In the recently published paper, Trustworthy AI Inference Systems: An Industry Research View, researchers from academia and industry offer input on the design, deployment, and operation of trustworthy AI inference systems.
The researchers dissect and review current AI inference systems, proposing that customers, institutions and regulators are the critical components in the current AI and privacy landscape at the macro level; while at the micro level,trends in AI and privacy continue to affect how researchers and AI practitioners think about data privacy technologies, both reflecting and influencing the interactions between customers, institutions, and regulations.
The researchers also identify complex infrastructures that include the cloud, end devices and base stations as contact points between customers, institutions, regulations and AI systems, where the deployment of trustworthy AI inference systems can address most concerns associated with data privacy and IP protection.
The paper suggests AI inference systems adopt Privacy-Enhancing Technologies (PETs) that can harness customers’ data at any time. Such systems should also leverage appropriate security protection mechanisms for AI models while offering customers timely, informed, and customized inferences to aid in their decisions.
The researchers say current trends in privacy research and technology development can be expected to evolve technologies that are more suitable and ready for mass deployment. For example, most research in the field focuses on developing accurate privacy-preserving classifications using CNN. In services such as Machine Learning-as-a-Service (MLaaS) that are offered by all major cloud service providers, privacy technologies ensure that customers can issue classification queries without exposing any potentially sensitive information. In some cases, however, both model coefficients and the functional form of the service providers’ model are not available for clients, “hence, alternative hybrid approaches may be required to make privacy preserving MLaaS practically viable.”
It is believed that such solutions will surface more opportunities and techniques to sustain required system performance while preserving security and privacy properties. Particularly, in areas such as benchmarks, hardware, standardization, etc., the researchers foresee architectural solutions with security mechanisms and advanced cryptographic techniques as viable innovations for building trustworthy AI inference systems.
The paper Trustworthy AI Inference Systems: An Industry Research View is available on arXiv.
Reporter: Fangyu Cai | Editor: Michael Sarazen
This report offers a look at how China has leveraged artificial intelligence technologies in the battle against COVID-19. It is also available on Amazon Kindle. Along with this report, we also introduced a database covering additional 1428 artificial intelligence solutions from 12 pandemic scenarios.
Click here to find more reports from us.
We know you don’t want to miss any story. Subscribe to our popular Synced Global AI Weekly to get weekly AI updates.