Guigang Zhang,et al.  Deep Learning , 2016, Int. J. Semantic Comput..

[21]  Alex Graves,et al.  Asynchronous Methods for Deep Reinforcement Learning , 2016, ICML.

[22]  Daniel Soudry,et al.  No bad local minima: Data independent training error guarantees for multilayer neural networks , 2016, ArXiv.

[23]  Max Tegmark,et al.  Why Does Deep and Cheap Learning Work So Well? , 2016, Journal of Statistical Physics.

[24]  Samy Bengio,et al.  Understanding deep learning requires rethinking generalization , 2016, ICLR.

[25]  Ohad Shamir,et al.  Distribution-Specific Hardness of Learning Neural Networks , 2016, J. Mach. Learn. Res..