Patrick L.

I bring disruptive and innovative solutions to your Big Data challenges.

I specialize in Machine Learning, Deep Learning, Computer Vision, NLP, and RESTful Web App Design. Specifically, I can solve your data analytics problems and design a web application as an entry point for your users to interact with the product.

Among the projects I have completed:

Automatic verification of medical certificates for the organization of a running race

- Identification of printed and handwritten text with SVMs trained on BoVW of SIFT descriptors

- Use of an OCR module (pytesseract) for printed text transcription (accuracy: 65%), and a convolutional recurrent neural network (CRNN) for handwritten text

Prediction of Bitcoin prices with Bayesian neural networks.

- Implementation of a Bayesian Recurrent Neural Network (BNN) in Keras for Bitcoin price prediction

- BNN advantage: provides model uncertainty unlike a standard recurrent neural network (SNN), while improving performance

- Extraction of relevant time series data from the web using Selenium and BeautifulSoup

- Results: BNN(LSTM+MonteCarloDropout)-RMSE: 0.099, SNN(LSTM+Dropout)-RMSE: 0.111

Recommendation of tags when entering questions on StackOverflow.

- Development of a tag suggestion system that automatically assigns relevant tags to a question for IT development Q&A

- Training and comparison of supervised (SVM, MaxEnt, Multinomial Naive Bayes) and unsupervised (LSA, PLSA, LDA) models

- Results: Supervised model: SVM + tf-idf vectorizer - F1 score: 56%, Unsupervised model: PLSA + count vectorizer - F1 score: 14%