How much can self-supervised models learn about semiconductor data without being told what to look for?
Self-supervised learning is responsible for many recent breakthroughs in machine learning. Nearly every major development in NLP relies on self-supervised pretraining to some extent, and in vision, techniques like contrastive learning are gaining popularity as well. On ImageNet style data (and even non-curated datasets of natural images), self-supervised learning has been shown to be a powerful tool for learning useful representations of images without the need for large amounts of labeled data.
It is often unclear, however, how effectively these techniques can be applied to images in other domains. How well can self-supervised learning be used in semiconductor manufacturing, where data is large, unlabeled, and highly imbalanced? This work provides thorough evaluations of some of the most popular self-supervised learning techniques on semiconductor wafer map data. Everything is done using lightly, an excellent package for self-supervised learning on images using PyTorch. More than a dozen joint embedding and masked image modeling frameworks are put to the test on the WM-811K and MixedWM38 datasets.
In order to set up the necessary environment:
- review and uncomment what you need in
environment.yml
and create an environmentssl_wafermap
with the help of conda:conda env create -f environment.yml
- activate the new environment with:
conda activate ssl_wafermap
NOTE: The conda environment will have ssl_wafermap installed in editable mode. Some changes, e.g. in
setup.cfg
, might require you to runpip install -e .
again.
Optional and needed only once after git clone
:
-
install several pre-commit git hooks with:
pre-commit install # You might also want to run `pre-commit autoupdate`
and checkout the configuration under
.pre-commit-config.yaml
. The-n, --no-verify
flag ofgit commit
can be used to deactivate pre-commit hooks temporarily. -
install nbstripout git hooks to remove the output cells of committed notebooks with:
nbstripout --install --attributes notebooks/.gitattributes
This is useful to avoid large diffs due to plots in your notebooks. A simple
nbstripout --uninstall
will revert these changes.
Then take a look into the scripts
and notebooks
folders.
- Always keep your abstract (unpinned) dependencies updated in
environment.yml
and eventually insetup.cfg
if you want to ship and install your package viapip
later on. - Create concrete dependencies as
environment.lock.yml
for the exact reproduction of your environment with:For multi-OS development, consider usingconda env export -n ssl_wafermap -f environment.lock.yml
--no-builds
during the export. - Update your current environment with respect to a new
environment.lock.yml
using:conda env update -f environment.lock.yml --prune
├── AUTHORS.md <- List of developers and maintainers.
├── CHANGELOG.md <- Changelog to keep track of new features and fixes.
├── CONTRIBUTING.md <- Guidelines for contributing to this project.
├── Dockerfile <- Build a docker container with `docker build .`.
├── LICENSE.txt <- License as chosen on the command-line.
├── README.md <- The top-level README for developers.
├── configs <- Directory for configurations of model & application.
├── data
│ ├── external <- Data from third party sources.
│ ├── interim <- Intermediate data that has been transformed.
│ ├── processed <- The final, canonical data sets for modeling.
│ └── raw <- The original, immutable data dump.
├── docs <- Directory for Sphinx documentation in rst or md.
├── environment.yml <- The conda environment file for reproducibility.
├── models <- Trained and serialized models, model predictions,
│ or model summaries.
├── notebooks <- Jupyter notebooks. Naming convention is a number (for
│ ordering), the creator's initials and a description,
│ e.g. `1.0-fw-initial-data-exploration`.
├── pyproject.toml <- Build configuration. Don't change! Use `pip install -e .`
│ to install for development or to build `tox -e build`.
├── references <- Data dictionaries, manuals, and all other materials.
├── reports <- Generated analysis as HTML, PDF, LaTeX, etc.
│ └── figures <- Generated plots and figures for reports.
├── scripts <- Analysis and production scripts which import the
│ actual PYTHON_PKG, e.g. train_model.
├── setup.cfg <- Declarative configuration of your project.
├── setup.py <- [DEPRECATED] Use `python setup.py develop` to install for
│ development or `python setup.py bdist_wheel` to build.
├── src
│ └── ssl_wafermap <- Actual Python package where the main functionality goes.
├── tests <- Unit tests which can be run with `pytest`.
├── .coveragerc <- Configuration for coverage reports of unit tests.
├── .isort.cfg <- Configuration for git hook that sorts imports.
└── .pre-commit-config.yaml <- Configuration of pre-commit git hooks.
This project has been set up using PyScaffold 4.1.4 and the dsproject extension 0.7.2.