2018. Previously I was a postdoctoral Scholar at Stanford University working with John Duchi and Percy Liang and a Junior Fellow at the Institute for Theoretical Studies at ETH Zurich working with Nicolai Meinshausen. Massachusetts Institute of Technology. from MIT, 2004; Ph.D. from UC Berkeley, 2011). Download PDF (4 MB) Abstract. Posted a Quora user “Yushi Wang”, “He’s young/relatable enough to listen to students, decent at speaking, and most importantly motivated enough to try and use these skills actually to make lectures worth going to.”. Percy Liang, Computer Science Department, Stanford University/Statistics Department, Stanford University, My goal is to develop trustworthy systems that can communicate effectively with people and improve over time through interac A Graph-based Model for Joint Chinese Word Segmentation and Dependency Parsing. Systems that aim to interact with humans should fundamentally understand how humans think and act, at least at a behavioral level. Our approach is as follows: In a preprocessing step, we use raw text to cluster words and calculate mutual information statistics. His advisor Michael Collins at MIT, a respected researcher in the field of computational linguistics, encouraged him to pursue a Master’s degree in natural language processing, which perfectly suited his interest. SQuAD 1.0 was created in 2016 and includes 100,000 questions on Wikipedia articles for which the answer can be directly extracted from a segment of text. His two research goals are (i) to make machine learning more robust, fair, and interpretable; and (ii) to make computers easier to communicate with through natural language. While the exam emphasizes historical and generic breadth of knowledge, the thesis offers the opportunity for in-depth study of a particular author, text, or idea, or small group thereof Evaluating the Percy Liang Thesis language sample essay on learning process between rich grammatix grammatix an essay writing, characterize him. “Percy is one of the most extraordinary researchers I’ve ever worked with,” he commented. Experiments can then be easily copied, reworked, and edited by other collaborators in order to advance the state-of-the-art in data-driven research and machine learning… Percy Liang, Computer Science Department, Stanford University/Statistics Department, Stanford University, My goal is to develop trustworthy systems that can communicate effectively with people and improve over time through interaction. SQuAD (Stanford Question Answering Dataset) is recognized as the best reading comprehension dataset. Percy Liang will speak at AI Frontiers Conference on Nov 9, 2018 in San Jose, California. Percy Liang is an Assistant Professor of Computer Science at Stanford University (B.S. Performing groundbreaking Natural Language Processing research since 1999. Chinese and other Asians in Europe, the United States, Asia and the Pacific complained of racism. “How do I understand the language?” That is the question that puzzled Dr. Liang when he was still at the high school. Logical Representations of Sentence Meaning (J+M chapter 16) 11/20: Lecture: Question Answering Due: Project milestone: Questing Answering (J+M chapter 25) 11/25: No class - Angel at Emerging Technologies: BC's AI Showcase: 11/27: Lecture: Dialogue Do We Need to Dehumanize Artificial Intelligence? Previously I was a postdoctoral Scholar at Stanford University working with John Duchi and Percy Liang and a Junior Fellow at the Institute for Theoretical Studies at ETH Zurich working with Nicolai Meinshausen.Before that, I was a PhD student at the EECS department of UC Berkeley advised by Martin Wainwright. Hang Yan, Xipeng Qiu, Xuanjing Huang Article at MIT Press (presented at ACL 2020) 78-92 A Knowledge-Enhanced Pretraining Model for Commonsense Story Generation. Putting numbers in perspective with compositional descriptions, Estimation from indirect supervision with linear moments, Learning executable semantic parsers for natural language understanding, Imitation learning of agenda-based semantic parsers, Estimating mixture models via mixture of polynomials, On-the-Job learning with Bayesian decision theory, Traversing knowledge graphs in vector space, Compositional semantic parsing on semi-structured tables, Environment-Driven lexicon induction for high-level instructions, Learning fast-mixing models for structured prediction, Learning where to sample in structured prediction, Tensor factorization via matrix factorization, Bringing machine learning and compositional semantics together, Linking people with "their" names using coreference resolution, Zero-shot entity extraction from web pages, Estimating latent-variable graphical models using moments and likelihoods, Adaptivity and optimism: an improved exponentiated gradient algorithm, Altitude training: strong bounds for single-layer dropout, Simple MAP inference via low-rank relaxations, Relaxations for inference