Can you trust synthetic data?

• Learn more about the technology behind synthetic data generation → ibm.biz/learn_more_watsonx-ai
• Learn more about how generative AI play a role in synthetic data generation → ibm.biz/AI-and-synthetic-data
Synthetic data can be created with generative AI to help shape decision making. Learn about why synthetic data exists and how data quality, data privacy, and deployment set up and are pivotal to being able to trust it.
• AI news moves fast. Sign up for a monthly newsletter for AI updates from IBM → ibm.biz/AI-newsletter
• Get started for free on IBM Cloud → ibm.biz/buildonibmcloud
#syntheticdata #generativeAI

[@BooleanDisorder]

I was thinking that these hosts are really good at writing backwards, but then I realized the video is probably mirrored. 😂

[@jessetwumboafo]

It’s not mirrored, they are using a lightboard

[@michaelcharlesthearchangel]

AI can be used to reverse a mirror effect.

[@marzuno]

They're using lightboard: czcams.com/video/Uoz_osFtw68/video.html

[@lekanfagbuyi7963]

Working on a project using synthetic data meaning this video came at the right time. 👌🏾

[@sk3ffingtonai]

Since AI is "all about the data", more and more videos about data integrity and privacy best practices would be well appreciated. Thank you.

[@yvanbarbeau2999]

Fantastic job ! Hats off !

[@ranggayogiswara5148]

Can you explain that new york, texas thing to non-American? I don't understand.

[@LeeYeon-qv1tz]

When creating Synthetic data, we make sure that city, state, and country are not shuffled. They cannot be shuffled. For a customer, we cannot have new york in a column and alaska in another column since they are geographically far from each other.

[@SiegelBantuBear]

🙏🏼

[@Alice8000]

How do you guys write backwards like that??

[@MrRaja]

Practice

[@shehab9487]

but how data is synthesized from the original real data ?

[@MrRaja]

By looking at and using the private data 🤣

[@RomansProkopjevs]

One approach is to initialise a neural network with random junk as parameters and adjust these parameters to fit well the real-world data but not overfit. The output will be similar to the real-world data but not the same.
Another approach is to use classical statistical methods and extend them with an overfitted neural network. For example, an ARIMA model can generate artificial residuals, which the neural network can augment.

[@Alice8000]

Does your GF trust you?