Google Transcribes House Numbers, Google acquires DeepMind, Google Brain, Facebook sentiment analysis from user comments and posts, Google hired Geoffrey Hinton, Facebook’s Image recognition, Facebook Hires Yann Lecun, Yahoo Acquires Startup LookFlow, Baidu opens a deep learning lab in the Silicon Valley, Netflix moves into deep learning research to improve personalization, Amazon’s announcement on to their mobile shopping app that allows users to point their phone at an item and automatically check the item’s price and availability on Amazon. These are some of the recent moves by the technology giants who all having their vision on future. So, Deep learning is one of the research area which is rapidly growing.
What is deep learning ?
Deep learning is one of the ways to do big data analytics where computers learn without any human inputs/training.
The layman definition would be, teaching computers to think which is better than teaching a system explicitly. Deep Learning uses its core ideas from neural networks and then builds upon them. This is as similar as a baby learns sounds, then words and then sentences.
Neurons in Action
Human brain will think with the help of neurons to identify the relationships, representations etc. With the similar idea, Neural Networks identifies the relationship of each node (neuron) from the starting node where input dataset has been given. This process will be repeated for each node to find the relationship between each and every node. The same learning process repeats as the neural network layer increases which leads to deep learning. How Deep learning achieves Accuracy? Just like how a new born baby repeatedly learns its environment and mature over the age, deep learning process will learns continuously in a massive data environment to identify the hidden relationships and representations of its neurons. In this learning process, over the period, the model will be matured to give accurate results. Ex. It can Identify a “cat” can be closely related to lion, but not to anaconda.
Slow learners
Since this deals with unsupervised learning to identify relationships between each and every node, the learning process will be very slow and can be achieved using GPUs. Python is used extensively for deep learning. Theano is one of the python library which has a feature train the deep learning model using GPU.
Applications on Text data
-
Concept Matrix is nothing but a measure of semantic distances between words. Having concept matrix and applying deep learning on the given text content helps in efficient categorization. i.e, people like Obama, Lincoln can be categorized in to president.
-
Most of the sentiment analysis models work looking at individual words to give positive for postive words and negative points for the negative words and finds the sentiment by summing it up. Here the information might be lost since the words order has been ignored. Using deep learning models can develop a representation of entire sentences based upon the sentence structure. Here the sentiment analysis depends on how the words form the meaning of entire sentence/phrase. So, the model will be more accurate and cannot be fooled.
-
Paraphrase detection.
An Emerging trend
Apple’s SIRI, IBM’s WATSON are some technology milestones in deep learning. As we can see, the big technology companies of the world are making faster moves into deep learning. We can expect machines which can think ahead of humans in the near future.
– Santhosh Kumar
thebigdatascience 