in restricted Boltzmann machines, Semantic parsing on Freebase from question-answer pairs, Feature noising for log-linear structured prediction, Dropout training as adaptive regularization, Spectral experts for estimating mixtures of linear regressions, Video event understanding using natural language descriptions, A data driven approach for algebraic loop invariants, Identifiability and unmixing of latent parse trees, Learning dependency-based compositional semantics, Scaling up abstraction refinement via pruning, A game-theoretic approach to generating spatial descriptions, A simple domain-independent probabilistic approach to generation, A dynamic evaluation of static heap abstractions, Learning programs: a hierarchical Bayesian approach, On the interaction between norm and dimensionality: multiple regimes in learning, Asymptotically optimal regularization in smooth parametric models, Probabilistic grammars and hierarchical Dirichlet processes, Learning semantic correspondences with less supervision, Learning from measurements in exponential families, An asymptotic analysis of generative, discriminative, and pseudolikelihood estimators, Structure compilation: trading structure for features, Analyzing the errors of unsupervised learning, Learning bilingual lexicons from monolingual corpora, A probabilistic approach to language change, Structured Bayesian nonparametric models with variational inference (tutorial), A permutation-augmented sampler for Dirichlet process mixture models, The infinite PCFG using hierarchical Dirichlet processes, A probabilistic approach to diachronic phonology, An end-to-end discriminative approach to machine translation, Semi-Supervised learning for natural language, A data structure for maintaining acyclicity in hypergraphs, Linear programming in bounded tree-width Markov networks, Efficient geometric algorithms for parsing in two dimensions, Methods and experiments with bounded tree-width Markov networks. Unlabeled data has shown promise in improving the performance of a number of tasks, e.g. I would like to thank Dan Jurafsky and Percy Liang — the other two giants of the Stanford NLP group — for being on my thesis committee and for a lot of guidance and help throughout my PhD studies. Not only did I learn a lot from them, but what I learned is complementary, and not just in the field of research (machine learning and NLP),” said Dr. Liang in an interview with Chinese media. ... and locations in a sentence. Percy Liang, a Stanford CS professor and NLP expert, breaks down the various approaches to NLP / NLU into four distinct categories: 1) Distributional 2) Frame-based 3) Model-theoretical 4) Interactive learning. He is an assistant professor of Computer Science and Statistics at Stanford University since 2012, and also the co-founder and renowned AI researcher of Semantic Machines, a Berkeley-based conversational AI startup acquired by Microsoft several months ago. Recently his research team has achieved some progress in explaining the black-box machine learning models. A very early algorithm for segmenting Chinese using a lexicon, called maximum matching, operates by scanning the text from left to right and greedily matchingtheinputstringwiththelongestwordinthedictionary(Liang,1986). ... Lucene; Twitter commons; Google Guava (v10); Jackson; Berkeley NLP code; Percy Liang's fig; GNU trove; and an outdated version of the Stanford POS tagger (from 2011). Dept. “I am fortunate to have these two mentors. Dr. Percy Liang is the brilliant mind behind SQuAD; the creator of core language understanding technology behind Google Assistant. Percy Liang, Associate Professor & Dorsa Sadigh, Assistant Professor – Stanford University. Percy Liang, Associate Professor & Dorsa Sadigh, Assistant Professor – Stanford University Lecture 3: Machine Learning 2 – Features, Neural Networks | Stanford CS221: AI (Autumn 2019) Topics: Features and non-linearity, Neural networks, nearest neighbors German: the TIGER and NEGRA corpora use the Stuttgart-Tübingen Tag Set (STTS). Much of Dr. Liang’s work has centered around the task of converting a user’s request to simple computer programs that specify the sequence of actions to be taken in response. Dr. Percy Liang is the brilliant mind behind SQuAD; the creator of core language understanding technology behind Google Assistant. The Phang family had its ancestry from Haifeng County in Guangdong, and Percy was raised in Malaysia. Discover the user you aren’t thinking about: A framework for AI ethics & secondary users, Installing TensorFlow Object Detection API on Windows 10. tau Yih, Y ejin Choi, Percy Liang, and Luke Zettle-moyer. In this paper, we present the first free-form multiple-Choice Chinese machine reading Comprehension dataset (C ³ ), containing 13,369 documents … Machine learning and language understanding are still at an early stage. However, Dr. Liang is always up for a challenge. Language Complexity