AI in the Beauty Industry with Emmanuel Acheampong, roboMUA Co-Founder

Model Drift

Model drift is a significant issue in AI, particularly in computer vision. Emmanuel Acheampong explains that as more people use the model, it degrades faster, necessitating frequent refreshes to maintain accuracy. He emphasizes the importance of continuously training and refreshing the model APIs in production to combat this drift 1.

The more people use the model, the quicker the model degrades. Refreshing the model multiple times is something we learned as part of the process.

--- Emmanuel Acheampong

Legal implications and government regulations also play a role in managing model drift, with various governments working to create laws around data usage 1.

   

Synthetic Data

Creating diverse datasets is crucial for AI training, especially when existing data is scarce. Emmanuel Acheampong shares how they initially used celebrity images to build a dataset for different skin shades, later incorporating synthetic data to expand it 2.

We obviously have to use synthetic data. Right? So 100 skin shades. Like, if you google right now, hey, dark skin with, like, red undertones, the data would be very few.

--- Emmanuel Acheampong

He predicts an inflection point where generative AI data will exceed traditional internet data, raising questions about the future of AI training 3.