In my opinion, statistics is the core of various disciplines (eg. Machine learning, Signal Processing, Six Sigma).

Machine learning uses a lot of statistical methods as in Support Vector Machine, Cluster Analysis, Bayesian networks, Deep Learning.

However, i think that statistics will continue to have an own identity indipendently from the field of application and, in the same way, Machine learning and other disciplines will continue to develop also own methods that won't be statistical based. 

In my opinion, statistics is the core of various disciplines (eg. Machine learning, Signal Processing, Six Sigma).

Machine learning uses a lot of statistical methods as in Support Vector Machine, Cluster Analysis, Bayesian networks, Deep Learning.

However, i think that statistics will continue to have an own identity indipendently from the field of application and, in the same way, Machine learning and other disciplines will continue to develop also own methods that won't be statistical based. 

Recall that much of machine learning (neural networks, statistical learning theory, evolutionary/genetic algorithms, (un)supervised learning, etc.) grew out of work in physics, eugenics, biology and information theory.

Information theory was introduced by Shannon in 1948 (and used terms like entropy because it was borrowing from physics).

The basic artificial neuron that almost everybody starts out with when first learning neural network algorithms is either the McCulloch–Pitts neuron or something close to it, and that dates back to 1943 (almost a decade before the biological mathematical model upon which much of neuroscience is based, namely the Hodgkin–Huxley model).

Reinforcement learning (and a lot of work in other learning paradigms) comes right out of outdated psychology (behaviorism). Groundbreaking work was conducted in the 1930 by Tolman & Honzik (and superior work in the 40s by Toman, Richie, and Kalish).

Correlation is so frequently measured by Pearson's r that it is often assumed this was the coefficient used, and this coefficient (along with regression and much else) grew out of Pearson's work in eugenics, such as his 1896 Regression, Heredity, and Panmixia or helping found the (originally) mathematical eugenics journal Biometrika (one of the leading statistical journals). Fisher, another founder of modern statistics, likewise saw his work as mathematical eugenics (he contributed 200 papers to the journal Eugenics Review alone, and was the Galton Professor of Eugenics). His most important contribution was the 14 editions of his Statistical Methods for Research Workers, and Fisherian statistics (null-hypothesis testing that so dominates so many sciences) and the statistical tests revolved around experimental design for Eugenics research.

Partition functions are from classical physics (along with Gibbs sampling, density functions, probability distributions, the gradient, many of the optimization methods such as Lagrange multipliers, etc.). "Statistical mechanics" basically was statistics before it was called "statistical mechanics".

Finally, many of the modern techniques that should be replacing t-tests, ANOVAs, and much of the GLM actually were developed by statisticians in the early 20th century (and, in fact, many of the most popular statistical measures/tests were known to be inferior to other methods at the time, but we lacked computers). Permutation tests were developed in the 30s by Fisher, Pitman, and Edgington. And I've yet to meet a Bayesian who wasn't aware that Bayes' theorem was developed centuries ago, even if they don't recall exactly what Thomas Bayes did develop and how important Laplace was.

In other words, machine learning already is (more or less) statistics, and like statistics it grew out of other fields and methods that are outdated for their original purposes (for example, most of the statistical methods that are in use elsewhere but developed for measurement error are too primitive for their original purposes). Also, many statistical methods should not really be used anymore and some modern statistics textbooks (e.g., A. J. Izeman's Modern Multivariate Statistical Techniques) include chapters on SVMs, ANNs, etc.

I would outline my views with the following reduction:

Machine Learning => 'Teach' simulators to solve problems => SIMULATION =>Interpret results => Statistics

Machine Learning Methods

– Classification Trees

– Neural Networks

– Random Forests

– Ridge and Lasso Regression

– Support Vector Machines

Hastie T., Tibshirani R., Friedman J. The Elements of Statistical Learning: Data Mining, Inference, and Prediction. 2nd Edition. New York: Springer