Deep learning is a new field of AI that’s already steadily becoming invaluable to businesses around the world.
AI technology is controversial, to say the least. Elon Musk is concerned enough about it that he’s put his money into a billion-dollar non-profit organization purely to keep an eye on the technology. The late great professor Sir Stephen Hawking, meanwhile, was equally skeptical as to whether the impact of the technology would be positive or negative.
However, the truth is that AI technology is currently already being used all over the world in many businesses to make them more efficient and cost-effective. The idea of intelligent, self-aware artificial life-forms is still a long way away!
Today, we’re going to take a look at the major AI breakthrough of the last couple of years – one that’s made AI applicable to business: deep learning.
What does it mean for the future of business?
Let’s get started…
So, what is deep learning?
We’ll keep this as non-technical as we can!
Essentially, deep learning is a class of machine learning algorithms comprising a collection of artificial neurons (known as a ‘deep neural network’) that are trainable mathematical units.
Essentially, the deep neural network is capable of ‘learning’ complex mathematical functions, which is of course the foundation of all computing.
You might be surprised to learn that actually, ‘deep’ networks have existed since the 1950s, but as our technology has advanced they have leapt forward exponentially in terms of power. Essentially, it has only become commercially viable over the past few years.
To give you some perspective, a ‘traditional’ deep neural network might have 8 layers and around 60 million parameters.
However, some of the most cutting-edge networks now work at up to 400 million parameters!
Why is the capability growing so quickly?
The main reason for the astonishing capacity of new deep learning networks is actually the recent increase in ‘big data’ – indirectly, at least.
With big data becoming more and more important in the world of business, technology has naturally developed to allow for very large data sets, and these sets are more and more available.
Essentially, there is only so much a learning network can do with small datasets. But the larger the data sets become – and they’ve been growing continually over the past few years – the performance capabilities of deep neutral networks can take huge leaps forward.
Why is the technology so invaluable to modern businesses?
Deep learning technologies lend themselves very, very well to certain tasks.
One of the best examples is identity verification and facial comparison. Identification fraud and theft has grown hugely in the last decade or so, and financial institutions are naturally keen to use any solutions capable of combatting this threat.
Deep learning is an ideal weapon in this fight, because it is very well suited to object recognition and image classification.
Under the old system, customers would have to visit a bank branch and present ID in order to verify their identity. This is inconvenient, as anyone who’s been forced to use their lunch break to go to the bank will tell you!
However, it’s also inconvenient for the banks themselves. In fact, figures suggest that manual ID checks cost banks hundreds of millions of dollars each year.
Deep neural networks are already a practical solution to this problem. It’s now possible for customers to take a selfie and then submit the picture directly to their bank, where a deep learning algorithm will verify the authenticity of the submission.
The process, despite being remarkably complex on a technical level, takes just seconds, saves the customer time and effort, and saves the financial institutions a huge amount of money and time.
What other examples are there?
ID and verification are just one of the areas where deep learning can be invaluable to businesses. Here are some of the others:
• Customer and client satisfaction. Machine learning has already started analyzing user activity in order to actively track customer happiness. Essentially, companies can spot when a customer is likely to close their account before they do so, allowing them to be proactive in helping the customer out. Anyone in business knows it’s far cheaper to keep an existing customer than it is to get a new one, so this is a deceptively valuable use of the technology.
• Market trend analysis. So many business successes are built on being able to spot trends and patterns before they appear. Algorithms are simply more effective at spotting trends than humans are, which can be invaluable in sectors as diverse as marketing and finance.
• Calculating risk. This is another invaluable financial use. Learning algorithms can accurately assess risk by looking at all the relevant data (defaults, spending patterns, financial data, etc.) in order to assess the risk of lending to a customer. This can also be invaluable in terms of tailoring products to suit individual demographics.
• Health monitoring. The advent of smart watches has made health learning a reality. A combination of apps can be used to set personal data through to the learning algorithms, allowing healthcare professionals to spot anomalies earlier on. The case of elderly people, early alerts can be invaluable.
• Customer service and delivery. The retail sector has benefited hugely in terms of efficiency from deep learning. Smart machines are now capable of easing the burden on employees in departments where previously, staff resources would have decided response time. By deciphering the intent and meaning within emails and delivery notes, machines enable staff to be more active in ensuring customers are happy.
