Семинар специалиста из Германии Сергиенко Романа по обработке естественного языка

30 декабря состоялся научный семинар специалиста из Германии Сергиенко Романа Борисовича по обработке естественного языка.

Dr. Dr. rer. nat. ROMAN SERGIENKO
Machine Learning Engineer for Natural Language Processing
Gini GmbH, Munich, Germany

Topic: NATURAL LANGUAGE PROCESSING

1. General overview of NLP tasks (named entity recognition, text classification, semantic analysis, machine translation etc.)
2. Text Vectorization
3. Term Weighting
4. Word embeddings
   • Word vectors
   • Sub-word vectors
   • Content-based word vectors
5. Neural Networks for NLP
6. Milestones of Deep Learning
   • Unsupervised pretraining (encoder-decoders)
   • Modern normalization and regularization
   • Modern gradient methods and activation functions
7. NN architectures for NLP
   • Feedforward NNs (example: relevance detection of news from Internet)
   • Recurrent NNs (example: Bi-LSTM for named entity recognition)
   • Convolutional NNs (example: character-level CNN)
   • Attention mechanism and Transformers
   • Chargrid for 2D Text (example: information extraction from invoices)