These are words from companies like yours who have invested in Data science but are wondering whether they are getting as much value as they could
“Hiring a data science team is impossible. We can’t find people with the right skills, and when we do, we don’t know how to evaluate them.”
“We don’t want some black box machine learning technology that no one can understand.”
“WE WANT TO BE MORE DATA-DRIVEN, BUT WE DON'T KNOW WHERE TO START. IT’S JUST NOT IN OUR CORPORATE CULTURE.”
“WE’RE NOT SURE IF OUR DATA HAS ANY VALUE.”
“WE’RE WILLING TO INVEST IN DEEP LEARNING, BUT IT FEELS LIKE ONLY GOOGLE AND FACEBOOK CAN DO IT AT STATE-OF-THE-ART LEVEL”
“THESE BI REPORTS ARE ONLY TELLING US WHAT WE ALREADY KNEW.”
“WE CAN’T PREDICT WITH CONFIDENCE BECAUSE THE BUSINESS TEAM CAN’T EXPLORE THE DATA AND DEVELOP THEIR OWN MODELS.”
“WE WANT TO PARALLELIZE OUR DATA OPERATIONS, BUT IT'S GOING TO BE TOO COSTLY.
OUR MODELS TAKE TOO LONG TO RUN, AND OUR RESULTS GET OBSOLETE FASTER THAN WE CAN GENERATE THEM”
"WE CAN’T MATCH OUR CRM DATA TO OUR LIVE SYSTEM, AND WE CAN’T REACT IN REAL-TIME.”
"REACTIVE STREAM ANALYTICS SOUNDS GREAT, BUT WE DON'T HAVE EXPERTISE IN-HOUSE. MACHINE LEARNING ON STREAMS IS IMPOSSIBLY HARD"
You believe there’s a competitive advantage in being more data-centric. You are getting close. You invest in your data, but decisions aren’t getting better just like magic. Now you need the culture and the strategy.
If you are ready for a change, our team has expertise on both the business and technology sides of data science.
DSR has developed the following solutions:
'Live Smile' recognition system for a marketing company that wanted to understand the (facial) reactions of their audiences to their advertising campaigns. Imagine a family sitting in their living room watching TV. The TV has a camera that tracks the faces of everyone sitting on the sofa. When an advert is shown on TV, they know whether it 'works' for each different customer segment that this family represent: teenager, middle-aged parents (male and female), senior, retired grandfather, etc.
General Question answering on visual input. For any picture, the user can type any question (for example, 'How many sheep are in the picture? or 'What color is the flower?') and the system will produce an answer that is on average 67% correct. That's for 'any' question (something that was considered sci-fi only a couple of years ago). For more restricted questions (say, only about position, or about counting etc), performance goes up to 85%.
A tool to identify twitter paths to targeted influencers. 'How to influence the right person on twitter'. Imagine you are a DJ. In the hope that a label 'MaxCool' would offer you a contract, you tweet your new sounds immediately after you finish them. 'MaxCool' doesn't follow you. In fact, no label is following you as of today. You are a nobody! Or are you? Well, a bunch of your followers are 'in the know' guys. They are trend setters, and sometimes they have followers themselves that may be connected to the label of your dreams. If only everyone in the chain down to MaxCool would retweet your productions! We worked on a personalized recommender that finds the right people to influence and the exact chain of people that connects you to them.
The use cases above use the following technologies:
- Deep learning. We have significant expertise in analyzing live video, which is more demanding than static photographs.
- Natural language processing (NLP), particularly question answering systems that use conversational language and rely on context.
- Large graph processing. Graphs that are hundreds of gigabytes are not a problem anymore with modern technology.
- Stream processing. Data that changes in real time is the new normal (aka 'fast data').