• Price changing. In today’s hyper competitive retail environment, prices change quickly, and can fluctuate a great deal. Machine learning plays a huge part in helping online stores to remain competitive at all times, and stops them being caught out by competitor promotions. It also ensures customers can always get the best possible deals quickly.
Deep learning is still in its early days.
Traditional machine learning has been around long enough that the algorithms and technology used are generally pretty well understood. A tool like scikit-learn, for instance, will allow any skilled developer to build a traditional algorithm with relative ease.
However, deep learning is still in its early days. There are tools available – Facebook’s PyTorch and Google’s Tensorflow are two examples – and it is possible to create examples of deep learning quite easily.
There’s a difference, though, between getting an example to work and actually building an algorithm that makes a genuine commercial difference to a business. The truth is that a huge amount of experience is required to get results with deep learning. It pays to consult an expert.
It requires a lot of resource
It’s no coincidence that the best examples of deep learning (which we’ll go through in the next section) come from companies with huge resources like Google and Facebook.
Data and compute power are the two main reasons that deep learning is possible in 2018, as we already said. So without the two, your company simply won’t be able to get the same results.
GPUs can reduce the speed of calculations remarkably, and TPUs even more so. Companies like AWS and Google Cloud have both started leasing them to companies looking to get a head start.
You can still get some results without the latest technology, but the truth is that using older algorithms, you’re looking at taking weeks to do what the latest TPU could do in hours.
If a business wants to implement machine learning, it’s necessary for them to be truly agile in the way they operate.
By analyzing data, for instance, a business could discover that they’d be far more efficient if they completely changed the way their infrastructure worked, or if they changed their customer service processes.
What’s more, it’s likely that the business will have to experiment; to try out different processes depending on what the algorithms suggest.
Unfortunately, some older businesses are notorious for their inflexibility.
Learning how to be efficient can require time, effort and money upfront, with a mind to receiving greater rewards later on.
Some businesses may be unable – or indeed – unwilling to do this.
There are two options if you wish to introduce deep learning into your business:
• Consult an expert
• Hire an in-house team
A lot of businesses – especially larger ones – will always prefer to have an in-house team. However, this can be very expensive.
The average salary for a data scientist with deep learning experience in the US is usually over $100,000, and that’s not including bonuses, benefits and the like.
Plus, of course, deep learning isn’t a one-man job. Essentially, it will require the company to take on a whole new team – data engineers, a good manager who equally understands the technical side as well as the business goals, the purchase of expensive hardware, etc. The cost of building an internal AI team is often cost prohibitive for most organizations.
This is why it’s often preferable to hire an expert AI consultancy firm instead, that already has the people and processes in place that are ready to tackle your data challenges.
Siri and Cortana
Any time you use voice recognition systems, machine learning is in operation. That’s how the technology learns to mimic human interaction.
The theory is, of course, that the longer the machines listen to you talk, the more they begin to understand the nuances and semantics of the way you use language.
You can debate the merits of Facebook’s privacy controversy as much as you like, but there’s no denying their use of machine learning has pioneered the technology.
For instance, did you know that there’s a reason you’re not prompted to tag your friends as often as you used to be? It’s because Facebook’s machine learning has become strong enough that it now often recognizes the people in your photos automatically, based on whether you’ve appeared with them in photos before.
In the last couple of years, Google maps has become more advanced in its use of data algorithms to measure directions.
Using the anonymous location data from smartphones, the search engine algorithms are able to analyze the actual speed of traffic on the roads.
The result? That the maps app can suggest the fastest route at that specific time, rather than just suggesting the one that should be the fastest route in general.
As anyone who’s followed their GPS only to end up in traffic will tell you, this is invaluable.
Netflix was another pioneer of machine learning technology. In fact, it’s a truly integral part of their recommendation engine.
By analyzing what viewers watch, Netflix is able to suggest other programs they might like, and is more likely to keep them watching.
In terms of financial ROI, Netflix is a perfect example of deep learning. In fact, the company has valued their algorithms at about $1 BILLION per year, because of the impact they have on customer retention.
If you’re interested in learning more about the potential impact of deep learning on your business, contact Iconic Solutions today.