It’s a good time to be a machine learning engineer. According to Indeed's best jobs in the US report for 2018, the position of machine learning engineer ranked fourth, behind commercial project manager (first), full stack developer (second) and computer vision engineer (third). In fact, machine learning engineers have seen an increase of 166 percent in the number of job postings from 2014 to 2017.
As AI and machine learning are being bandied about, many organizations find themselves asking if they are ready to bring in a machine learning engineer. And according to a Wired report, not all organizations are there yet.
Artificial Intelligence (AI) experts are in short supply, and big tech companies have made moves to address this. Google and Amazon each launched AI consulting services. According to the Wired report, Google CEO Sundar Pichai said last fall that there are only “a few thousands” capable of creating sophisticated machine-learning models. “If you’re a random manufacturing company in the midwest you may have money, but it’s hard to attract a $250,000-a-year Stanford PhD to work for you,” Diego Oppenheimer of Algorithmia told Wired.
So how do you attract machine learning engineers? CMSWire caught up with a few machine learning engineers to discuss their roles, their visions for technology landscapes at their companies and what they look for in a workplace. And having a firm grasp on your organization's data is the theme that resonated greatly.
Related Article: Machine Learning Will Transform the Advertising Industry
Understanding of Data Needs and Timelines
Sergei Makar-Limanov, machine learning engineer at mabl, works for, what he calls, a small startup. Their goal is to use machine learning and AI in order to improve and automate testing of websites and apps. In the past, Makar-Limanov has held titles leaning more toward data science. “I am, in general, excited about practical solutions to hard business problems,” Makar-Limanov said. “So if I am faced with a challenge where machine learning can add value I am excited to work on it.”
What does he look for from an organization as a machine learning engineer? What matters to him most as far as seeing the organization as appealing to join? “Aside from being in a supportive work environment,” Makar-Limanov said, “it's important that the upper management understands and supports the investments that are needed to make machine learning projects successful.”
This, Makar-Limanov said, means organizations need to have a good understanding and be supporting of the following:
- The kind of data that is needed, and the general approach to collecting data. Organizations would be better served to have a robust data warehousing solution that can support the necessary data queries, or a willingness to invest in architecting and building such a solution.
- A different approach to timelines. Most organizations subscribe to the agile methodology with two- or three-week sprints measuring milestones, but many machine learning-based projects don't fit well into such timelines with initial development often taking a lot longer, according to Makar-Limanov.
- A willingness to invest in a framework that measures the results of machine-learning-based applications. “Machine learning and AI algorithms are never right 100 percent of the time,” Makar-Limanov said, “so in order to understand improvement one needs to measure live performance and compare it against existing solutions.”
Encouragement to Participate in the Development Process
Technology needs vary from organization to organization. Makar-Limanov has worked at places that had in-house Hadoop clusters with Hive or Spark on top, and at other places which used the AWS Stacks with Redshift or Athena back-ends. Currently he’s working on the GCP infrastructure with BigQuery, DataFlow and Google Cloud Machine Learning. “Business wise I think that it's very important for the business to be data-driven and to recognize that engineering resources need to be devoted to data gathering and that data scientists need to participate in the development of the product to make sure that the right kind of information is being gathered,” Makar-Limanov said. “For example adding captchas to identify suspected robot traffic even when this might not be a strict business requirement.”
Related Article: Data, Search and the Foundation of Machine Learning
Patience With Testing
Eric Moller, chief technology officer AtomicX, a company that is using machine learning for chatbot technology, said his focus now is on applying machine learning techniques to conversations and text data to make better predictions. What does he look for from an organization as a machine learning engineer? “Access to lots of interesting data sets, smart coworkers and buy-in from the top,” Moller said. “Machine learning is still a fairly scientific pursuit. Applying it to a new domain often means, a lot of failure needs to be tolerated before you start seeing results.”
An organization also needs good, structured data for machine learning engineers to be successful. “A lot of companies are jumping on the machine learning bandwagon without any of the necessary prerequisites to make good of these new technologies,” Moller said. “Most important: you need to be collecting a lot of structured data and searching for deep patterns where classic statistical/regression techniques aren't providing adequate results.”
Making the Impossible Possible
Moller said he’s thrilled to see his company apply machine learning to problems that were always "impossible" to tackle using traditional computing paradigms. “We've mastered image recognition; machines can do it better than humans now,” he said. “At AtomicX we're applying machine learning to text and conversational systems. It's a very difficult problem since there's so much collective human experience embedded into our language.”
Moller’s also bullish on generative design, another aspect of machine learning he said has “incredible potential.” “We're already building lighter, stronger and more efficient structures and designs by simply feeding in the parameters we want optimized,” he said.
Related Article: Machine Learning's Secret Sauce? Getting Your Data Right
Incorporating Knowledge from Science, Mathematics
Esteban Pellegrino, chief scientist at Zimperium, leads the company’s efforts on machine learning and AI in Zimperium’s z9 mobile security engine. He sees his role as changing the world of mobile security by leveraging advanced modeling techniques to solve complex security problems. Pellegrino was initially trained as a nuclear engineer with a degree from Balseiro Institute. His role as a chief scientist is to evolve his team’s security solution in the mobile space. He works with the research, development and product teams to find suitable solutions that are flexible enough for future architectures of connected apps and devices — where machine learning could be part of the solution, or not.
Machine learning, Pellegrino said, combines “several areas of science” in order to find solutions to engineering and industrial problems. Machine learning engineers should be able to “easily incorporate new knowledge from diverse areas of science and mathematics.” They should also, Pellegrino added, be comfortable finding solutions to unsolved problems.
The most important thing for an organization to have for machine learning engineers is access to enough data in the scope of the problem the organization is trying to solve. “It is kind of the chicken/egg problem,” Pellegrino said, “because machine learning algorithms should be evaluated before having a clear answer if it is the correct technology and if the available data is enough — and a machine learning engineer is needed for that.”
Machine learning engineers want to be creative enough to find out-of-the-box solutions and construct data models with useful learning representations to efficiently solve the problem at hand, he added. “This might be a very unique iterative process for the organization that's going to change the infrastructure to better adapt it to suitable architectures for an efficient data extraction, processing and learning in the working pipeline," he added.
Related Article: How Machine Learning Will Tame the Explosion of Unstructured Data
The Future: ML’s A Tool for Machines
To some extent, machine learning will help us to optimize the human brain power we collectively use as a society to improve our way of living, Pellegrino predicted. "Even more," he said, "some problems are too complex and out of the scope for humans, that the only way to approach this is with machine learning algorithms.”
In the future, we’ll see machine learning becoming a tool for machines rather than humans. Machines, Pellegrino said, will dynamically learn and build models to have a better representation of reality without human intervention. “I think that's the most promising field of machine learning and artificial intelligence in general,” he said. “Creating autonomous agents that can create machine learning models on-the-fly as we, humans, do now.”