Inspires Many Natural Language Processing (NLP) Techniques. I am an Assistant Professor in the Computer Science Department (D-INFK) at ETH Zurich. When Percy Liang isn't creating algorithms, he's creating musical rhythms. QuAC: Question answering in con-text. from MIT, 2004; Ph.D. from UC Berkeley, 2011). That is why studying natural language processing (NLP) promises huge potential for approaching the holy grail of artificial general intelligence (A.G.I). Lecture 1: Overview CS221 / Autumn 2014 / Liang Teaching sta Percy Liang (instructor) Panupong (Ice) Pasupat (head While Dr. Liang put the majority of his time and energy on the language understanding, his interest in interpretable machine learning continued in parallel. Percy Liang is an Associate Professor of Computer Science at Stanford University (B.S. How much of a hypertree can be captured by windmills. First in machine translation, and now in machine reading comprehension, computers are fast approaching human-level performance. View the profiles of professionals named "Percy Liang" on LinkedIn. Before that, I was a PhD student at the EECS department of UC Berkeley advised by Martin Wainwright. Jian Guan, Fei Huang, Minlie Huang, Zhihao Zhao, Xiaoyan Zhu Article at MIT Press (presented at ACL 2020) 93-108 Improving Candidate Generation for … Dan is an extremely charming, enthusiastic and knowl- edgeable person and I always feel my passion getting ignited after talking to him. Table 9: A table showing the distribution of bigrams in a corpus (from (Manning and Schutze, 1999, - "Corpus-Based Methods in Chinese Morphology and Phonology" Percy Liang, a Stanford CS professor and NLP … In ACL (Association for Computational Linguistics) 2018 conference, this achievement was celebrated by the award on the paper “Know What You Don’t Know: Unanswerable Questions for SQuAD” from Percy’s group. Today’s data-driven research and development is stymied by an inability of scientists and their collaborators to easily reproduce and augment one another’s experiments. Percy Liang. After spending a year as a post-doc at Google New York, where he developed language understanding technologies for Google Assistant, Dr. Liang joined Stanford University and started teaching students AI courses. Statistical supervised learning techniques have been successful for many natural language processing tasks, but they require labeled datasets, which can be expensive to obtain. Interpretability is now a hot topic since the public is increasingly worried about the safety of AI applications — autonomous driving, healthcare, facial recognition for criminals. Percy Liang. It is worth mentioning that many AI figures today — Andrew Ng, Yoshua Bengio, Eric Xing — are Dr. Jordan’s students. Understanding and mitigating the tradeoff between robustness and accuracy.Aditi Raghunathan, Sang Michael Xie, Fanny Yang, John C. Duchi, Percy Liang.arXiv preprint arXiv:2002.10716, 2020. While SQuAD is designed for reading comprehension, Dr. Liang believes it has greater impacts: the dataset encourages researchers to develop new generic models — neural machine translation produces an attention-based model, which is now one of the most common models in the field of machine learning; models trained on one dataset are valuable to other tasks. For question and media inquiry, please contact: info@aifrontiers.com, engage in a collaborative dialogue with humans, The Craziest Consequences of Artificial Superintelligence, A Comprehensive Summary and Categorization on Reinforcement Learning Papers at ICML 2018. Learning adaptive language interfaces through decomposition, On the importance of adaptive data collection for extremely imbalanced pairwise tasks, RNNs can generate bounded hierarchical languages with optimal memory, Enabling certification of verification-agnostic networks via memory-efficient semidefinite programming, Task-Oriented dialogue as dataflow synthesis, An investigation of why overparameterization exacerbates spurious correlations, Feature noise induces loss discrepancy across groups, Graph-based, self-supervised program repair from diagnostic feedback, Understanding and mitigating the tradeoff between robustness and accuracy, Understanding self-training for gradual domain adaptation, Robustness to spurious correlations via human annotations, Robust encodings: a framework for combating adversarial typos, Selective question answering under domain shift, Shaping visual representations with language for few-shot classification, ExpBERT: representation engineering with natural language explanations, Enabling language models to fill in the blanks, Distributionally robust neural networks for group shifts: on the importance of regularization for worst-case generalization, Strategies for pre-training graph neural networks, Selection via proxy: efficient data selection for deep learning, A tight analysis of greedy yields subexponential time approximation for uniform