Deep Learning Overview

I’ve watched a few videos of this course in Udacity to give myself an insight of what Deep Learning is. I do Machine Learning but I don’t know much about Deep Learning yet. What I’m going to do here is to discuss the important and not-so-technical points I got from the videos.

Deep Learning is a branch of Machine Learning.

It has emerged as a central tool to solve perception problems including recognizing images, what people are saying, helping robots interact with the world, computer vision, and speech recognition.

Deep Learning is a much better tool to solve problems like discovering new medicines, understanding natural language, and understanding documents like ranking them for search.

The course is divided to four parts

  1. Logistic Classification; Stochastic Optimization; Data and Parameter Tuning
  2. Deep Networks; Regularization
  3. Convolutional Networks
  4. Embeddings; Recurrent Models

And it uses Python! Unfortunately as of this writing, I don’t know how to write code in Python (but I do R) so I’ll skip the exercises and just watch the video lectures for now.

Neural Networks became important because of Speech Recognition, Computer Vision, and Machine Translation. This further led to Deep Learning.

Classification is the central building block of machine learning. This is where Regression, Ranking, Reinforcement Learning, and Detection build up from. Example of Classification is detecting whether pedestrians or non-pedestrians are present in an image. A Logistic Classifier is a linear classifier. It takes an input, in our example it can be the pixels of an image and applies a linear function to them to generate its predictions. A simple logistic classifier can be then turned into a Deep Network.

In neural networks, increasing the size of hidden layers in the middle is not efficient because it gets hard to train. Adding more layers and making the model deeper (rather than wider) leads to parameter efficiency. Another one is that natural phenomena tend to have a hierarchical structure which deep models naturally capture. Deep models are applicable if the data is large enough to train.

I think this will suffice for an overview. More helpful videos are present in Udacity’s Deep Learning course but I may have to review Python first so I can add more. Thanks for dropping by!


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s