What is the difference between the Machine learning and Deep learning ?

Deep learning and machine learning are similar and yet different ideas in artificial intelligence with the latter being a niche of the former.

Machine Learning is the general term that includes the algorithms and methods that allow a computer to learn patterns based on data without being programmed on a case-by-case basis. It consists of several strategies such as:

• Linear regression, logistic regression.
• Random forests and decision trees.
• Support vector machines
• Clustering algorithms
• Neural networks with few layers.

Deep Learning is a specific branch of Machine Learning which trains artificial neural networks with more than one hidden layer (usually three or more), allowing them to automatically learn hierarchy and learn increasingly complex patterns at each deeper level. Deep learning models are much more complex as compared to traditional machine learning, with many containing millions or sometimes billions of parameters, and they typically need much larger datasets to be high-performing. Although the traditional ML may involve hand-selected and hand-engineered features, deep learning may automatically find useful features in raw data, making it particularly useful with unstructured data such as images, text, and audio. But such a power carries a price because deep learning requires substantially more computational resources, increases the time to train the model, and requires specialized hardware.

Conversely, the classical models of machine learning are smaller, faster to train and understand, and can be applied to smaller datasets and case studies where explain ability is a key concern. The two methods are capable of solving similar problems, however, deep learning is effective in complicated tasks such as image recognition, natural language processing and speech recognition. In effect, although machine learning is the larger toolkit that allows computers to learn, deep learning is the powerful tool in that toolkit which has limited resources, but is best used in complex pattern recognition tasks.