Artificial intelligence is already revolutionizing scientific and cosmological exploration. In a recent research breakthrough, the scientists believe that AI could make the search for new Physics much faster and less expensive.
Here is the surprising downside too. The discovery of new Physics through the transfer learning technique is not without risks.
In the study published in the Journal of Cosmology and Astroparticle Physics, the researchers explored the utility of transfer learning in investigating cosmological theories beyond the standard model.
The standard cosmological model (ΛCDM) explains many large-scale features of the universe; scientists believe it is incomplete.
To explore mysterious phenomena including modified gravity, massive neutrinos and dark energy particles, there is a great need to explore new physics that requires generating massive numbers of virtual universes via computer simulation. As per scientists, producing these complex simulations is not everyone’s cup of tea. It would require high computational power.
The potential solution lies in “transfer learning.” Researchers applied transfer learning to make AI training more efficient. Instead of training the neural network directly on highly complex models, the AI is first “pre-trained” on simpler, less expensive ΛCDM simulations.
Adrian Bayer, a cosmologist at the Flatiron Institute and Princeton University, co-author of the study, said, “It’s basically a shortcut. Usually people train the AI directly on the most computationally expensive simulations. What we do instead is first use simpler and less expensive ΛCDM simulations to give the AI an idea of what’s happening, and only afterward move to the more complex models.”
According to first author Veena Krishnaraj,, this strategy will prevent the AI from having to “digest everything at once,” thereby reducing the number of expensive simulations required by more than a factor of ten.
“The negative transfer is not random. It is driven by underlying physical degeneracies in the model.So this is something we need to be aware of and try to mitigate,” says Krishnaraj.
The researchers have only tested this technique simulation. If proven successful at large scale, it could revolutionize future cosmological research.