Big Data: Deep Learning with Tensor Flow

Will Google’s TensorFlow Fuel Your Next Breakout Product?

In the era of self-driving cars and the Internet of Things (IoT), it can be easy to get caught up in the “window effect”. Maybe you’ve been there already…on the outside of that window looking in jealousy. That feeling of “sure, deep learning and big data utilization are all perfectly well for Google and other tech giants, but it’s entirely out of my company’s league.” If you’ve thought that far into the matter, you’ve probably also reflected on the massive upside potential that these technologies hold (probably with a healthy dose of entrepreneurial jealousy, if we had to guess). 

But now you can stop wishing and start innovating. In a move that holds the keys to revolutionary change across countless sectors, Google recently open-sourced TensorFlow, its deep learning software. Deep learning fuels everything from self-driving cars to IoT sensors and much more. Will TensorFlow and big data integration fuel your next breakout product? Read on for four especially promising sectors for 2018’s deep learning revolution. 

1. Digital Maps

Do you have an incredible idea for the next big digital map? Have you always known you could one-up Google Maps if you only had the proper tech at your disposal?  Well now, with the open-sourcing of TensorFlow, you have a chance to test your mettle. 

Google uses TensorFlow (and now so can you) to reverse engineer traffic speed from anonymous location data. As opposed to the advertising industry, which gathers and then analyzes large data swaths in light of set marketing objectives, this particular use case is an example of “instant integration” of big data. In other words, data is gathered, correlated and processed, and then instantly reformatted for the user’s navigational benefit.      


This nearly instantaneous capability impacts a slew of other map features, from real-time accident updates to proper route selection. Recent experimenters with TensorFlow have taken things off the real road and onto the digital one for a bit of driving practice. Check out this enlightening post about one coder who modified the system to play Mario Kart with no human interference or aid. 

This innovation provides a great look at the decision-making framework implicit in TensorFlow, and how that will shape developers’ objectives, mindsets, and parameters. It also offers valuable insights into cross-system connections with TensorFlow.   

2. Fraud Detection

This is easily one of the most relevant and timely applications of TensorFlow. (Remember last year’s horrendous Equifax hack that we’re still picking up the pieces from?) 


Admittedly, deep learning has been fueling fraud detection for a few decades now. Strictly rule-based models very simply won’t work in this context: there are individuals on the other end of this equation who will gleefully dispatch those with one flick of the wrist. Deep learning is needed (and needed badly). 

As discussed in the Mario Kart article above, the magic of deep learning is that it allows you to set fluid “primary parameters,” and readjust them at will. In the Mario Kart example, the primary parameter that TensorFlow used to make decisions was the kart’s predicted turn vector. The predicted turn vector was then compared with the actual turn vector, and the system made the required adjustments in real time. 

So in the context of credit fraud, total monthly spending could (and should) be a primary parameter. Monthly spending is fluid; however, so it’s disastrous to work with concrete values here. If you set the expected value for this metric at $500/month, say, and an individual charges $500.50 to their Visa, they would get a fraud notice and potentially have their card locked over a fifty cent discrepancy. That’s why set-value thinking falls flat here. 

3. Biometrics

Biometrics are on track to impact everything from credit fraud rates to security clearances of all kinds. As cutting edge as biometrics may seem (and they are), there’s still plenty of room for enterprising entrepreneurs to up the ante. 


Points of failure still exist even in advanced biometrics. Facial recognition is one of the leading biometrics of 2018, and a great example (featured prominently on the iPhone X). One of its major drawbacks is its poor performance in bright sunlight, which eliminates your phone’s visual access to depth information. 

This is because the sun is a far more powerful infrared projector than the invisible IR laser your smartphone uses. Much like our earlier example, with Google’s TensorFlow using GPS locations to reverse engineer traffic speed, an enterprising individual or team could develop metrics for measuring moment-to-moment sunlight exposure, and calibrate your phone’s infrared levels accordingly.

Remember, now that TensorFlow has been open-sourced and a flood of updates and innovations are sure to follow, big data analysis relative to light exposure may not be so far-fetched. There’s a clear and present need for it, to say the least.          

4. Virtual Agents

This use case for big data and machine learning is one of the most intriguing. Virtual agents in online business settings are on the rise, with a growing number of companies opting for virtual customer care. But there’s more to the story than meets the eye. In many cases, virtual agents aren’t yet stand-alone entities. Rather, a human agent waits in the digital wings, and the virtual agent kicks any inquiries it fails to comprehend over to its human helper. The human agent then gives proper clarification to the virtual one, and the conversation continues between it and the site visitor. 

Clearly, this is not the ideal state of affairs, and large profits await the company who can nix the virtual agent’s reliance on a human counterpart. This can be accomplished with TensorFlow: Google’s software combines big data and machine learning for active real-time performance. This is precisely the combination that many virtual agents need in order to move from “text recognition” regurgitation machines to sophisticated digital agents capable of fluid conversations. There are rich rewards awaiting companies who successfully deploy their resources in this space.    

Summing Up

Google’s TensorFlow innovatively and imaginatively melds the realms of big data and machine learning. Whether you’re interested in advanced biometrics or creating the most seamless virtual agent on the market, you have your work cut out for you now that TensorFlow is available to all. 

Do you require assistance on your network or want to deploy your very own?
Then click the button below to get started!