Russian Scientists Claim AI Training Breakthrough

Researchers in Moscow say their new method could speed up drug discovery and make neural networks far more efficient.

Russian scientists at the Higher School of Economics have unveiled a new approach to training generative flow neural networks, according to state news agency TASS.

The researchers argue that their system can dramatically improve how artificial intelligence handles complex, unstructured tasks — particularly in fields like drug development and transportation planning.

The core innovation lies in how the method manages the relationship between two critical components of a neural network: the forward and backward models. Balancing the two has long been a challenge. Now, the researchers say, their system allows the models to interact more flexibly, almost like holding “negotiations” in real time.

That shift could mean faster training cycles, lower computational costs, and broader applications for AI across industries. By reducing the heavy demand on computing power, the approach makes it easier to scale neural networks for real-world problems.

Experts note that the method seems especially promising in situations with high levels of uncertainty, where the backward model plays an outsized role in decision-making. That versatility, researchers argue, could make the system valuable in multiple scientific and industrial domains.

The announcement highlights both the pace of AI research worldwide and Russia’s ambitions to remain a player in the field, despite limited visibility on the global stage.