agsandrew - Fotolia
It was 2015 when Dj Das realized machine learning "is not something that lives in its own cave" and that his company, Third Eye Consulting, had an opportunity to provide services in the space.
Today, 90% of the company's customer projects all have a data science component -- machine learning, natural language processing, deep learning and artificial intelligence, said Das, CEO of Third Eye.
"The biz opportunity is crazy,'' he said. "It is as [big] as the internet wave because everything can have a machine learning back end. Machine learning can help us in every [part] of our lives, not just from a consumer perspective but an enterprise perspective. It's crazy at the enterprise level."
That is due in large part to the fact that enterprises sit on huge amounts of data that they haven't yet analyzed, Das said. Now the compute power is available to make that happen. "I definitely believe this is a once-in-a-lifetime opportunity,'' he said of machine learning projects. "It's going to be crazy, but there is definitely a lot of work required in actual implementation, and we're in the early stage of the algorithms."
Several cloud providers are making machine learning as a service possible with new offerings. Last September, Google announced its Cloud Machine Learning product was publicly available in beta. The company also offers a Google Data Engineer certification aimed at businesses, partners and data scientists "who want to design, train and deploy accurate machine learning models to gain insights previously out of reach."
And in a Microsoft blog post, Harsha Bennur, U.S. partner technology strategist, Data and Analytics, discussed the "new monetary stream" for partners to provide customers with intellectual property and domain expertise as a service. Azure Machine Learning is "democratizing data and intelligence by making the tools and resources accessible to everyone,'' Bennur wrote.
SADA Systems also believes now is the time for providers to offer machine learning as a service. "We see a huge opportunity to help businesses [that] do not understand the value or possibilities of machine learning,'' said Simon Margolis, director, cloud platform, at SADA Systems. "We're able to pass on the value of these models to our customers and further assist in their fine-tuning for unique use cases."
Machine learning projects: Practical applications
Among the practical applications of machine learning are training a quality assurance model in manufacturing, for example, to determine whether a machine object is flawed or based on an image taken at the end of production, Margolis said. "It could also be applied to more complex problems such as determining if there's a causational trend to be found in blood samples of cancer patients. The overall theme is that machine learning allows us to 'train' models to make decisions based on complex inputs, often unnoticed by humans."
Das pointed to a Los Angeles-based company that Third Eye is working with that sells mobile data. The company gives away free data but makes money when customers upgrade and purchase data, he said. "The survival of the company depends on the transition of users to the premier package." Consequently, the company is conducting email, text message and ad campaigns to their user base to get them to transition to paid packages, and it needs to figure out which campaign will be most effective at incentivizing a user to convert to the premium package, he said.
Third Eye is analyzing user data, which includes a user's past interactions with the company. "You create a profile of every user. … You can target the user based on your understanding of that user." The prediction of what package to offer and to whom and at what time is a machine learning component, he said.
Third Eye is also working with a computer hardware company that has set up sensors on its hardware to identify the probability of a server's CPU crashing. The company is "ingesting about five data points," and Third Eye will extend that to about 100 data points that the company can analyze to understand how the component is performing. "They get data from the hardware once or twice per hour, and [Third Eye] will change that to 15 seconds," Das said.
CEO, Third Eye ConsultingDj Das
Similarly, machine learning projects will also be big in the automotive industry, which is collecting data from cars, analyzing it and making predictions on many things, such as when a car will fail, Das said. "It can tell a driver that based on current emissions we can predict in the next four months the car will need a service check." Machine learning can also tell an insurance company if a driver can be deemed as safe or unsafe based on the driver's profile, he noted.
One of the most common machine learning use cases is in the retail industry. Amazon, for example, is already using machine learning to look at a person's buying behaviors and make recommendations for other items he might like to purchase.
"Machine learning can help an industry in all these ways, and it's all beautiful. Everyone wins, and the technology is already there,'' Das said. "It's not something in the sky; it's something you can grab right now."
SADA has many customers using machine learning -- but not necessarily in the way people think of. For example, Margolis said, Google has a Vision API, which is a "pre-trained" model for visual analysis. SADA's customers "are able to feed the API an image and learn what the image is, where in the world it is located, and which items are in the foreground versus the background." The company will be able to help its customers "ingest images of people and determine how many people are present and what their individual sentiments are."
Right now, very few of SADA's customers have started training their own models for unique use cases, he said, "but I'd expect this to grow as more businesses learn of the benefits in using such technology."
Skills needed to pursue machine learning projects
Das is now looking to start a channel and said the components he has created can be resold and put on any cloud because all are Java based. The skills that partners will need include experience analyzing data and providing data insights.
"[Partners] need to know the pains of migrating data,'' he added. "It's dirty work. Ninety percent of the time the data is in bad shape -- some pieces are in the cloud, some are on an Excel spreadsheet on someone's laptop and some are in some other country. They have to take all of that and stitch it together."
Das said he always sees holes in data. "We need somebody like that to work with and extend their value proposition to the customer on top of their current offerings," he said, adding that it should be someone who can stitch data together to ensure data integrity and quality.
Ultimately, machine learning is a branch of data science and, as such, engineers with a data science background are especially well prepared to help build out training programs for models and help businesses implement them, Margolis said.
There are also more tools coming out for the machine learning community, which he said will help more traditional developers take advantage of machine learning.
"Even developers with little exposure to machine learning previously are able to quickly ramp up, thanks to detailed documentation and tool sets associated with Google Machine Learning and stand-alone Tensorflow,'' an open source software library for machine learning tasks, Margolis said.
Get insight into monetizing customers' data
Learn about opportunities for robotic process automation
Read about machine learning's educational challenges for users