Recently, I want to write something about data in physics

Click For Summary

Discussion Overview

The discussion revolves around the intersection of data and physics, particularly in the context of machine learning and its implications for modeling and understanding physical phenomena. Participants explore the role of data in physics, the potential for machine learning to replace traditional mathematical models, and the challenges associated with explaining results derived from data-driven approaches.

Discussion Character

  • Exploratory
  • Debate/contested
  • Technical explanation

Main Points Raised

  • Some participants suggest that the rise of machine learning in computer science prompts questions about the role of data in physics and how physicists might parallel computer scientists in their empirical approaches.
  • There is a proposal to investigate whether traditional mathematical formulas in physics could eventually be replaced by neural networks that predict experimental outcomes.
  • One participant expresses concern about the limitations of machine learning in science, emphasizing the need for explanations behind discoveries, which may not be achievable through data alone.
  • A specific example is provided regarding a program that extracted equations of motion from data without prior physics knowledge, which led to discoveries in cellular behavior but faced publication challenges due to a lack of theoretical explanation.
  • Another participant notes that the program mentioned has found practical applications in agriculture, indicating a successful outcome despite the initial theoretical concerns.

Areas of Agreement / Disagreement

Participants express differing views on the implications of using machine learning in physics, with some highlighting its potential and others raising concerns about the inability to explain results. The discussion remains unresolved regarding the future role of machine learning in replacing traditional physics models.

Contextual Notes

Limitations include the dependence on definitions of data and models in physics, as well as unresolved questions about the theoretical underpinnings necessary for scientific publication.

Pring
Messages
47
Reaction score
0
Nowadays, the machine learning of computer science is hot. It is based on data, and drove by data. Thus, a question is naturally coming out: the data in physics, and the models of data. I think it is a really empirical way to know how physicists do the same thing as the computer scientists. So could you give some examples about the data in physics or give the links?
 
Last edited:
Physics news on Phys.org
Good. Is thee anything we can help you with? A question, perhaps ?

Did you notice your post got an 'advanced' tag, meaning PhD level ?
 
Pring said:
Nowadays, the machine learning of computer science is hot. It is based on data, and drove by data. Thus, a question is naturally coming out: the data in physics, and the models of data. I think it is a really empirical way to know how physicists do the same thing as the computer scientists.
Are you investigating the question, whether the current physics (mathematical formulas) will be once replaced by neutral networks, which predict the outcome of experiments?
 
A.T. said:
Are you investigating the question, whether the current physics (mathematical formulas) will be once replaced by neutral networks, which predict the outcome of experiments?
That's a horror idea! I just think of the key idea 'from data to model' in physics instead of investigating it.
 
There is a fundamental weakness in using machine learning methods in science. Once we discover something we need to be able to explain it but that might not be possible.

Cornell has a program that extracted the equations of motion for a compound mechanical pendulum using lots of data with no knowledge of physics. It worked so they tried it on some cellular data and discovered some behavior in cells never before quantified as an equation. However the biologists said they couldn't publish because they didn't have a theory to explain the result.

https://www.wired.com/2009/04/Newtonai/
 
  • Like
Likes   Reactions: bhobba
jedishrfu said:
https://www.wired.com/2009/04/Newtonai/

If you follow that link deep enough, you discover it spawned a company, which was then bought by an agricultural seed company, which uses it to evaluate hybrid seeds to market to farmers! You may even be eating the results in your next meal.

So it appears there was a use for the program after-all!
 

Similar threads

Undergrad Wired article: Physicists will soon rule Silicon Valley
  • · Replies 7 ·
Replies
7
Views
3K
Undergrad Physicists set to revolutionize big data, AI
  • · Replies 16 ·
Replies
16
Views
2K
Python Why backpropagation dominates neural networks instead of interpolation
  • · Replies 4 ·
Replies
4
Views
3K
Physics Looking to transition from data science to computational physics
  • · Replies 3 ·
Replies
3
Views
5K
Can I manage to do both: Computer Science and Research in Physics?
  • · Replies 2 ·
Replies
2
Views
2K
Need some advice on my future career
  • · Replies 1 ·
Replies
1
Views
479
Which MSc to do after BSc physics if you love to code?
  • · Replies 10 ·
Replies
10
Views
3K
Admissions Computational physics career change
  • · Replies 2 ·
Replies
2
Views
2K
Graduate Physics - Description is description
  • · Replies 22 ·
Replies
22
Views
2K
Programs Diploma thesis - Neural Network Application in Physical Problems
  • · Replies 1 ·
Replies
1
Views
2K