Algorithmic biases: neural networks are also influenced by hardware
• Researchers have demonstrated that the fairness of AI models is dependent on hardware platforms used for their deployment. Some hardware configurations have been found to introduce demographic biases which are highly problematic, notably for healthcare applications.
• Model compression is proposed as a key solution for deploying neural networks on devices with limited hardware resources, such as AI capable PCs or edge computing devices.
• The adoption of co-design frameworks for hardware and software architectures will play an essential role in the drive to optimise the fairness and performance of AI models. The integration of non-volatile memory (NVM) devices and noise reduction in neuromorphic systems are also promising avenues for future development.
Read the article
• Model compression is proposed as a key solution for deploying neural networks on devices with limited hardware resources, such as AI capable PCs or edge computing devices.
• The adoption of co-design frameworks for hardware and software architectures will play an essential role in the drive to optimise the fairness and performance of AI models. The integration of non-volatile memory (NVM) devices and noise reduction in neuromorphic systems are also promising avenues for future development.
Automated intervention reports for augmented technicians thanks to generative AI
Read the article
Controversies around AI: from ethical questions to legal regulation
Read the article