Nowadays, Machine Learning has become one of the supports of information technology and with that, a rather central, albeit usually hidden, part of our lives. Machine learning is around us, from detecting potential fraudulent credit card transactions, google page ranking, floor cleaning robots and face recognition to helping doctors diagnose patients. We likely find machine learning applications a dozen times a day and may not even realize it.
If we think about the ever increasing amounts of data becoming available there is good reason to believe that smart data analysis will become even more pervasive as a key feature for technological progress.
At the same time, type of data is a key ingredient to characterize learning problems. This is a great help when finding new challenges, since quite often problems on similar data types can be solved by using very similar approach. One sample can be found for natural language processing and bioinformatics which use very similar tools for strings of natural language text and for DNA sequences
From a high perspective Machine learning is about design algorithms that enable computers to learn, these algorithms can be classified into taxonomy, based on the desired outcome of the algorithm, more specifically we can found:
Supervised Learning: This is the most common ML approach. At supervised learning the value we want to predict is actually in the training data. In ML the goal of classification is to group items that have similar feature values, into groups. From classification approach we can find following specific algorithm types
|
Linear Classifiers – Logical Regression – Perceptron – Support Vector Machine – Naïve Bayes Classifier Quadratic Classifiers |
Neural Networks K-Means Clustering Boosting Decision Tree – Random Forest Bayesian Networks |
Unsupervised learning: Value we want to predict doesn´t belongs to training data. The labelled samples are not available. Unsupervised learning appears to be much harder than supervised one. In this case the goal is enable computers to learn how to do something that we don’t tell it how to do. Unsupervised learning has produced many successes like machines capable of driving cars.
Semi-supervised learning: This approach combines both worlds in order to generate the proper classifiers or functions
Reinforcement learning: In this approach algorithms learn policies of how to respond given an observation of the world. Every action produces some impact in the environment, and the environment generates feedback that guides the learning algorithm
Transduction: It´s very similar to supervised learning, but does not explicitly creates a function or classifier: instead, it tries to predict new outputs based on training inputs, training outputs, and new inputs.
For a wide variety of business tasks, machine learning is clearly superior to traditional software approaches. Iridium is a lead tech company when dealing with machine learning, neural networks and cognitive services that can help you to bait your business competitors with the power of Machine Learning
We work with standard frameworks and libraries such as TensorFlow and DeepLearning4J writing algorithms with Python and R