Saturday, November 21, 2026
Translate
PrivacySubmit newsShare your story Newsletter Advertise
Breaking
AI news

MIT Method Stops Physics AI Models From Spiraling Out of Control

A new MIT machine learning approach prevents chaotic system models from drifting into physically impossible predictions over time.

MIT Method Stops Physics AI Models From Spiraling Out of Control

Researchers at MIT have developed a new machine learning approach that prevents AI models of chaotic systems from generating impossible, unbounded predictions over time. Published in the Proceedings of the National Academy of Sciences, the technique introduces rigorous physical constraints to improve long-term forecasting.

Hard-Constrained Neural Networks

Modeling complex processes like weather, climate patterns, and turbulent flows often causes machine learning models to drift into physically impossible predictions. Small errors can accumulate over time, leading simulations to produce values that grow without limit.

To counter this, researchers at MIT’s Laboratory for Information and Decision Systems developed a technique that embeds physical constraints directly into the models. Instead of relying on soft incentives during training, the team formulated energy-based constraints using a projection layer.

The TechUpscale Brief. The day's cyber, AI and tech news in one short email, every weekday morning. Free. Unsubscribe anytime.

Enforcing Energy Dissipation

The approach allows the model to learn an energy-like function along with a confined boundary for chaotic behavior. Outside this boundary, the energy-like quantity decreases, drawing the simulated system back toward the bounded region.

According to the researchers, this method establishes formal theoretical guarantees that predictions remain bounded over arbitrarily long forecast windows. The modular projection layer is model-agnostic, and the team has made their open-source code publicly available online.

Key facts and where they come from
  • MIT researchers developed a new approach for machine learning models of chaotic systems.
    researchers with MIT’s Laboratory for Information and Decision Systems (LIDS) have found a way to introduce physics-based restrictions on machine learning models
  • The approach uses a projection layer to enforce energy dissipation as a hard constraint.
    We then add a projection layer, a component that adjusts how the model predicts the system will evolve, to enforce energy dissipation as a hard constraint.
  • The findings were published in the Proceedings of the National Academy of Sciences.
    The new findings are published in the Proceedings of the National Academy of Sciences (PNAS)

The catch

  • The announcement does not say how the approach performs computationally compared to standard unconstrained neural network architectures.
  • The team has not yet tested the method on full-scale realistic weather systems, though they plan to do so.

The open question: Can this projection layer approach scale effectively to extremely high-dimensional real-world weather forecasting systems?

Tell us what you think · Discuss it in the AI room

Reader poll: Will physics-constrained architectures become standard for all AI-based climate and weather forecasting models?

A show of hands from readers, not a scientific poll. Say why in the comments.

Add TechUpscale as a preferred source on Google

The TechUpscale Brief

The day's cyber, AI and tech news in one short email, every weekday morning. Free. Unsubscribe anytime.

I'm most interested in

Can this projection layer approach scale effectively to extremely high-dimensional real-world weather forecasting systems?

Comments are moderated. Be civil and stay on the story: no personal attacks, hate, spam or ads, and no one else's private details. Your email is never shown. Privacy policy.

More AI