A team of scientists from New York University has developed a neural network that can explain how it makes its predictions. This work sheds light on the principles of the functioning of neural networks, which are the basis of artificial intelligence and machine learning.
The main direction of the study is associated with the specific use of neural networks actively used in recent years – the solution of complex biological problems. The basis was the study of the processing process of RNA, which plays a key role in transmitting information from DNA to functional RNA and protein products.
“Many neural networks remain black boxes because they cannot explain their work, which causes fears regarding their reliability,” says Odda Regev, a professor of computer science from the Institute of Mathematical Sciences of Kurono of New York University. He adds that thanks to the new method that improves the volume and quality of data for teaching machines, an interpreted neural network has been developed, which can accurately predict complex outcomes and explain how it comes to its predictions.
To create their own regions and his colleagues, they relied on existing data on RNA splashing. Their model, in some way, the equivalent of a powerful microscope, allows scientists to track and quantify the processing process of RNA.
Reghev emphasizes: “Based on the approach of ‘developed for interpretation, we created a neural network model that provides an understanding of RNA splashing.” Researchers revealed that a small structure in RNA resembling a hair hair clip can reduce splashing.
These conclusions were confirmed by a series of experiments: when the RNA molecule took the form of the hairpin, the processing process stopped, and vice versa, if this structure was violated, the placing was resumed.