Conferences, Lectures, & Seminars
Events for November
-
AI Seminar-The Crisis in Statistics and the Reliability of Published Results
Fri, Nov 04, 2016 @ 11:00 AM - 12:00 PM
Information Sciences Institute
Conferences, Lectures, & Seminars
Speaker: Alan Garfinkel, UCLA
Talk Title: The Crisis in Statistics and the Reliability of Published Results
Series: Artificial Intelligence Seminar
Abstract: Medicine and biology are currently in a crisis. Many if not most published studies contain false or irreproducible information, and articles have appeared in major journals with titles like, Why Most Published Research Findings Are False.
A major contributor to this crisis is Bad Statistics. A large fraction of published papers use statistical methods that are simply wrong. The most frequent error is the use of formula-based statistical tests, like t-tests, regression, ANOVA etc. to give p values on data for which they cannot be used, because the data is either markedly non-Gaussian or too small to tell. These p-value calculation errors are currently being addressed by major journals, which have recently greatly tightened their statistical reviewing, and are insisting upon appropriate statistical methods, such as resampling-based test for non-Gaussian data.
But a much deeper criticism focuses on the very idea of p-values, however calculated. Many respected sources are calling for an end to p-values as the test for publishability. Phenomena like p-hacking are common, and advanced thinking now holds that the very idea of p-values is the problem. Several journals are now refusing to accept p-values as evidence of the existence of a phenomenon, and even the American Statistical Association has issued warnings about p-values.
We will review the situation, assess the extent of the damage, and discuss proposed fixes for this serious problem.
Biography: Dr. Garfinkel graduated from Cornell and received his PhD from Harvard in philosophy and mathematics. He is particularly interested in nonlinear dynamics and its applications to medicine.
Host: Gully Burns
Webcast: http://webcastermshd.isi.edu/Mediasite/Play/00f1e452277f416186bf9b6743e650131dLocation: Information Science Institute (ISI) - 11th Flr Conf Rm # 1135, Marina Del Rey
WebCast Link: http://webcastermshd.isi.edu/Mediasite/Play/00f1e452277f416186bf9b6743e650131d
Audiences: Everyone Is Invited
Contact: Peter Zamar
This event is open to all eligible individuals. USC Viterbi operates all of its activities consistent with the University's Notice of Non-Discrimination. Eligibility is not determined based on race, sex, ethnicity, sexual orientation, or any other prohibited factor. -
AI Seminar-TASK-GUIDED HETEROGENEOUS INFORMATION NETWORK EMBEDDING
Thu, Nov 10, 2016 @ 11:00 AM - 12:00 PM
Information Sciences Institute
Conferences, Lectures, & Seminars
Speaker: Yizhou Sun, UCLA
Talk Title: TASK-GUIDED HETEROGENEOUS INFORMATION NETWORK EMBEDDING
Series: Artificial Intelligence Seminar
Abstract: One of the challenges in mining information networks is the lack of intrinsic metric in representing nodes into a low dimensional space, which is essential in many mining tasks, such as anomaly detection, recommendation, and link prediction. Moreover, when coming to heterogeneous information networks, where nodes belong to different types and links represent different semantic meanings, it is even more challenging to represent nodes properly for a particular task. In this talk, I will introduce our recent progress of network embedding approaches that are designed for heterogeneous information networks and guided by specific tasks, and discuss (1) how to represent nodes when different types of nodes and links are involved; (2) how different tasks can guide the embedding process; and (3) how heterogeneous links play different roles in these tasks. Our results on several application domains, including enterprise networking, social network, bibliographic data, and biomedical data, have demonstrated the superiority as well as the interpretability of these new methodologies.
Biography: Yizhou Sun is an assistant professor at department of computer science of UCLA. Prior to that, she was an assistant professor in the College of Computer and Information Science of Northeastern University. She received her Ph.D. in Computer Science from the University of Illinois at Urbana-Champaign in 2012. Her principal research interest is in mining information and social networks, and more generally in data mining, machine learning, and network science, with a focus on modeling novel problems and proposing scalable algorithms for large-scale, real-world applications. Yizhou has over 60 publications in books, journals, and major conferences. Tutorials on mining heterogeneous information networks have been given in several premier conferences, including EDBT 2009, SIGMOD 2010, KDD 2010, ICDE 2012, VLDB 2012, ASONAM 2012, and ACL 2015. She received 2012 ACM SIGKDD Best Student Paper Award, 2013 ACM SIGKDD Doctoral Dissertation Award, 2013 Yahoo ACE (Academic Career Enhancement) Award, 2015 NSF CAREER Award, and 2016 CS@ILLINOIS Distinguished Educator Award.