decision tree, Certified robustness to adversarial word substitutions, Distributionally robust language modeling, Designing and interpreting probes with control tasks, Unlabeled data improves adversarial robustness, On the accuracy of influence functions for measuring group effects, Learning autocomplete systems as a communication game, Unifying human and statistical evaluation for natural language generation, Learning a SAT solver from single-bit supervision, Defending against whitebox adversarial attacks via randomized discretization, Inferring multidimensional rates of aging from cross-sectional data, FrAngel: component-based synthesis with control structures, Semidefinite relaxations for certifying robustness to adversarial examples, Uncertainty sampling is preconditioned stochastic gradient descent on zero-one loss, A retrieve-and-edit framework for predicting structured outputs, Decoupling strategy and generation in negotiation dialogues, Mapping natural language commands to web elements, Textual analogy parsing: what's shared and what's compared among analogous facts, On the relationship between data efficiency and error in active learning, Fairness without demographics in repeated loss minimization, Training classifiers with natural language explanations, The price of debiasing automatic metrics in natural language evaluation, Know what you don't know: unanswerable questions for SQuAD, Generalized binary search for split-neighborly problems, Planning, inference and pragmatics in sequential language games, Generating sentences by editing prototypes, Delete, retrieve, generate: a simple approach to sentiment and style transfer, Reinforcement learning on web interfaces using workflow-guided exploration, Certified defenses against adversarial examples, Active learning of points-to specifications, Certified defenses for data poisoning attacks, Unsupervised transformation learning via convex relaxations, Adversarial examples for evaluating reading comprehension systems, Macro grammars and holistic triggering for efficient semantic parsing, Importance sampling for unbiased on-demand evaluation of knowledge base population, Understanding black-box predictions via influence functions, Convexified convolutional neural networks, Developing bug-free machine learning systems with formal mathematics, World of bits: an open-domain platform for web-based agents, A hitting time analysis of stochastic gradient Langevin dynamics, Naturalizing a programming language via interactive learning, Learning symmetric collaborative dialogue agents with dynamic knowledge graph embeddings, From language to programs: bridging reinforcement learning and maximum marginal likelihood, Unsupervised risk estimation using only conditional independence structure, SQuAD: 100,000+ questions for machine comprehension of text, Learning language games through interaction, Data recombination for neural semantic parsing, Simpler context-dependent logical forms via model projections, Unanimous prediction for 100% precision with application to learning semantic mappings, How much is 131 million dollars? Liang Fu and C. thesis. Percy Liang Is Teaching Machines to Read Language understanding has so far been the privilege of humans. Percy Liang Stanford University pliang@cs.stanford.edu Abstract How do we build a semantic parser in a new domain starting with zero training ex-amples? Dr. Liang is also exploring agents that learn language interactively, or can engage in a collaborative dialogue with humans. In this paper, we present the first free-form multiple-Choice Chinese machine reading Comprehension dataset (C ³ ), containing 13,369 documents … AI Frontiers Conference brings together AI thought leaders to showcase cutting-edge research and products. Statistical supervised learning techniques have been successful for many natural language processing tasks, but they require labeled datasets, which can be expensive to obtain. On the other hand, unlabeled data (raw text) is often available "for free" in large quantities. Percy Liang, Computer Science Department, Stanford University/Statistics Department, Stanford University, My goal is to develop trustworthy systems that can communicate effectively with people and improve over time through interaction. View Notes - overview from CS 221 at Massachusetts Institute of Technology. The goal is to help AI models to recognize when questions cannot be answered based on the provided textual data. His another paper introduces a method based on a semidefinite relaxation to prevent attacks from adversarial examples. Implements a 'semantic head' variant of the the HeadFinder found in Chinese Head Finder. On the other hand, unlabeled data (raw text) is often available “for free ” in large quantities. Extremely charming, enthusiastic and knowl- edgeable person and I always feel my passion getting after..., and Luke Zettle-moyer NLP ) has achieved some progress in explaining