DL Greens Functions

I plan to use CNN's to speed up chemistry calculations.

Comment

Inspiration

Previous papers.

What it does

Learns the energy of a system.

How we built it

Using python.

Challenges we ran into

Conversion to images.

Accomplishments that we're proud of

Producing a working model.

What we learned

How to process data for a task that the data is not designed for.

What's next for DL Greens Functions

Refinement.

Built With

Share this project:

Updates

posted an update —

Introduction: I am trying to map Green's function matrices to energies. I arrived at this topic because if this relationship can be represented by a DL model it will accelerate quantum calculations. This is a regression task. Challenges: Finding a reasonable method to convert my data into images was difficult. Insights: I have trained the model and can show it's unoptimized performance. I expected my model to perform better which may indicate that I need more data. Plan: I am on track. I need to dedicate more time to data creation. I am thinking of changing the system I am studying.

Log in or sign up for Devpost to join the conversation.

posted an update —

Title- Deep Learning Energy Estimates Using Green's Function

Who- Kenneth Berard, Kberard

Introduction- I am trying to map Green's function matrices to energies. I arrived at this topic because if this relationship can be represented by a DL model it will accelerate quantum calculations. This is a regression task.

  1. Broecker, P., Carrasquilla, J., Melko, R.G. et al. Machine learning quantum phases of matter beyond the fermion sign problem. Sci Rep 7, 8823 (2017). https://doi.org/10.1038/s41598-017-09098-0 #Data- The data I will use can be generated in any quantity I need by running a simulation. I will process the data similar to how reference 1 describes. #Methodology- My model is a convolution neural network. I provide the model with a set of images and labels, then I perform a regression task. One challenge will be finding a way to represent the data which contains all required physical information. A backup plan will be to use a feed forward neural network to learn the original matrix. #Metrics- A success would be a low MSE error. I plan to compare different systems and DL architectures. The models performance will be assessed using a testing data set. My base goals are to implement a DL framework that has a reasonable loss. My target is to export a model that makes accurate predictions on any Green's function provided. My stretch is to replace the current method with my method. #Ethics- Deep learning is a good approach to this problem because the amount of data involved is large and the relationship is complex. My data set is from running a simulation. There are no concerns with how it was collected. It has no social implications as this is the computation of a complex numbered quantum matrix.

Log in or sign up for Devpost to join the conversation.

posted an update —

Introduction: I plan to convert Green's Functions to images. I will then train a CNN on these images to learn a mapping to the energy. Challenges: Finding a reasonable method to convert my data into images was difficult. Insights: I have trained the model and can show it's unoptimized performance. I expected my model to perform better which may indicate that I need more data. Plan: I am on track. I need to dedicate more time to data creation. I am thinking of changing the system I am studying.

Log in or sign up for Devpost to join the conversation.