Both the number of parameters and size of models (see Figure 1) that refers to width and depth of the neural networks is skyrocketing.
Richer data means better predictive capabilities for businesses to anticipate customer preferences, trends, fraud, the climate – anything, really. But to analyze the data more effectively, companies need more computing power. Accelerated computing provides the computing power businesses need. In this article, we will introduce accelerated computing and its challenges in detail.
Figure 1. Increase of number of parameters
What is accelerated computing?
Accelerated computing is a computing approach employed in academic, research and engineering applications in which calculations are executed on specialized processors (called accelerators) together with traditional CPUs to achieve faster execution times. Accelerators are sophisticated microprocessors and AI chips built for data parallelism. In practice, they reduce execution times via offloading demanding work that can slow down CPUs, processors that normally perform tasks serially. Since accelerated computing combines CPUs and other types of processors on an equal scale, it is also called heterogeneous computing.
Accelerated data processing is already widespread. It is used in PCs, smartphones, and cloud services. Nowadays, companies also use accelerated computing to run their deep learning models. Neural networks need many parallel arithmetic operations and accelerated computing satisfies that need.
Why is accelerated computing important now?
Accelerated computing is one of the techniques that let deep learning models flourish since it provides effective computing power.
Deep learning and neural networks are not new inventions. In fact, Frank Rosenblatt, “father of deep learning,” invented it in 1962. There were lots of practical improvements in the use of DL in the 1980s. However, as Figure 2 shows, DL has only recently become popular because before the 2010s two things were missing:
- Training data: Researchers used to rely on hand-labeled data for machine learning. However, data generation has increased significantly, with the number of new data per year doubling every two years since the 2010s.
- Relevant hardware & cheap computing power: Neural networks require AI chips, processors that can support computing power needs of neural networks. Until the 2010s, such tools were non-existent or extremely expensive, so they could not be widely deployed.
- Storage: Thanks to technological developments we have more storage areas for data in both cloud platforms and in on premise areas.
- Deep neural network techniques: Novel developments in deep learning techniques.
Figure 2. The popularity of deep learning
What are accelerated computing use cases?
Accelerated computing is used in almost all industries. It helps in effective deployment of chatbots, fraud prevention and detection, weather forcasting customized sales/marketing efforts, etc. Here we present 3 examples:
Vision Banco: Effective credit risk calculation
Credit risk analysis requires significant computing power because so many customer-related and external parameters affect the accuracy of the analysis.
Vision Banco cooperated with H2O.ai for development of an algorithm for fast credit score analysis. However, Vision Banco’s existing infrastructure could not run the ML model effectively. They needed GPU acceleration to handle the demands of the model. With this in mind, Vision Banco started to use IBM Power Systems AC922.
The accelerated computing initiative enabled Vision Banco to increase the number of credit products accepted per customer. In addition, the duration of credit risk analysis was reduced from about a week to a few hours.
American Express: Fraud detection and prevention
According to McAfee, fraud costs the world $600 billion annually. Such an amount is equivalent to the entire GDP of some G20 countries. As a result, companies have been using AI applications to effectively detect and prevent fraud for some time. The problem is that fraudsters keep inventing new methods of theft that make monitored ML models useless. Behavioral analytics can enable better fraud prevention but it requires too much computing power.
By using accelerated computing hardware, American Express has improved its recurrent neural networks that detect anomalies in customer behavior. As a result, American Express provides a real-time fraud detection system and improved its accuracy.
The Weather Company: More accurate weather forecast
In 2016, IBM acquired The Weather Company. The caotic nature of the climate made the acquisition a perfect experiment to test IBM’s hardware capabilities. The Weather Company realized that by using high-speed GPU technology, they could accelerate weather forecasting to produce global, high-resolution weather forecasts for the next 15 hours and update them every hour and supply more accurate weather forecasts.
Such more accurate weather forecasting improves the entire value chain and our standard of living. For example, farmers can be informed about frost and protect their crops, urban emergency systems can prevent damage caused by extreme weather, and so on.
What are the challenges of accelerated computing?
Accelerated computing is simply a subset of hardware investment. Therefore, it is useful to highlight the challenges of hardware to understand the disadvantages of accelerated computing:
- Development time of hardware systems are long. For instance, even Intel, a giant tech company could build the Nervana neural network processor in 3 years.
- Development cost of hardware tend to be high since it takes long time and require to employ high skilled engineers.
- Updating features: When you build a hardware system, it is hard to change or maintain.
In-house hardware production is therefore only possible for companies that have experience, resources and the necessary human capital. Companies that lack these resources can buy hardware from third parties. You can check our article on top 10 AI Chip-makers article and our sortable list of companies working on AI chips to help you to find a correct vendor.
Don’t hesitate to ask your questions about accelerated computing:
Cem has been the principal analyst at AIMultiple since 2017. AIMultiple informs hundreds of thousands of businesses (as per similarWeb) including 60% of Fortune 500 every month.
Cem's work has been cited by leading global publications including Business Insider, Forbes, Washington Post, global firms like Deloitte, HPE, NGOs like World Economic Forum and supranational organizations like European Commission. You can see more reputable companies and media that referenced AIMultiple.
Throughout his career, Cem served as a tech consultant, tech buyer and tech entrepreneur. He advised businesses on their enterprise software, automation, cloud, AI / ML and other technology related decisions at McKinsey & Company and Altman Solon for more than a decade. He also published a McKinsey report on digitalization.
He led technology strategy and procurement of a telco while reporting to the CEO. He has also led commercial growth of deep tech company Hypatos that reached a 7 digit annual recurring revenue and a 9 digit valuation from 0 within 2 years. Cem's work in Hypatos was covered by leading technology publications like TechCrunch and Business Insider.
Cem regularly speaks at international technology conferences. He graduated from Bogazici University as a computer engineer and holds an MBA from Columbia Business School.
To stay up-to-date on B2B tech & accelerate your enterprise:Follow on
Next to Read
Your email address will not be published. All fields are required.