Nice to meet you, I'm Yuval!
I am a masters student at The Cooper Union working on a increasing the accuracy of flood-map models through increasing the temporal resolution of satellite imagery.
Some of my favorite work:
S.P.P.A.C.Y. (Tensorflow, tf.data, Google Earth Engine- Python API) - We designed and made a machine learning algorithm that takes satellite imagery and predicts crop-yield on a pixel-wise basis. The project had us wrestle with lack of sufficient labels (we could only find aggregate yield values per county), as well as with creating a clean costume dataset. What was really cool, in my opinion, was that we used histograms to help generalize the aggregate, county-level, yield labels to any scale input.
Data Science for Social Good (Pandas, Scikit-learn, Scipy, Group-lasso, Seaborn) - Volunteer work to interpret data and increase costumer retention. We did a lot of data cleaning & feature selection, collaborated with key personnel to define metrics to evaluate program performance, and trained models to predict most useful cohorts to study further (using residual analysis). At the end of our collaboration, the analysis we performed helped the team redesign their intake surveys to track metrics they care about better.
DanceMuse (Bash, PyTorch, Slurm, FFmpeg, OpenPose) - Year long project dedicated to increasing the ease of use of dance-generation deep learning models so that they can be a tool for the artist community. Through collaboration with alumni from the Cooper Union Art School, we decided that the best way to do this is to pipeline dataset creation and training, which is what we did! [Repo]
DefensiveLayer (Tensorflow, Foolbox) - My first try at independent deep-learning based research (and I loved it!). My partner and I decided to investigate the use of intra-network layers as a way to defend against adversarial attacks, turns out they help! (but also reduce accuracy, so this ends up being a tradeoff).
Financial Signal Processing (course) (Markov chains, Monte Carlo Simulations, Stochastic Differential Equations, Time-Series Analysis, ARMA, GARCH) - A collection of projects dealing with finance but using a lot of really cool techniques based on signal processing. I loved learning about all the tools used to model financial data, and even more so loved to code them up myself and learn to use them.
Other Work I Have Done, in Far Less Detail:
- Digital Image Processing - Super-resolution (PyTorch, OpenCV) - Independent study
- Frequentist Machine Learning Projects (Sklearn, Xgboost, Mlxtend, Surprise, NumPy, Pandas, Seaborn, Matplotlib) - Course projects
- Remote Sensing projects (TensorFlow, Sklearn, Pandas, NumPy, Seaborn, Matplotlib, Rasterio, GDAL, Folium) - Course projects
- Natural Language Processing (NumPy, Pandas, NLTK, Logging, Argparse, PyTorch) - Course projects
- Sound & Space (Matlab) - Coursework
There is nothing better than chatting and sharing my work with others :)