Data Science,Deep Learning,Neural Network

How Celebal Leverages Deep Learning to Solve Business Problems

17 Oct , 2016  

Deep Learning – is yet another buzz word and is the successor of Machine Learning and now a predecessor to Artificial Intelligence or simply put – AI. The intriguing thing is the number of articles that exist in our enterprise ecosystem, all emphatically pressing on the implementation of deep learning to solve real world problems. However most of them restricts themselves to uncomfortable algebra and calculus which too seems to be “inspired” from Stanford Courses (In case you are not aware of those, try googling CS231n and CS224d, they are awesome)

The challenging thing however is the capability to solve real “Business” problems. I quoted business because there is plethora of tutorials using Deep Learning to solve problems like Computer Vision for Auto driven cars or someone designing a TIC TAC TOE using reinforcement learning or someone predicting a Movie Review Sentiment etc. I have often realized that this gets the customer excited, but only at a curiosity level.

We at Celebal are building products that solve Business Problems using state of the Art Algorithms in Deep Learning for ex Restricted Boltzmann Machines in conjunction with Recurrent Neural Networks to build a Semantic Search Engine which allows users to Search for related objects at the same time, for example using this, a user in Manufacturing Industry searching for a inventory spare parts also finds the list of dependent objects in a hierarchical way. It also provides the list of replaceable parts similar to one you searched. Another such example is Document Summarization, Celebal is helping few legal organizations summarize their Cases using Deep Learning and save their effort and time quite tangibly. We have been able to achieve a cognitive compression of 22X using our platform, again leveraging algorithms from the Deep Learning Stack. We are also implementing the same concept at various organizations that are in the clinical research business and are looking to summarize drug discovery documents.


One great addition to Deep Learning last year was Tensorflow. Though it started low and almost lost the battle to CNTK but by the end of first quarter 2016, Tensorflow fought back and today it sits at the heart of Deep Learning models in most of the products that you would see in the ecosystem. Tensorflow extension to spark has been really awesome and we are now building few more products in Text Analytics space.

Having said that, all the major use cases of Machine learning in Modern Business ecosystem like Customer Churn, Segmentation, Offer Prediction, Fraud analytics etc. are going to be the major opportunities but it is evident already that any technology with a wow factor does not guarantee customer’s approval. It has to align to a business requirement, in other words – “Its Always the Use Case that derives an implementation not the Technology”.


Leave a Reply

Your email address will not be published. Required fields are marked *