The field of computer vision has made great strides over the last two decades, with significant contributions from deep learning techniques during the last five years. Today, as self-driving cars that perform complex, real-time visual recognition tasks are within technological reach, let’s take a look at the most fundamental vision task - that of classification. In this post, we will classify Devanagari vowels using TFLearn, a deep learning framework built on top of TensorFlow.

One of the many things our brains do very well during language processing, is the task of abstracting out details, keeping only information that is relevant, and then proceeding to form a quick summary. For example, we could read a sizable paragraph of text (like an article), and come up with a one-line summary or a heading that best describes it.

Many times, we can also gauge the underlying mood that a passage of text conveys. For example, upon reading a product review on Amazon or eBay, we get an idea of whether the buyer was “happy”, “angry”, “disappointed” or something else. After reading a movie review, we know broadly whether the movie is “good” or “bad”. Our brains are naturally adept at extracting “emotion” from mere text, which necessitates the following question: Can a computer do it ? Given text (be it a single line or several pages), can a machine capture the emotion or sentiment that the text conveys ?

This is the fundamental question that Sentiment Analysis attempts to answer. It is an area of study that falls under the wider field of NLP.

In this post, we will attempt to address this question using movie reviews as an example.

The gradient descent algorithm is widely used across several machine learning techniques, and forms the backbone of neural networks and backpropagation.
In this post, let’s try to understand how gradient descent works, and also get a mathematical intution behind this ubiquitous technique!