National Edge AI Hub Webinar Series: Towards Reliable and Efficient Edge AI: Compression and Certified Verification of Neural Representations

As artificial intelligence is increasingly deployed in resource-constrained environments, ensuring systems are both efficient and trustworthy is becoming a critical research challenge. In this webinar, Dr Mohammad Kazemi (Imperial College London) will explore recent advances in the compression and certified verification of neural representations, highlighting how these techniques can support the development of reliable, resource-efficient Edge AI systems.

Title:

Towards Reliable and Efficient Edge AI: Compression and Certified Verification of Neural Representations

Speaker: Dr Mohammad Kazemi, Imperial College London
Date: November 24th
Time: 2 PM GMT

Please register below to get invited to the webinar via Microsoft Teams. 

Abstract:

Deploying artificial intelligence in resource-constrained environments requires addressing challenges beyond predictive accuracy, particularly computational efficiency, memory requirements, and reliability. Implicit neural representations (INRs), which encode signals and continuous data through neural networks, offer a flexible and powerful approach to representing complex information. However, making these representations compact and ensuring their reliable behaviour remain important challenges.

In this webinar, I will discuss recent research on the compression and certified verification of implicit neural representations, focusing on our CertINR framework. I will introduce the underlying mathematical ideas, explain how compression and formal verification can be addressed, and present the associated theoretical and experimental findings. The talk will also explore the relevance of these techniques to trustworthy and resource-efficient Edge AI, highlighting potential applications and open research directions.

Bio: Dr. Mohammad Kazemi is a Research Associate in Edge AI at the Department of Computing, Imperial College London, working with Professor Julie McCann in the Adaptive Emergent Systems Engineering (AESE) group. Previously a Marie Skłodowska-Curie Fellow at Imperial, his research spans machine learning, information theory, and wireless communications, with an emphasis on developing efficient and reliable intelligent systems. His current work includes neural representation learning, formal verification, and task-oriented communication for distributed sensing applications.