Several years ago, an angry man stormed into a Minnesota Target store and began shouting at the location manager. The man was angry because Target had sent a series of coupons for baby cribs and infant clothing to his 16-year old daughter, and he believed the megastore was encouraging his teenager to get pregnant. The truth of the matter was even more troubling. It turns out his daughter already was pregnant, he just didn’t know it yet. So how did Target?
Businesses have long made good practice of collecting customer details in order to generate hyper-targeted marketing incentives. Now, in the age of big data, these same businesses have the capacity to crunch huge volumes of statistical information for the purpose of building accurate customer portraits. Hundreds of raw data points – from credit card transactions to loyalty rewards – get pumped into a computer, which then generates purchase predictions based on consumer history, habits, and subtle behavioural changes.
In the case of the Target pregnancy debacle, an algorithm may have picked up that the girl had purchased pregnancy tests, zinc and magnesium supplements or fragrance free body lotion – products that, when purchased in tandem, are often dead giveaways for an early-stage pregnancy – and issued a coupons booklet to entice her to build her nursery with Target products.
While this particular marketing tactic crossed the privacy line and was deservedly criticized, the takeaway is that the technology to harness relevant and effective insights into customer behaviour has reached profound new levels.
We’re working on deep learning methodologies in our Research & Innovation Lab that can take the machine learning capabilities described in the Target case to the next level.
School’s Out for Server
To qualify that extraordinary claim, it’s important to provide some context. In traditional machine learning, computers analyze data points to create predictive models on everything from customer behavioural trends and how consumers will react to a product, to pinpointing fraud and risk potential. But traditional machine learning requires human experts to tell computers how to interpret the data – a process known as feature engineering – and its analytic capacity plateaus after it hits a certain volume of information.
Deep learning, a subfield of machine learning, teaches computers to learn like humans. This gives machines an exponentially higher capacity to absorb knowledge from data by performing in a fashion analogous to the way the human brain processes, represents, and uses information.
In simpler terms, we’re teaching computers how to think like us. And the higher the amount of information, the better deep learning works. Now instead of thousands of parameters, we can use millions or billions of parameters, making the system virtually impossible to over-saturate.
Perhaps even more impressively, deep learning can model any type of data – like text, video, or voice – across a wide spectrum of fields, allowing the computer to combine and exchange insights from multiple domains and eliminating the need for the trial-and-error results of human guesswork. Incredible.
Facing the Future
Problems that have confounded the most brilliant minds in the artificial intelligence community for decades such as image understanding, speech recognition, and natural language processing, are now becoming a reality and the world’s most innovative corporations are already investing heavily in a deep learning future.
For example, Facebook currently has the world’s most advanced face recognition technology, a natural application of deep learning that can identify faces in a photo at near-human accuracy.
Big deal, right? That’s hardly new terrain.
But the same technology can also extrapolate information about the person in the photo by recognizing how he or she interacts with the other people in the images and what activities he or she likes to do. From there, Facebook can construct a far more personalized and accurate portrait of the individual, which can then be used to map preferences and personality traits that get turned into hyper-targeted, relevant advertising.
As a clue to how powerful and ubiquitous deep learning will be in the near future, Google has already started to snap up the best scientific minds at top universities with deep learning expertise and enticed these world-class academics to work for them exclusively.
Shaping the Future of Human Experience
When you start to think of it in this context, the example of Target sending nursery coupons to a teen in Minneapolis seems quaint. But deep learning opens up a new frontier of technologies that have the potential to shape the future of human experience. We’re passionate about the research we’re doing in deep learning to harness this incredible technology for business innovation in industries like retail, telecommunications, finance, and healthcare.
Just what will those innovations look like? Stay tuned to find out!
About the Author
With an exemplary reputation in his field, David's vast experience as a technology consultant and solution architect has spanned more than 15 years in some of the most complex business environments throughout Canada and around the world. David's focus on creative solutions is based on a strong belief that traditional software development practices are flawed, and his team routinely demonstrates a better approach with open source, agile methods and cloud computing.Follow on Twitter More Content by David Suydam