Potentials and Challenges
of Synthetic Image Data in Manufacturing Companies in Switzerland
The Swiss MEM industry is experiencing robust growth in the application of artificial intelligence, especially in the field of computer vision solutions. This technology is being advanced by various companies in collaboration with different institutions such as colleges and universities. The majority of use cases we have encountered so far contribute to increasing efficiency and quality, although the position in the value chain can vary greatly. Last quarter, we at Synthetic Future had the opportunity to accompany a paper from the CAS AI Management program at HWZ, Zurich University of Applied Sciences in Business Administration. We acted as industry partners with the aim of better understanding the users of computer vision solutions in order to adapt our service to market requirements and thus make informed strategic decisions.
The Insights Gained
The Swiss MEM industry has promising potential for the use of artificial intelligence (AI) and computer vision (CV). The sector is characterized by high innovation capability and openness to technology. CV is already well-established and is used for quality checks and sustainability. Companies view data as a strategic asset and increasingly want to collect and manage end-customer data to improve their digital services. Although real-time data has been primarily used so far, synthetic image data could gain importance in the coming years as acceptance for CV with ML and deep learning models continues to grow.
The MEM industry uses computer vision (CV) in production to increase efficiency, product quality, and sustainability. CV technology is primarily used for quality and conformity testing as well as quality assurance, while other application areas are used sporadically. CV enables versatile solutions for efficiency improvement in various industries.
For the use of synthetic data, five factors were identified in particular:
1. Security
Data security is of utmost priority, as synthetic data is viewed as “global data” and integrated into the production lines of target customers through digital service packages, thereby meeting strict standards and governance requirements to ensure the development team has secure access to the data for training the ML models.
2. Robustness
The dependence of algorithms on data and the robustness of the data are crucial, as changes in the data can lead to different decisions. Therefore, it is important to evaluate the robustness of the data when developing algorithms to ensure traceable decisions.
3. Quality
The quality of synthetic data is crucial for the results of the developer teams of ML models, as they have high demands on the quality of the results and the acceptance criteria depend on reliable, safe, and efficient data suppliers who must provide high-quality and consistent data to train accurate models.
4. Interpretability and Explainability
Computer vision applications with complex deep learning algorithms require strategies to enable a better understanding of machine learning, as the unclear use of ML (such as with Deep Fakes) is perceived as a threat. The question of compromises in the accuracy of image processing to ensure explainability is challenging and requires clear documentation of the decisions made.
5. Acceptance and Trust
As an external data provider, it is important to consider the different levels of knowledge and acceptances of synthetic data among target customers and to assist in clarification, while also gaining the trust of the development teams in the unknown data provider to enable close collaboration for the development and training of ML models.
Summary of Opportunities and Challenges in Market Entry
Synthetic Future has potential in the MEM industry, as CV technologies are widespread and acceptance for ML-based solutions is increasing. The technological openness of companies allows the integration of synthetic data, while unused use cases offer further opportunities. However, there are also challenges such as limited access to industry insights and the heterogeneity of the market. The limited scalability of the data product and marketing pose further obstacles. Overcoming the “Not-invented-here syndrome” in development teams is crucial to achieving successful collaborations.
Think Connectedly, Act Together
Synthetic Future would like to express its heartfelt thanks to Nicole Unternährer Rogenmoser. Her work on the potentials and challenges of synthetic image data in the Swiss MEM industry has provided us with valuable insights and findings. It was a pleasure for us to participate in this project as industry partners. We also extend our gratitude to HWZ for the fruitful collaboration. Looking forward to more joint projects!
