Lamini’s Post

"But what we do is "data transport acceleration" and through many awkward conversations I have learned that there is a big difference between "data migration" and "data transport." - Seth Noble, CEO, Data Expedition, Inc. Read what he means in our latest blog: https://bit.ly/3VeYDPK

𝐆𝐫𝐚𝐝𝐢𝐞𝐧𝐭 𝐁𝐚𝐬𝐞𝐝 𝐂𝐨𝐦𝐦𝐮𝐧𝐢𝐜𝐚𝐭𝐢𝐨𝐧 𝐢𝐧 𝐅𝐞𝐝𝐞𝐫𝐚𝐭𝐞𝐝 𝐋𝐞𝐚𝐫𝐧𝐢𝐧𝐠 𝐢𝐬 𝐬𝐭𝐢𝐥𝐥 𝐯𝐮𝐥𝐧𝐞𝐫𝐚𝐛𝐥𝐞 : Even when ensuring that only gradients are communicated, there are ways to reconstruct the labels and data to some extent. The methodology involves comparing successive intentional gradients given to the global server and their gradient values, which reveal the difference in model weights at the server. This yields information that can be used to reconstruct the data and labels. 𝐏𝐨𝐭𝐞𝐧𝐭𝐢𝐚𝐥 𝐒𝐨𝐥𝐮𝐭𝐢𝐨𝐧: 1) Avoid using "cross-entropy" and its variants, as they are among the most vulnerable activation functions to deep gradient leakage. 2) Differential Privacy or other Random Masking Techniques can reduce the extent to which gradients can reveal information. 𝐃𝐞𝐞𝐩 𝐆𝐫𝐚𝐝𝐢𝐞𝐧𝐭 𝐋𝐞𝐚𝐤𝐚𝐠𝐞: https://lnkd.in/gTsKvnv2 𝐈𝐦𝐩𝐫𝐨𝐯𝐞𝐝 𝐃𝐞𝐞𝐩 𝐆𝐫𝐚𝐝𝐢𝐞𝐧𝐭 𝐋𝐞𝐚𝐤𝐚𝐠𝐞: https://lnkd.in/g-vcSyfY

Good news for data governance professions! 🎊 Snowflake just added more data governance features to the platform in the form of aggregation and projection policies (now generally available). 🎉 Aggregation Policies protect the privacy of individual rows by requiring analysts to run queries that aggregate data rather than retrieving individual rows and projection policy defines which users should be blocked from projecting columns and which users should be allowed. Check out the release notes here - https://lnkd.in/giCfnYTT

Ready for a demo? Learn about privacy-enhancing technologies and their application in real-world business use cases to help you appropriately balance the trade-off between data privacy and utility.

Imagine if someone put Differential Privacy into the world's ~7th most popular database! Oh wait... we did. Differential Privacy is now GA at Snowflake. Come learn how to utilize Snowflake Horizon's amazing privacy features to unlock the value in your most sensitive data.

Ian Connolly makes this sound so, so easy - and I can tell you that it was not. Every single customer using the AI Data Cloud can unlock insights and new value their most sensitive data. If you have sensitive data that you've previously not felt comfortable with sharing and you'd like to better collaborate with your partners, other companies, and/or coworkers within the same organization then attend this session!

Just finished the course “Understanding and Prioritizing Data Privacy”!

Just finished the course “Understanding and Prioritizing Data Privacy” by Michelle Dennedy! 

Inrupt's Solid technology is highlighted in BBC News today—check out the article to learn how Solid Pods are transforming industries like recruitment, media and healthcare:

Just finished the course “Understanding and Prioritizing Data Privacy” by Michelle Dennedy!