If you see ripples in a pond, you can usually deduce where the pebble that hit the water is. This is an example of an inverse problem in science. These problems are used in various ways, from localizing tumors in medical scans to identifying the locations of earthquakes.

However, working backward from an outcome to find the cause is extremely difficult. The outcome can be noisy with errors in the data that make it almost impossible to find the right answer.

The Wall That Current AI Faces

Artificial intelligence has been used to solve these inverse problems. AI can identify the parameters of these problems that humans might otherwise miss. However, standard forms of AI struggle with the physics of these problems and with data noise.

At Penn Engineering, AI was tested for its ability to solve inverse problems. The test results were not great. The AI was slow, used up a massive amount of memory while solving these problems, and mostly failed to return the correct answer.

The Mollifier Solution

A group of researchers at the University of Pennsylvania has found a way to clear the air in this situation. They have developed “Mollifier Layers,” which use a mathematical trick from the 1940s to smooth out the data that enters the AI.

By filtering out data noise, the AI can focus on understanding the problem’s physics. This small change results in a massive difference in how the AI understands the problem.

Faster, Leaner, and More Accurate AI

As the researchers at Penn Engineering have tested this model of AI with Mollifier Layers, there have been impressive results. This AI model is 10 times faster than current models while using 90% less memory. It is also significantly more likely to find the correct answer. Where current AI finds the right answer 20% of the time, AI with Mollifier Layers finds the answer 100% of the time.

Furthermore, because this model does not require a massive supercomputer to perform its calculations, it has a variety of potential applications. In the weather industry, it could produce much better weather forecasts. In the medical field, it could help scientists understand the role that certain genes play in triggering diseases. In the world of infrastructure, AI can identify cracks in bridges and buildings before they become visible to the naked eye. This innovation in AI allows people to finally begin to find answers to the world’s most complex mysteries.