Improved learning in human evolutionary systems with dynamic contrastive learning

What is it about?

We assume data will shift when making predictions involving human interactions. We use contrastive learning as a pre-training mechanism to embed the incoming data according to the time period that it occurred. The resulting embedding dimension contrasts the data according to time period.

Featured Image

Photo by Darius Bashar on Unsplash

Photo by Darius Bashar on Unsplash

Why is it important?

If you assume your data will be constant, or slightly drift, you will overfit. This is especially true for data involving human interactions. That is because humans generally interact in mixed-motive games, where evolutionary pressures require constant shifting of strategies.