Dear appliedAI Ecosystem,
appliedAI is happy to welcome Dr. Han Xiao from Tencent on the 8th of January 2019 from 9:00 am to 12:30 pm at the UnternehmerTUM, Lichtenbergstraße 6, 85748 Garching, room Anton ground floor. He will talk about his work at Tencent and his recent launch of the German-Chinese Association for Artificial Intelligence (GCAAI).
Furthermore, we welcome Philipp Dufter from CIS at LMU Munich, who is doing exciting work on Multilingual Contextualized Embeddings, which is a promising direction in Natural Language Processing.
Agenda:
9:30 - 9:45: Introduction to the appliedAI Initiative
9:45 - 10:45 Han Xiao (Tencent): "Neural Information Retrieval Applications at Tencent AI Lab"
10:45 - 11:00 Coffee break
11:00 - 11:45: Philipp Dufter: "Multilingual Contextualized Embeddings"
11:45 - 12:30 Networking
Abstract of talks:
1. Neural Information Retrieval Applications at Tencent AI Lab
Recent advances of deep learning in NLP make us rethink the traditional information retrieval system, in particular, those represent query and the doc using hand-crafted features and store an inverted index, e.g. Elasticsearch, Lucene. By contrast, more recently proposed neural models can learn representations of language from raw text that bridge the gap between query and document “automagically". In this talk, I will give an overview of how Tencent AI Lab apply neural IR to real applications such as Chatbot, Lyrics Search, and Patent Search. Techniques I will cover including representation, indexing, storage, evaluation, and real-time serving. I’d like to show you that much of our understanding of the traditional approach from decades of research can be extended to these modern deep learning models and that old inverted index may still be your good friend.
Speaker: Dr. Han Xiao
Dr. Han Xiao is a Senior Scientist III at Tencent AI Lab, leading a team on neural information retrieval. He is also the Chairman of German-Chinese Association of Artificial Intelligence. Before joining Tencent, Han was a senior research scientist at Zalando from 2014 to 2018, building search and recommendation systems. Han is the creator of Fashion-MNIST, one of the most popular AI open-source projects of 2017. Dr. Han Xiao completed Ph.D. and M.Sc. in computer science and both at Technical University of Munich in Germany. His doctoral dissertation (2014) is about adversarial and robust machine learning.
2. Multilingual Contextualized Embeddings
Monolingual word embeddings which are based on local context have been a success story in NLP for many years. Since then they have been extended in several directions: multilingual embeddings have been created, embeddings are deemed useful for domain adaptation and quite recently contextualised word embeddings gained traction. In this talk we will introduce the basics of word embeddings and subsequently dive deeper to discuss multilingual and/or contextualised word embeddings. Special focus will be on their usefulness for NLP-applications and on caveats when using them.
Speaker: Philipp Dufter
Philipp Dufter is currently a doctoral student in natural language processing supervised by Prof. Hinrich Schütze at Ludwig-Maximilians-Universität Munich. Prior to his PhD he completed a BSc in Mathematics with a focus on probability theory at Technical University of Munich. He continued his studies with a MSc in Business Analytics at Imperial College in London and wrote his Master's Thesis about stochastic and robust optimisation. His current work in computational linguistics combines the passion for probability theory and optimisation with highly relevant and fascinating linguistic applications, such as word learning word representations across 1000+ languages or making word embeddings more interpretable.
Please do not forget to register by clicking the "Register" button.
We are looking forward to welcoming you and to a successful day.
Best regards,
the appliedAI Team
