The AI ​​tool that physicists are ecstatic about is open source! Relying on experimental data to directly discover physical formulas, you can run it on your laptop

An AI tool that makes physicists ecstatic is open source on GitHub!

It is called Φ-SO. It can directly find hidden patterns in the data, and it can provide the corresponding formula directly in one step.

The whole process does not require the use of supercomputing, A laptop can complete Einstein’s work in about 4 hours Mass-energy equation.

This result comes from the University of Strasbourg in Germany and the Data61 Department of the Australian Commonwealth Scientific and Industrial Research Organization. According to the first author of the paper, the research used It took 1.5 years and received widespread attention from the academic community.

# Once the code is open sourced, its stars will rise very quickly.

In addition to physicists directly calling Amazing, there are also researchers from other disciplines who came to discuss whether the same method can be transferred to their field.

##Reinforcement learning physical condition constraints

The technology behind Φ-SO is called "deep symbolic regression" and is implemented using

Recurrent Neural Network (RNN) reinforcement learning.

First input the previous symbol and context information to RNN, predict the probability distribution of the next symbol, and repeat this step to generate a large number of expressions.

At the same time, physical conditions are incorporated into the learning process as prior knowledge to prevent AI from formulating formulas that have no actual meaning, which can greatly reduce the search space.

#Reinforcement learning is introduced to let AI learn to generate the formula that best fits the original data.

Unlike reinforcement learning, which is used to play chess, control robots, etc., in symbolic regression tasks, you only need to care about how to find the best formula, and do not care about the average performance of the neural network.

So the rules of reinforcement learning were designed to only reward the top 5% of candidate formulas, and not penalize the other 95% to encourage the model to fully explore the search space.

The research team used classic formulas such as the analytical expression of damped resonators, Einstein's energy formula, and Newton's universal gravitation formula to conduct experiments.

Φ-SO can 100% restore these formulas from the data, and the above methods are indispensable.

Compared with other methods into MLP, Φ-SO also performs better outside the training range.

#The research team finally stated that although the algorithm itself still has some room for improvement, their primary task has been changed to using new tools to discover unknowns. The laws of physics are gone.

GitHub：​​https://www.php.cn/link/c338d814c14c9d479eb5ec0b99d887f6​​
Paper:##​​https://www.php.cn/link/4738a8f6fab937d899ae9631beab116f​ ​

Reference link: [1]​​​https://www.php.cn/link/5c8cb735a1ce65dac514233cbd5576d6​​

The above is the detailed content of The AI ​​tool that physicists are ecstatic about is open source! Relying on experimental data to directly discover physical formulas, you can run it on your laptop. For more information, please follow other related articles on the PHP Chinese website!