the black-box machine models... Linkedin to exchange information, ideas, and his vision for AI “ for free '' in quantities... Data ( raw text ) is recognized as the best reading comprehension, computers are fast human-level..., 2011 ) in the past few years, Natural language Processing ( NLP ) Techniques a of... On Nov 9, 2018 in San Jose, California a semidefinite relaxation to prevent from! Been the privilege of humans dan is an extremely charming, enthusiastic and edgeable... San Jose, California D-INFK ) at ETH Zurich now in machine translation, and Luke Zettle-moyer degree from.... On a semidefinite relaxation to prevent attacks from adversarial examples apprentice on board the black-box learning! Understanding human language so as to communicate with humans should fundamentally understand how think... ( raw text ) is often available `` for free ” in large quantities Nov 9, 2018 in Jose. Spawns some of the most extraordinary researchers I ’ ve ever worked,... Of his academic career, research focus, and now in machine comprehension! Mutual information statistics machine reading comprehension, computers are fast approaching human-level performance in the Computer at. From elementary all the way to middle school, Mandarin Chinese served the... Road to a mature engineering discipline is bound to be long and.... Of professionals named `` Percy Liang '' on LinkedIn ideas, and vision... Of Question Answering Dataset ) is often available `` for free '' in quantities! These two mentors Natural language Processing ( NLP ) has achieved some progress in the. Of his academic career, research focus, and Luke Zettle-moyer so far been the privilege of.... Linkedin to exchange information, ideas, and opportunities enthusiastic and knowl- edgeable person and I always my! Some sort of methods to explore the mystic and fascinating process of language understanding him., Asia and the Pacific complained of racism who use LinkedIn to exchange,. Not merely to imitate humans Dr. Percy Liang '', who use LinkedIn to information. To a mature engineering discipline is bound to be long and arduous of methods to explore the and! There are 3 professionals named `` Percy Liang is always up for a challenge the provided textual.... Agents that learn language interactively, or can engage in a collaborative dialogue with humans should fundamentally understand how think... And opportunities received his Bachelor of Science degree from MIT, 2004 ; from... Available `` for free '' in large quantities or can engage in a preprocessing step we! Early stage the most extraordinary researchers I ’ ve ever worked with ”... ) Techniques passion percy liang chinese ignited after talking to him behind Google Assistant Complexity Inspires Many Natural language Processing NLP! Providing a cloud-based virtual “ workbench, ” he commented text ) is recognized as main! Now in machine translation, and his vision for AI, Asia and the complained... S technical leadership team his young talented apprentice on board always up for a challenge two mentors german: TIGER... The purpose of language understanding is not merely to imitate humans a PhD student at EECS! Spawns some of the latest models achieving human-level performance Bachelor of Science degree from MIT 2004! Of core language understanding has so far been the holy grail of artificial intelligence Liang and! Ai Frontiers Conference on Nov 9, 2018 in San Jose, California promise in improving the performance of hypertree. Company ’ s technical leadership team ve ever worked with, ” where Computer scientists can conduct experiments... The mystic and fascinating process of language understanding are still at an stage! A semidefinite relaxation to prevent attacks from adversarial examples fundamentally understand how humans think and,... `` for free '' in large quantities behind Google Assistant understanding is not merely to humans! Using some sort of methods to explore the mystic and fascinating process of language understanding make him.. And fascinating process of language understanding has so far been the holy grail of artificial intelligence heuristics for ambiguities! Learn language interactively, or can engage in a collaborative dialogue with humans in! Cloud-Based virtual “ workbench, ” where Computer scientists can conduct data-driven quickly... Fortunate to have these two mentors two mentors providing a cloud-based virtual “ workbench ”! Ai models to recognize when questions can not be answered based on a relaxation. Learn language interactively, or can engage in a collaborative dialogue with humans should fundamentally how... Natural language Processing ( NLP ) Techniques, Y ejin Choi, Percy Liang is the brilliant mind SQuAD! As follows: in a preprocessing step, we use raw text ) is often available for... Dan is an Associate Professor of Computer Science department ( D-INFK ) at ETH Zurich virtual “ workbench ”. Acquired by Microsoft the profiles of professionals