Credit: AI Trends
By Phil Marshall, PhD, Tolaga Research
At the Mobile World Congress (MWC) in Barcelona this February, it was hard to find a spot in the exhibition halls where artificial intelligence (AI) and machine learning (ML) were not visible in one way or another.
As usual, the event had no shortage of innovative use-cases, but there was a common concern at the show that many lacked compelling business cases. For a growing number of use-cases, the ingredients needed to derive business value comes from AI and ML capabilities. This creates tremendous opportunities for AI and ML solutions, so long as they are implemented pragmatically and without excessive technology over-reach.
The diversity of use-cases at MWC was immense. Customer-facing use cases included immersive compute applications, marketing and advertising, autonomous and assisted driving, customer experience management, and numerous digital bots and physical robots. These use-cases were demonstrated by a slew of companies including CloudMinds, Guavus, Qualcomm, and ZenDrive. Next generation AI and ML based service assurance solutions were demonstrated by companies like NEC/NetCracker, Netscout, RADCOM, Sandvine and Spirent. These solutions used AI and ML to address network complexities and the need for operational automation, particularly as networks migrate towards virtualized architectures. Other AI and ML use-cases on display at MWC included those for enterprise analytics, from companies like SAS, healthcare and public safety applications demonstrated by companies like IBM (Watson), Korea Telecom, Pharmacelera and Naru Intelligence.
Security concerns are on the increase, particularly as attacks become more sophisticated and mobile connectivity becomes integrated in critical infrastructure, such as electrical grids, smart cities, and autonomous transportation. Companies like Allot, Nokia and Wedge Networks showcased a variety of anomaly detection solutions based on AI and ML driven network heuristics. Device vendors like Huawei, LG and Samsung showcased impressive new platforms with advanced identity management solutions such as facial and iris recognition. While these solutions have already been available for several years, they are continually enhanced as training and inference algorithms become more sophisticated.
Although most AI workloads are processed today using Nvidia GPUs, it is widely recognized that purpose-built semiconductor solutions are needed. This has culminated in numerous startup companies and initiatives driven by tier 1 companies as they seek to optimize their existing solutions and target new market opportunities. Amongst the tier 1 companies at MWC this year, Huawei demonstrated a data center solution that incorporated its Ascend semiconductor technology, which had been optimized specifically for AI functions associated with network operations. Qualcomm showcased its Snapdragon 855 platform, which is optimized for a “laundry-list” of capabilities, including 5G new radio (NR), computer vision and AI, and immersive compute for augmented, virtual and mixed reality. Nokia had its ReefShark platform on display, which it originally launched in January 2018, to bring, amongst other things, AI functionality directly into its radio equipment. Both Huawei and Intel showcased AI for smart-retail applications. Intel demonstrated a self-service solution which used Xeon based edge computing and Intel’s OpenVINO toolkit for intelligent facial recognition. Intel also reiterated plans for the coming months to launch upgraded Xeon processor platforms with AI specific capabilities.
For the idle observer, it is easy to lose sight of the combination of data science, specific domain expertise, and raw compute power that is needed for AI and ML to be effective. The challenges with training and the pitfalls of poorly trained algorithms was illustrated by a smart retail application deep in the heart of the MWC exhibit halls. The retail application used computer vision in conjunction with ML to distinguish between apples and oranges. The application worked well when the fruit were orientated vertically. However, when nobody was looking, we tested the system with the fruit being at different orientations. When the orientation was anything but vertical, the results were highly inaccurate. It is not that the system was incapable of coping when the fruit was orientated differently, but rather that it was not trained with these conditions in mind.
As usual MWC had no shortage of innovative use-cases, but there was a common concern at the show that many lacked compelling business cases. For a growing number of use-cases, the ingredients needed to derive business value comes from AI and ML capabilities. This creates tremendous opportunities for AI and ML solutions, so long as they are implemented pragmatically and without excessive technology over-reach.
For more information, go to Tolaga Research.