By Artem Koval, Data Analytics and AI/ML Practice Lead, ClearScale
Interest in machine learning (ML) application development is increasing as organizations see the benefits ML offers. It’s enabling rapid innovation, driving efficiencies, and helping to meet customer needs at an unprecedented pace and scale.
The problem is that many businesses lack the resources — both the technical and the human kind — to develop ML apps. In part 3 of our 3-part blog series on ML, we discuss how the cloud is changing that.
The Resource Challenge
On the human resources front, few companies have in-house staff with the expertise to implement and support ML projects. Demand for those with ML skill sets is high and the field of ML is still relatively new. Therefore, finding professionals with the specialized skills required to build, train, and deploy ML models is difficult and expensive.
Most companies also lack the necessary infrastructure. ML entails training algorithms. That’s a compute-intensive, time-consuming task that requires extensive parallel computing resources.
Training an algorithm using traditional CPU-based processors typically takes days. Powerful graphics processing units (GPUs) significantly reduce processing time. That’s because they are specifically designed for ML, as well as artificial intelligence (AI) workloads. However, they’re expensive, and prices are expected to continue rising thanks to tariffs, the effects of COVID-19 on manufacturing and supply chains, and other factors. It’s particularly difficult to justify the costs if ML and AL projects are only conducted on an occasional basis.
ML also requires vast amounts of data — often in the petabytes — and immediate access to new data for algorithm training. That, in turn, drives the need for huge storage capacity. Those storage demands can vary greatly, based on the application and where it is in its lifecycle.
The Cloud and ML
The good news is that the cloud provides resources that facilitate ML app development. The result: it’s easier for companies to pursue ML projects — on their own or in conjunction with cloud-native app development companies.
There’s cost-efficiency courtesy of the cloud’s elasticity, pay-per-use model, and inexpensive data storage. For organizations just dabbling in ML to explore its potential, using the cloud can enable cost-effective testing and implementation of small projects, scaling them up or down as needs and demand change.
The pay-per-use model simplifies access to more complex capabilities without having to purchase new hardware. Organizations can access powerful GPUs and flexible, scalable storage without investing in expensive hardware. Without infrastructure to purchase and maintain, those organizations can also trade CapEx for OpEx.
Data Lakes, Microservices, and More
Yet another benefit: Data isn’t hampered by dependence on an established in-house repository. The cloud’s elasticity allows for customizing the amount of data and where it’s stored without requiring costly upgrades and system changes. In addition, data lakes housed in the cloud offer access to bigger, better data for training without straining your in-house resources.
Building a model in-house requires a significant investment of time in data collection, feature engineering, and model development. Using the cloud for ML also allows for leveraging cutting-edge technologies, services, and platforms—including pre-trained models and accelerators—to help speed up time to innovation and deliver results.
The cloud also enables the use of modern app development components and processes, such as microservices to automate a product or service or for targeted development. Additionally, the cloud allows for packaging applications into containers, making it easy to replicate results across multiple platforms.
Data lakes housed in the cloud give your organization access to bigger, better data for training without straining your in-house resources. As you deploy newer models and race towards the continuous innovation ideal, building on the resources available outside of your organization gives you a strategic advantage.
Depending on the cloud platform used, ML features can be implemented without requiring any special ML expertise or a team of data scientists. SDKs and APIs are available so ML functionalities can be directly embedded.
AWS: The Leader in ML
The big three cloud providers —Amazon Web Services (AWS), Azure, and Google Cloud— all support using either regular CPUs or GPUs to train models. They also all offer a broad spectrum of tools to facilitate ML app development.
AWS, in particular, is known for its in-depth service capabilities in the cloud ML and AI space. Developers without ML skills can use pre-trained AI services with continuously learning APIs. Those with more experience who want advanced functions can deploy AWS’s fuller-featured ML platform.
AWS also offers pre-trained AI services for computer vision, language, recommendations, and forecasting to build, train, and deploy ML models at scale. Tools that facilitate AI-based services include Amazon CodeGuru, Amazon Fraud Detector, Amazon Lex, Amazon Polly, Amazon Textract, and Amazon Comprehend, among others.
AWS also has several offerings in the hardware category, such as the AWS DeepLens wireless video camera. It can run deep learning models on what it sees and perform image recognition in real-time.
The Bottom Line
What it comes down to is that the barriers to pursuing ML app development — and taking advantage of the benefits of ML — have been lowered thanks to cloud resources. Of course, there’s a lot more that goes into successful ML app development than tapping the cloud.
Experience using specialized ML-centric cloud resources, as well as proficiency in cloud-native app development, can facilitate problem-solving and innovation. That lets you overcome the challenges that inevitably arise, and accelerate your time to market. Expertise with specific cloud platforms and the tools they offer – such as those available from AWS, also can yield significant benefits.
That’s why many organizations choose to partner with ClearScale. As an AWS Premier Consulting Partner, we have proven expertise in using AWS services to develop ML-powered solutions. Whether it’s Amazon Forecast to more accurately forecast supply chain demand or Amazon Personalize to deliver tailored search results, ClearScale is well versed in identifying and implementing the right AWS services to solve specific business needs.