La «caja negra» de la IA científica no es rival para un método de 200 años
Las transformadas de Fourier revelan cómo una red neuronal profunda aprende física compleja.
Una de las herramientas más antiguas de la física computacional, una técnica matemática de 200 años conocida como análisis de Fourier — podría revelar información importante sobre cómo una forma de inteligencia artificial llamada red neuronal profunda aprende a realizar tareas que involucran física compleja, como modelar el clima y la turbulencia, según un nuevo estudio.
El descubrimiento, realizado por investigadores de ingeniería mecánica de la Universidad de Rice, se describe en un estudio de acceso abierto publicado en la revista Nexo PNASque es una publicación hermana de procedimientos de la Academia Nacional de Ciencias.
«Este es el primer marco riguroso para explicar y guiar el uso de redes neuronales profundas para sistemas dinámicos complejos como el clima», dijo el autor correspondiente del estudio, Pedram Hassanzadeh. «Podría acelerar en gran medida el uso del aprendizaje científico profundo en la ciencia del clima y conducir a predicciones más confiables del cambio climático».
En el artículo, Hassanzadeh, Adam Sobel y Ashish Chattopadhyay, ambos exalumnos, y Yivi Gauan, becaria de investigación postdoctoral, detallan su uso del análisis de Fourier para estudiar una red neuronal de aprendizaje profundo que ha sido entrenada para reconocer flujos de aire complejos en el atmósfera. o agua en el océano y predecir cómo estos flujos cambiarán con el tiempo. Su análisis, dijo Hassanzadeh, reveló «no solo lo que aprendió la red neuronal, sino que también nos permitió relacionar lo que aprendió la red directamente con la física del sistema complejo que estaba modelando».
Redes neuronales profundas Notoriamente difícil de entender A menudo se consideran «cajas negras». «Esta es una de las principales preocupaciones cuando se usan redes neuronales profundas en aplicaciones científicas. La otra es la generalización: estas redes no pueden operar con un sistema que no sea para el que fueron entrenadas».
El marco analítico que su equipo presenta en el documento, dijo Hassanzadeh, «abre la caja negra, permitiéndonos mirar dentro para comprender qué aprendieron las redes y por qué, y también permitiéndonos relacionar eso con la física del sistema que se aprendió». .»
Sobel, el autor principal del estudio, comenzó la investigación como estudiante de pregrado en Rice y ahora es estudiante de posgrado en la Universidad de Michigan.[{» attribute=»»>New York University. He said the framework could be used in combination with techniques for transfer learning to “enable generalization and ultimately increase the trustworthiness of scientific deep learning.”
While many prior studies had attempted to reveal how deep learning networks learn to make predictions, Hassanzadeh said he, Subel, Guan and Chattopadhyay chose to approach the problem from a different perspective.
“The common machine learning tools for understanding neural networks have not shown much success for natural and engineering system applications, at least such that the findings could be connected to the physics,” Hassanzadeh said. “Our thought was, ‘Let’s do something different. Let’s use a tool that’s common for studying physics and apply it to the study of a neural network that has learned to do physics.”
He said Fourier analysis, which was first proposed in the 1820s, is a favorite technique of physicists and mathematicians for identifying frequency patterns in space and time.
“People who do physics almost always look at data in the Fourier space,” he said. “It makes physics and math easier.”
For example, if someone had a minute-by-minute record of outdoor temperature readings for a one-year period, the information would be a string of 525,600 numbers, a type of data set physicists call a time series. To analyze the time series in Fourier space, a researcher would use trigonometry to transform each number in the series, creating another set of 525,600 numbers that would contain information from the original set but look quite different.
“Instead of seeing temperature at every minute, you would see just a few spikes,” Subel said. “One would be the cosine of 24 hours, which would be the day and night cycle of highs and lows. That signal was there all along in the time series, but Fourier analysis allows you to easily see those types of signals in both time and space.”
Based on this method, scientists have developed other tools for time-frequency analysis. For example, low-pass transformations filter out background noise, and high-pass filters do the inverse, allowing one to focus on the background.
Hassanzadeh’s team first performed the Fourier transformation on the equation of its fully trained deep-learning model. Each of the model’s approximately 1 million parameters act like multipliers, applying more or less weight to specific operations in the equation during model calculations. In an untrained model, parameters have random values. These are adjusted and honed during training as the algorithm gradually learns to arrive at predictions that are closer and closer to the known outcomes in training cases. Structurally, the model parameters are grouped in some 40,000 five-by-five matrices, or kernels.
“When we took the Fourier transform of the equation, that told us we should look at the Fourier transform of these matrices,” Hassanzadeh said. “We didn’t know that. Nobody has done this part ever before, looked at the Fourier transforms of these matrices and tried to connect them to the physics.
“And when we did that, it popped out that what the neural network is learning is a combination of low-pass filters, high-pass filters and Gabor filters,” he said.
“The beautiful thing about this is, the neural network is not doing any magic,” Hassanzadeh said. “It’s not doing anything crazy. It’s actually doing what a physicist or mathematician might have tried to do. Of course, without the power of neural nets, we did not know how to correctly combine these filters. But when we talk to physicists about this work, they love it. Because they are, like, ‘Oh! I know what these things are. This is what the neural network has learned. I see.’”
Subel said the findings have important implications for scientific deep learning, and even suggest that some things scientists have learned from studying machine learning in other contexts, like classification of static images, may not apply to scientific machine learning.
“We found that some of the knowledge and conclusions in the machine learning literature that were obtained from work on commercial and medical applications, for example, do not apply to many critical applications in science and engineering, such as climate change modeling,” Subel said. “This, on its own, is a major implication.”
Reference: “Explaining the physics of transfer learning in data-driven turbulence modeling” by Adam Subel, Yifei Guan, Ashesh Chattopadhyay and Pedram Hassanzadeh, 23 January 2023, PNAS Nexus.
DOI: 10.1093/pnasnexus/pgad015
Chattopadhyay received his Ph.D. in 2022 and is now a research scientist at the Palo Alto Research Center.
The research was supported by the Office of Naval Research (N00014- 20-1-2722), the National Science Foundation (2005123, 1748958) and the Schmidt Futures program. Computational resources were provided by the National Science Foundation (170020) and the National Center for Atmospheric Research (URIC0004).