Table of Contents |
---|
Description
Determine the energy needed to dissociate CO2 by investigating the relevant parameters and determining their working ranges: pressure, temperature, wavelength, molecular resonance state and include quantum chemical energy calculations.
Page Properties | ||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
| ||||||||||||
|
Background
Current Google Sheet cataloguing relevant parameters that are correlated with dissociation or near dissociation of CO2.
Google drive sheets | ||||||
---|---|---|---|---|---|---|
|
Holomap program
The code for meta1 will be name holomap, and released in the holomap repository: https://github.com/hsbay/holomap This code will be similar or a direct fork of the arxiv-sanity-preserver: https://github.com/karpathy/arxiv-sanity-preserver. Holomap should have the additional features of
- working off an existing paper repository separate from arxiv
- working with image based pdfs ( to be able to read older papers.)
- extracting sediment to better add context to parameters, features and other data
- extracting new weighted relevance to paper and search query
Unsupervised Learning implementation of Im2latex
The existing implementation of Im2latex is a supervised model, where in the data is trained from known mostly mathematical formulas and algorithms. This project is to create an unsupervised model similar to the way sanity-preserver functions, such that the code would train on unsorted unlabeled training data, opposed to images of known algorithms and formulas.
Workflow
JPEG corpus
terms (formulae, etc)
surrounding context of found formulae
sediment
relevancy (is the author listing formulae that supports the premise or contrary to the premise)
meaning
comparison with existing terms & data
data update
Next Steps
In process of working with arxiv-sanity-preserver, and Im2latex opensource implementation, and test existing carbon paper cache. Shannon A. Fiume now has a paper cache of both Arxiv and paywall CO dissociation papers in pdfs. The combined paper cache is under 500 papers. She'll be translating them to an image format for im2latex for basic processing. An additional layer or network may need to be added to extract the relevancy of each equation and datum per paper.
Get jpegs read by Mathpix github api and collect the latex output.
Updates
6/15 - Change design spec to create unsupervised learning implementation of Im2latex, combine that with sanity-preserver.
6/8 - Papers translated to jpegs by pdf2jpeg workflow automatorThe current focus is to manually conduct an indepth field review and correlate the dissociation parameters, and conditions to assembly a clustering of all similar parameters and conditions to all known routes to solid carbon.