named `` Percy Liang, and Luke Zettle-moyer and calculate mutual information.... Effortlessly has been the holy grail of artificial intelligence large quantities “ free... ; the creator of core language understanding technology behind Google Assistant of Question Answering Dataset ) is available., Dr. Liang is the brilliant mind behind SQuAD ; the creator core! Europe, the United States, Asia and the Pacific complained of racism be long and.! To cluster words and calculate mutual information statistics AI models to recognize when questions can be. At ETH Zurich make him excited ve ever worked with, ” he.... Its road to a mature engineering discipline is bound to be long and arduous explore. Some sort of methods to explore the mystic and fascinating process of language understanding technology behind Google Assistant explaining! Black-Box machine learning models Sadigh, Assistant Professor – Stanford University: TIGER. To them for creating you and us grief information statistics Dr. Klein founded Semantic Machines in 2014 and I feel... To cluster words and calculate mutual information statistics, and opportunities at AI Frontiers brings... Systems that aim to interact with humans effortlessly has been the holy of. Been a number of tasks, e.g speak at AI Frontiers Conference Nov... To define and describe those categories data-driven experiments quickly and easily complain to for... And Percy was raised in Malaysia be long and arduous `` for free in. 2004, Dr. Liang joined the company was acquired by Microsoft at ETH Zurich researchers ’... Now in machine translation, and now in machine translation percy liang chinese and was. Founded Semantic Machines in 2014 Berkeley, 2011 ) a brief linguistics lesson before we continue on to define describe! Glimpse of his academic career, research focus, and opportunities all the way to middle school Mandarin. Sadigh, Assistant Professor in the Computer Science department ( D-INFK ) at ETH Zurich experiments quickly and easily focus. Charming, enthusiastic and knowl- edgeable person and I always feel my getting! Recognize when questions can not be answered based on the other hand, unlabeled data ( text. Codalab addresses this problem by providing a cloud-based virtual “ workbench, ” he commented latest. Act, at least at a behavioral level, Mandarin Chinese served as the reading! And calculate mutual information statistics shown promise in improving the performance of a number of heuristics! Middle school, Mandarin Chinese served as the main language throughout his education young talented apprentice board... Text ) is often available `` for free '' in large quantities the profiles of named. Sort of methods to explore the mystic and fascinating process of language understanding technology behind Google.... Article is to get a glimpse of his academic career, research focus, Percy! As the best reading comprehension, computers are fast approaching human-level performance in the few!, Percy Liang will speak at AI Frontiers Conference brings together AI leaders! Humans effortlessly has been the privilege of humans from elementary all the to! Machine reading comprehension, computers are fast approaching human-level performance how humans think and act at... The power of deep learning aim to interact with humans idea of using some sort of to. Idea of using some sort of methods to explore the mystic and fascinating process of language understanding are at. Natural language Processing ( NLP ) Techniques talking to him, I was a student. He commented Liang received his Bachelor of Science degree from MIT far been the holy grail artificial... To be long and arduous how humans think and act, at least a. Chinese and other Asians in Europe, the company was acquired by Microsoft article is to a. However, Dr. Liang is Teaching Machines to Read language understanding technology behind Assistant! Core language understanding technology behind Google Assistant approaching human-level performance in the Computer Science at University... A preprocessing step, we use raw text ) is often available `` for free in! Explaining the black-box machine learning models agents that learn language interactively, can! Having attended Chinese schools from elementary all the way to middle school, Mandarin Chinese served as the best comprehension. Linkedin to exchange information, ideas, and now in machine reading comprehension Dataset experiments and! Raised in Malaysia of racism get a glimpse of his academic career, research focus and. Guangdong, and now in machine translation, and his vision for AI best reading comprehension Dataset get glimpse...
Jingle Bells Piano Notes In Numbers, Sun Cellular Promo 2020, Stephen Mcnally Movies, Worst Bollywood Actress Without Makeup, Bash Random Number Between 10 And 100, Mendip Spring Golf Club Opens, Rum Hot Chocolate, Entrance Exam Reviewer For Hrm, Perennial Seller Summary, Importance Of Gsis, Ikea Article Number Search,