主讲人：黄亮（Liang Huang）The City University of New York (CUNY)联系人：黄萱菁 xjhuang@fudan.edu.cn地点：张江校区计算机楼 405 室Bio:Liang Huang is currently an Assistant Professor at the City University of New York (CUNY). He graduated in 2008 from UPenn and has worked as a Research Scientist at Google and a Research Assistant Professor at USC/ISI. His work is mainly on the theoretical aspects (algorithms and formalisms) of computational linguistics, and related theoretical problems in machine learning. He has received a Best Paper Award at ACL 2008, several best paper nominations (ACL 2007, EMNLP 2008, and ACL 2010), two Google Faculty Research Awards (2010 and 2013), and a University Graduate Teaching Prize at Penn (2005). In his past life before switching to NLP he also co-authored the popular “Black Book” for programming contests. Fore more information, please visit: http://acl.cs.qc.edu/.讲座一：Linear-time Algorithms in Natural Language Understanding and Learning时间：8 月 25 日（周一）上午 10 点Abstract：Why are computers so bad at understanding natural language and why are human-beings so much better at it? Can we build a model to simulate human language processing so that computers can process human language the same way we humans do, i.e., fast, incremental (left-to-right) and accurate?In this talk I'll present a linear-time dynamic programming model for incremental parsing inspired by human sentence processing (from psycholinguistics) as well as compiler theory (LR parsing). This model, being linear-time, is much faster than, but also as accurate as, the dominant cubic-time algorithms. It overcomes the ambiguity explosion problem by approximate dynamic programming, which corresponds to local ambiguity packing in psycholinguistics.But how do we efficiently learn such a parsing model with approximate inference from huge amounts of data? We propose a general structured machine learning framework based on the structured perceptron that is guaranteed to succeed with inexact search and works well in practice. Our new learning algorithm can learn a large-scale state-of-the-art parsing model with dramatically reduced training time, thus having the potential to scaling up to the whole Web. More importantly, our learning algorithms are widely applicable to other structured domains such as bioinformatics.讲座二：Scalable Large-Margin Structured Learning: Theory and Algorithms时间：8 月 27 日（周三）上午 10 点、下午 1 点半Abstract:Much of NLP tries to map structured input (sentences) to some form of structured output (tag sequences, parse trees, semantic graphs, or translated/paraphrased/compressed sentences). Thus structured prediction and its learning algorithm are of central importance to us NLP researchers. However, when applying machine learning to structured domains, we often face scalability issues for two reasons:1. Even the fastest exact search algorithms for most NLP problems (such as parsing and translation) is too slow for repeated use on the training data, but approximate search (such as beam search) unfortunately breaks down the nice theoretical properties (such as convergence) of existing machine learning algorithms.2. Even with inexact search, the scale of the training data in NLP still makes pure online learning (such as perceptron and MIRA) too slow on a single CPU.This tutorial reviews recent advances that address these two challenges. In particular, we will cover principled machine learning methods that are designed to work under vastly inexact search, and parallelization algorithms that speed up learning on multiple CPUs. We will also extend structured learning to the latent variable setting, where in many NLP applications such as translation and semantic parsing the gold-standard derivation is hidden.Contents:• Overview of Structured Learning• key challenge 1: search efficiency• key challenge 2: interactions b/w search and learning• Structured Perceptron• the basic algorithm• the convergence proof (a geometric view)• voted perceptron, averaged perceptron and efficient implementation tricks• applications in tagging, parsing, etc.• Structured Perceptron under Inexact Search• convergence theory breaks under inexact search• early update• violation-fixing perceptron• applications in tagging, parsing, etc.Break• From Perceptron to MIRA• 1-best MIRA; geometric solution• k-best MIRA; hildreth algorithm• MIRA with all constraints; loss-augmented decoding• MIRA under inexact search• Large-Margin Structured Learning with Latent Variables• examples: machine translation, semantic parsing, transliteration• separability condition and convergence proof• latent-variable perceptron under inexact search• applications in machine translation• Parallelizing Large-Margin Structured Learning• iterative parameter mixing (IPM)• minibatch perceptron and MIRA讲座三： How to Succeed in PhD Study时间：8 月 29 日（周五）上午 10 点Abstract:• What are the most important skills to be learned in a PhD study?• Which year is the most difficult year?• How to choose a research topic?• How to improve your writing/presentation skills?• How to choose an advisor and what’s his/her role in your study?
