Gauge Newsletter January 2020 | Page 59

DEEP LEARNING “ To understand Deep Learning, you should have a crystal clear idea of Machine Learning. Machine Learning is one of the key subfields in the universal set of Artificial Intelligence. If someone asks what Machine Learning (ML) is, what would the perfect answer be? The common idea of the majority is that it is just a bunch of if-else statements, which is not correct. ML is the study of algorithms that try to learn from the given data which will be used to predict future scenarios for new data or instances. There are many traditional learning approaches such as linear regression, logistic regression, SVMs, etc. But deep learning has become the trending approach these days due to various reasons. ” Coming back to our question, what is deep learning? It can also be identified as a Machine Learning technique, but it has been inspired by the architecture of the human brain, which is made up of millions of neurons. The basis of deep learning is, creating artificial neurons using algorithms. These advanced algorithms are not new to this century. Due to the low processing power in the computers and lack of data, these algorithms have been ignored for a long time. But the transition from these traditional computers towards the modern GPUs and data collection techniques operated over the internet, boosted Deep Learning to the next level. How does this deep learning really work? That’s a tough question to answer. Let me slowly unpack this black box for you. I will answer it using a supervised learning problem approach where we have labeled data. To use these DL (Deep Learning) methods what we need is a big data set. And you have to pre-process this data which can be used to feed our Artificial Neural Networks (ANN). Then, you can design your neural network architecture. ‘Keras’ is one of the famous libraries (which is my personal favorite) that can be used to work with DL that runs on top of TensorFlow. GAUGE Magazine University of Peradeniya Page 57