• Conduct experiments for cutting edge pre-trained language models (e.g. BERT, XLNet) and compressed models (e.g. ALBERT) for different NLP tasks, such as text classification, semantic text similarity etc. • Propose domain adapted language models to improve performance of text classification in the finance/law domains • Conduct experiments for diversity of embeddings, strategies for imbalance data, optimize single deep learning models, different ensembles to improve performance of… Show more • Conduct experiments for cutting edge pre-trained language models (e.g. BERT, XLNet) and compressed models (e.g. ALBERT) for different NLP tasks, such as text classification, semantic text similarity etc. • Propose domain adapted language models to improve performance of text classification in the finance/law domains • Conduct experiments for diversity of embeddings, strategies for imbalance data, optimize single deep learning models, different ensembles to improve performance of multi-label industry/topic classification, Strong implementation of codes, implement prototypes to test new ideas and algorithms using tensorflow • Actively implement new prototypes for company’s ML library and new features for Kepler platform. Owner of text pipelines (text classification and semantic text similarity) in NLP framework design and deep experience in text annotation, text augmentation, text standarization, embedding reduction and ML models (CNNs, RNNs, Transformer and language models etc.) for text pipelines, experience with big data using dask • Experience with multiple language text: English, Spain, Portuguese, French etc., experience with Git, AWS, Dask etc. Show less

• Sentiment analysis: use NLTK to preprocess raw twitter text, initialize word embeddings (glove and word2vec), improve existing ML models such as CNN, LSTM with attentions, Transformer for text classification• Machine translation: employ Byte Pair Encoding to learn joint sub words out of the full vocabulary, implement ML models such as seq2seq, RNN, Transformer for machine translation, adapt codes in tensorflow for multi-GPUs• Ethereum price prediction: difference time series data to… Show more • Sentiment analysis: use NLTK to preprocess raw twitter text, initialize word embeddings (glove and word2vec), improve existing ML models such as CNN, LSTM with attentions, Transformer for text classification• Machine translation: employ Byte Pair Encoding to learn joint sub words out of the full vocabulary, implement ML models such as seq2seq, RNN, Transformer for machine translation, adapt codes in tensorflow for multi-GPUs• Ethereum price prediction: difference time series data to remove its trend, transform time series prediction to supervised learning issue, propose an ensemble of non-linear regression algorithms for ETH price prediction• Implementing the machine learning models with Python, based on frameworks (tensorflow, scikit-learn and pytorch)• Experience with multi-GPUs programming Show less

• Implement E-Commerce web site using Java 8, Spring MVC, J2EE, Hibernate, Eclipse, MySQL • Use Freemarker and JSP to implement the code generator module to automatically generate Java and JSP files, which largely speed up the speed

• Dedicate to natural language processing, and image processing and recognition• Propose new machine learning algorithms and apply to detect abnormal ECG, in order to work with physicians to reduce costs and improve precision of early detection of heart diseases.• Adapt and improve existing machine learning models for machine translation using C++• Have publications in top AI journals such as Machine Learning, IEEE Transactions on Biomedical Engineering• Particle Swarm Optimizer… Show more • Dedicate to natural language processing, and image processing and recognition• Propose new machine learning algorithms and apply to detect abnormal ECG, in order to work with physicians to reduce costs and improve precision of early detection of heart diseases.• Adapt and improve existing machine learning models for machine translation using C++• Have publications in top AI journals such as Machine Learning, IEEE Transactions on Biomedical Engineering• Particle Swarm Optimizer for Variable Weighting in Clustering High-dimensional Data, Machine Learning, 82: 43-70, 2011.• ECG Analysis Using Multiple Instance Learning for Myocardial Infarction Detection, IEEE Transactions on Biomedical Engineering, 59(12), 2012 Show less