VietAI TechTalk #3
Theme: Building AI that learns to help human users.
Talk 1 by Tuan Lai
Title: Recent advancements in NLP: from algorithms (RNNs) to applications (Question Answering systems).
Abstract: When evaluating a potential product purchase, customers may have many questions in mind. They want to get adequate information to determine whether the product of interest is worth their money. Building an automated QA system for product information can be a great way to improve customer’s experience. In the first part of this talk, I will describe how we build a system for answering questions related to product facts and specifications using deep learning and NLP.
Recurrent Neural Networks (RNNs) are a type of artificial neural networks that have shown great promise in many NLP tasks such as machine translation, language modeling, and question answering. Even though the Long Short-Term Memory (LSTM) and the Gated Recurrent Units (GRU) are two of the most popular kinds of RNNs, researchers have proposed various other kinds of RNNs over the years (e.g., Recurrent Additive Networks). In the second part of this talk, I will discuss the basic types of RNNs such as LSTM and GRU and their applications in NLP. After that, I will introduce other recently proposed RNNs.
Tuan Lai received his bachelor’s degree in Computer Science from KAIST in 2017. He graduated with honors and was ranked 1st in the department. He will start his PhD at the Purdue University from Fall 2018, working on Machine Learning and Natural Language Processing. He has interned at Google for two consecutive summers. In addition, he has interned at Adobe Research for 9 months, working on an intelligent shopping assistant. He has filed 1 patent and co-authored 5 papers that have appeared at conferences such as NAACL, ACL, COLING, ...
Talk 2 by Hung Ngo
Title: Meta Decision Processes for Human-Robot Collaboration Learning
Abstract: Humans, and some animals, learn to smoothly and effectively participate in collaborative tasks by first observing others, then asking questions sometimes, and gradually shifting to confident execution of the shared plan. We propose a novel computational framework, called meta decision processes, to capture the essences of such phenomena and transfer to human-robot cooperation settings. We formalize the meta decision processes (meta-DP) using partially observable Markov decision process (POMDP), and provide efficient approximate solutions for realtime, online planning. Simulated experiments from concurrent cooperation scenarios in relational domains and experiments on real robots show the effectiveness of the proposed framework.
Hung Ngo is both a research scientist and an educator working in the fields of AI, machine learning, and robotics. He was a PhD student of Juergen Schmidhuber at the Swiss AI Lab IDSIA, then a postdoc researcher of Marc Toussaint at MLR Lab, Stuttgart university. Follow him at fb.com/curiousAI.