Machine Learning evolved from the studies of pattern recognition and computational learning, and basically is the concept of building computers that can learn through experience without being explicitly programmed. Statistics, Data Science and programming lie at the core of machine learning. The ongoing Big Data hype and developments in new learning algorithms have further boosted the growth of machine learning. Data-intensive machine learning methods have been applied across numerous fields in science, technology and finance. There are many models available, which depend mostly on the data available, the approach applied, the output required and above all the varying circumstances. Here are some of the hottest trends in Machine learning as of Q2 2016:

## Deep Learning

Deep Learning is the biggest trend in machine learning right now, providing fast and powerful ways for machine learning and taking us one step closer to Artificial Intelligence (AI). Basically, deep learning is a more advanced and refined avatar of ANN (Artificial Neural Networks), which is a system of interconnected models or “neurons” as in the nervous system of an animal. These models have the capability to transfer data between them, as in real neurons, and are assigned “numeric weights” which can be tuned based on experience, making them adaptive to inputs. Earlier, it was not possible to build a multi-layer of such ANNs, a hindrance which deep learning has overcome, allowing us to experiment with 100+ layers of ANN. These sets of models which can be imagined as stacked layers of neurons, begin with the lowest layer taking the input of raw data/image, which is then sent to the next layer, such that the next layer has a more ‘abstract’ version of the data below it. This eliminates the need of manual extraction of data which used to be done earlier in *feature engineering*. With the onset of Big Data, more powerful data from a plethora of sources is available to power the deep learning models.

## Ensemble Learning

Ensemble Learning uses separate models and combines their results to solve a computational intelligence problem. Ensemble learning is mainly used to increase the probability of selection of the best outcome to increase the performance of the model, or to decrease the probability of selection of a poor outcome. The selection of learning models is very crucial in Ensemble learning as it affects the predictive performance of the output model. As the Model Complexity increases, the error in the model due to bias decreases, but error due to variance increases due to over-fitting. The most efficient model should maintain a balance between the two types of errors. There are many ensemble learning techniques that either give preference to reducing the error due to bias, as in *boosting technique* or that due to variance, as in *bagging*. Then there is *stacking,* where the learning model affects the bias and the variance error.

## Support Vector Machines (SVMs)

SVM is a new generation learning mechanism developed from statistical learning theory. It is based on the concept of a hyper plane that separates and defines decision boundaries. The output algorithm, as such, defines an optimal hyper plane that categorizes objects. Hence, SVMs have found wide-scale applications in character recognition and image classifications. The underlying principle of SVM is to transform a non-linear model into a linear model by mapping the original objects using mathematical functions called *Kernels* so as to make the set of data linearly separable. This basic idea is used for advanced classification and regression analysis in more complex multi-dimensional spaces.

### And what other machine learning trends would you add here?

*Featured image: www.lexalytics.com*

## Leave a comment