As a researcher working at the intersection of mental health, finance, and machine learning, I was thrilled to attend the Eyes-Off Data Summit 2024. This event delved into cutting-edge topics like data security, privacy-enhancing technologies (PETs) such as differential privacy, privacy-preserving machine learning (ML), Fully Homomorphic Encryption (FHE) and more. It featured a fascinating mix of panel discussions from industry leaders in privacy, AI, and AI regulation, alongside live demos showcasing real-world applications of PETs, privacy-preserving ML, and other innovations.
Here are some of the key insights I gained from the summit:
- Privacy is not a binary concept—it’s complex and nuanced.
- Responsibility with data is crucial, even when working for societal good.
- Direct access to sensitive data across the board increases exposure risks and simply doesn’t work.
- As models grow larger, they tend to memorise data from training, leading to potential information leaks.
- The biggest vulnerability in security is often the users themselves.
- Classical models may seem boring, but they work—and can be made privacy-compliant. We shouldn’t rush into shiny, new models without reason.
I also learnt about differential privacy and the barriers to adopting it, such as:
- Painkillers versus vitamins analogy i.e. is it a regulatory pain that needs to be solved or just a cool feature?
- Access limitations on platforms.
- Explanation and trust issues around adding noise to data.
- A general lack of awareness regarding differential privacy, its benefits and available solutions.
Overall, it was an eye-opening event, and I’m eager to apply these insights and explore privacy techniques in my research!
A huge thank you to Oblivious for organising such a captivating and unforgettable event.
#EODSummit2024 #EyesOffMyData #DataPrivacy #PrivacyTech
