kural_work

Machine learning work in PyTorch for the lab of Dr. Comert Kural (https://www.asc.ohio-state.edu/kural.1/).

Work in progress. Some tasks show performance, but none scaled into production yet.

Movie RNN
- Simplifying images
  - Sand-shaking algorithm
    - Perhaps implementing a signal-boosting algorithm (more below) would be preferrable
  - Cell outliner (NN)
    - Made use of K-means here
  - Crop cell to fixed-size image
  - Apply autoencoder (NN) to reduce dimentionality
- Continuous-time RNN
  - Currently not enough training data taken into account.
  - Model not stable.
    - Predictions diverge or converge too quickly to get useful information.
- Predict just one frame (NN)
  - Currently not enough training data taken into account.
  - Model is able to over-train and achieve desireable results, but unseen data produces unacceptable results.
Signal-boosting with U-nets
- Very easy to boost gross structure signal and flatten-out noise
- Some fine detail is lost, but I believe building a better classifier (NN) and using perceptual loss can help this.

NN : A neural network was used to accomplish this task. All NN architectures can be found in kural_core/models.py

Name		Name	Last commit message	Last commit date
Latest commit History 16 Commits
.vscode		.vscode
Obsolete Stuff		Obsolete Stuff
kural_core		kural_core
.gitignore		.gitignore
Autoencoder_with_noise.ipynb		Autoencoder_with_noise.ipynb
Bounding_box.ipynb		Bounding_box.ipynb
Classification after mitosis.ipynb		Classification after mitosis.ipynb
Convolutional Autoencoder.ipynb		Convolutional Autoencoder.ipynb
Exploring_lots_of_files.ipynb		Exploring_lots_of_files.ipynb
Modem_data_web_scraping.ipynb		Modem_data_web_scraping.ipynb
README.md		README.md
RNN_from_convolutional_autoencoder.ipynb		RNN_from_convolutional_autoencoder.ipynb
actin_detector.ipynb		actin_detector.ipynb
actin_snr_boost.ipynb		actin_snr_boost.ipynb
boilerplate.ipynb		boilerplate.ipynb
boundary_detection.ipynb		boundary_detection.ipynb
crops_and_points_for_each_movie.ipynb		crops_and_points_for_each_movie.ipynb
file_access.ipynb		file_access.ipynb
long_time_prediction.ipynb		long_time_prediction.ipynb
train_and_val_loss.pickle		train_and_val_loss.pickle
unsupervised_cell_detection.ipynb		unsupervised_cell_detection.ipynb

Provide feedback