1. physics-based AI - models take into account rules and principles of the real world

As AI is expanding into more and more areas of research, such as complex technical systems, AI models need to take physical constraints into account in order to be relevant overall. The combination of data and physics, for example via neural ODEs (ordinary differential equations) or PINNS (physics-informed neural networks), has great potential. Simulations are at the heart of physics-based AI: Complex models can be configured as variants within a simulation and allow developers to quickly switch between models in order to obtain the best possible and most accurate solutions.

Reduced order modeling (ROM) with physically based reduction models is also an important new trend. By using AI, simulations can be accelerated by replacing an extremely computationally intensive first-principles model of a system - while maintaining accuracy.

2. collaboration on AI - free access to AI continues to spread

Researchers, engineers and data scientists should continue to expand their cross-functional and cross-industry collaboration to think about solutions from different perspectives. To make the latest models available on demand and enable users to build on the latest research results in the shortest possible time, network-based version management services for software development projects such as GitHub are a good option. Open source solutions are also becoming increasingly popular, as engineering teams often work with models from different frameworks. Stronger networking between science, academic research institutions and companies is also driving AI research forward, benefiting researchers and users, for example in areas such as physics-based machine learning and biomedical image processing.

3. companies are focusing on smaller, easier to explain AI models

AI users are increasingly realizing that they need to be able to provide models, adapt them to the hardware and provide explanations for the decisions made by the models in order for these models to be relevant. The explainability of models and corresponding applications are therefore increasingly becoming the focus of engineers.

In order to meet the requirements for cost-effective devices with low power consumption and explainable outputs, engineers are therefore increasingly turning to traditional machine learning models and parametric models. These are compact and have low memory requirements and meet the application requirements by making the output easy to interpret. When newer, more memory-intensive models are required, quantization and pruning techniques offer ways to compress the models, reducing the model size with minimal impact on accuracy. So, if needed, engineering teams can leverage interpretability, quantization and pruning to extend the use of AI, including deep learning and traditional machine learning models, to traditional model development.

4 AI is becoming crucial for the design, development and operation of modern technical systems

As AI becomes more prevalent across all industries and applications, AI will become critical to future technological advancement and the development and operation of modern engineering systems. In more established fields of activity where AI has only recently been introduced, engineers often need additional background information on this technology as well as specific reference examples to integrate AI into their work. Based on proven examples, engineering teams can contribute data and their know-how to such examples and extend them to integrate AI specifically adapted to their task.

What challenges await AI engineers

What challenges do these developments entail? As different teams are often responsible for the creation and implementation of AI models, complex challenges arise in the AI environment that engineers still need to overcome. The selection of pre-processing algorithms and model training, for example, usually fall within the remit of data scientists, who focus on accuracy and robustness. However, engineers must also take many other criteria into account for successful porting to the target platform. Early testing of algorithms for a feasibility assessment (e.g. using PIL = Processor-in-the-Loop) can prevent already trained and sometimes very powerful models from having to be discarded in the end.

The training of the AI is also usually implemented in a different programming language than the implementation in the hardware. However, models from the training environment cannot simply be executed on the target hardware without further ado. To overcome barriers between scripting languages, there are runtime interpreters (such as TensorFlow Lite), machine learning compiler frameworks such as Apache TVM or automatic code generation in MATLAB/Simulink.

Finally, the safeguarding of AI models remains an important topic: while AI models are allowed to make mistakes in the training environment in order to learn and improve, errors after implementation on the hardware can lead to major damage in real existing systems. The question of reliable, objectively verifiable criteria for a model to be considered safe will remain an important area of research in the future.

Outlook

The introduction of AI has an impact on the entire organization, from interdisciplinary collaboration to the design of specific components. It is therefore crucial for engineers to identify use cases that align with their short- and long-term goals and implement them accordingly. As AI advances into all fields of activity, including safety-relevant areas, the focus will be on issues relating to model quality, language compatibility and security.