Trends such as artificial intelligence, deep learning, data analysis, IoT and other concepts are no longer new. But today, they are becoming increasingly intertwined, making technical advances possible that sounded like a dream of the future just a few years ago. Artificial intelligence (AI) is particularly worthy of mention here. However, with the increasing spread of AI in all sectors, there is also a growing need to make it available, accessible and applicable for engineers and scientists from different disciplines.

The complexity of ever larger data sets, cloud computing, implementation in embedded applications and larger development teams present developers with new challenges. To meet these challenges, solution providers will work on ways to increase collaboration and interoperability. This will lead to more productive workflows and less dependence on IT departments.

1. artificial intelligence - not just for data scientists

Engineers and scientists, not just data scientists, will drive the adoption of deep learning. Not only the openness to new techniques, but also the economic benefits that artificial intelligence and automation tools promise will drive the increasing use of AI by engineers and scientists.

New workflow tools simplify and automate data synthesis, labeling, pre-processing and delivery. These tools also extend the range of applications from image and computer vision to time series data such as audio, signal processing and IoT, For example, AI will be used outside of data science in unmanned aerial vehicles (UAV) to recognize objects in satellite images using AI or in cancer screenings for early disease detection.

2. industrial applications require specialization

Industrial applications are becoming an important field of application for AI, but also require greater specialization. In order for IoT and AI-driven applications such as smart cities, predictive maintenance and Industry 4.0 to go from visionary concepts to reality, a number of criteria must be met. Safety-critical applications, for example, require greater reliability and verifiability. Whereas advanced mechatronic systems require design approaches that integrate mechanical, electrical and other components.

Another challenge is that these specialized applications, such as systems for detecting overheating in aircraft engines, are often developed and managed by decentralized development and service teams. They are therefore not centralized under IT.

3. interoperability

To build a comprehensive AI solution, collaboration between different systems, programs or platforms is essential. The reality is that there is no single framework that can provide the best solutions for all application areas of AI. Currently, each deep learning framework focuses on a few applications and production platforms, while effective solutions need to bring together parts from several different workflows.

This creates friction and reduces productivity. Companies like ONNX.ai are tackling this problem. They offer an environment in which developers can freely choose the best tool, share their models more easily and use their solutions on a wide range of production platforms.

4. cloud computing

Public clouds in particular are increasingly being used as a host platform for AI. They will continue to evolve to reduce complexity and will reduce dependency on IT departments.

Powerful GPU (Graphics Processing Unit) instances, flexible storage options and production-ready container technologies are just three reasons why AI applications are increasingly cloud-based. For engineers and scientists, cloud-based development facilitates collaboration and enables on-demand use of computing resources instead of buying expensive hardware with a limited lifespan. However, cloud, hardware and software providers recognize that these technology platforms are often difficult for engineers and scientists to set up and use in their development workflows.

5. edge computing

Edge computing will enable artificial intelligence applications in scenarios where processing must be done locally. Edge computing for powerful, increasingly complex AI solutions in real time is made possible by advances in sensors and energy-saving computer architectures. Edge computing will be particularly important for safety in autonomous vehicles, which need to understand their environment and assess traffic situations in real time on this basis.

6. stronger cooperation

The increasing use of machine and deep learning in complex systems will require many more employees and greater collaboration. Data collection, synthesis and labelling will increase the scope and complexity of deep learning projects and require larger, decentralized teams.

System and embedded engineers need flexibility in deploying inference models in data centers, cloud platforms and embedded architectures such as FPGAs (Field Programmable Gate Array), ASICs (Application-Specific Integrated Circuit) and microcontrollers. In addition, they must have expertise in optimization, energy management and component reuse. Engineers who develop inference models need to bring this information together. They need tools to evaluate and manage the ever-growing amount of training data and for the lifecycle management of the inference models that they pass on to system engineers.

The author: Dr. Frank Graeber, Manager Application Engineering at Mathworks