Host: Emilio Ferrara
More Info: http://webcastermshd.isi.edu/Mediasite/Play/ce533b93329d4c65bb73697f53ebeb541d
Location: Information Science Institute (ISI) - 11th Flr Conf Rm # 1135, Marina Del Rey
Audiences: Everyone Is Invited
Contact: Peter Zamar
Event Link: http://webcastermshd.isi.edu/Mediasite/Play/ce533b93329d4c65bb73697f53ebeb541d
This event is open to all eligible individuals. USC Viterbi operates all of its activities consistent with the University's Notice of Non-Discrimination. Eligibility is not determined based on race, sex, ethnicity, sexual orientation, or any other prohibited factor. -
AI SEMINAR
Fri, Nov 18, 2016 @ 11:00 AM - 12:00 PM
Information Sciences Institute
Conferences, Lectures, & Seminars
Speaker: Pablo Barberá, School of International Relations at USC
Talk Title: Less is more? How demographic sample weights can improve public opinion estimates based on Twitter data
Abstract: Twitter data is widely acknowledged to hold great promise for the study of political behavior and public opinion. However, a key limitation in previous studies is the lack of information about the sociodemographic characteristics of individual users, which raises concerns about the validity of inferences based on this source of data. This paper addresses this challenge by employing supervised machine learning methods to estimate the age, gender, race, party affiliation, propensity to vote, and income of any Twitter user in the U.S. The training dataset for these classifiers was obtained by matching a large dataset of 1 billion geolocated Twitter messages with voting registration records and estimates of home values across 15 different states, resulting in a sample of nearly 250,000 Twitter users whose sociodemographic traits are known. To illustrate the value of this approach, I offer three applications that use information about the predicted demographic composition of a random sample of 500,000 U.S. Twitter users. First, I explore how attention to politics varies across demographics groups. Then, I apply multilevel regression and postratification methods to recover valid estimate of presidential and candidate approval that can serve as early indicators of public opinion changes and thus complement traditional surveys. Finally, I demonstrate the value of Twitter data to study questions that may suffer from social desirability bias.
Biography: Pablo Barberá joined the School of International Relations at USC as an Assistant Professor in 2016, after receiving his PhD in political science from New York University and spending a year as a Post-Doctoral Fellow at the Center for Data Science in New York University. His research interests include computational methods in the social sciences, automated text analysis, and social network analysis. He applies these methods to the study of social media and politics, comparative electoral behavior and collective action, and political representation. His work has been published in Political Analysis, PLOS ONE, Psychological Science, the Journal of Computer-Mediated Communication, Social Media + Society, and Social Science Computer Review. His current research agenda focuses on the role of social media platforms in the growth of social protests, the measurement of public opinion and political behavior using digital trace data, and how exposure to political violence and governments' counter-messages on social media affects ideological extremism and support for terrorist groups.
Host: Emilio Ferrara
Location: Information Science Institute (ISI) - 11th floor large conference room
Audiences: Everyone Is Invited
Contact: Kary LAU
This event is open to all eligible individuals. USC Viterbi operates all of its activities consistent with the University's Notice of Non-Discrimination. Eligibility is not determined based on race, sex, ethnicity, sexual orientation, or any other prohibited factor. -
NL Seminar-Incremental spoken dialogue system for reference resolution in images
Fri, Nov 18, 2016 @ 03:00 PM - 04:00 PM
Information Sciences Institute
Conferences, Lectures, & Seminars
Speaker: Ramesh R Manuvinakurike , USC/ICT
Talk Title: Incremental spoken dialogue system for reference resolution in images
Series: Natural Language Seminar
Abstract: In this talk, I will be speaking about our ongoing effort in the development of Eve, state-of-the-art incremental reference resolution in images based spoken dialogue agent. Incrementality is central to developing a naturally conversing spoken dialogue systems. Incrementality makes the conversations more natural and efficient compared to non-incremental alternatives. The performance of the Eve was found to be comparable to human performance and she conveniently outperforms alternative non-incremental architectures. However, building such a system is not trivial. It needs high-performance architectures and dialogue components (ASR, dialogue policies, language understanding etc.). I will also speak about future plans for enhancing Eve's capability. I also take a slight deviation and explore a different word level natural language understanding model for reference resolution in images in a dialogue setting.
Biography: Ramesh Manuvinakurike is a Ph.D. student at USC Institute for Creative Technologies working with Prof. David DeVault and Prof. Kallirroi Georgila. He is interested in developing conversational systems and has developed various such systems. His work with his colleagues on agent Eve won 'Best paper' award at Sigdial 2015.
Host: Xing Shi and Kevin Knight
More Info: http://nlg.isi.edu/nl-seminar/
Location: Information Science Institute (ISI) - 11th Flr Conf Rm # 1135, Marina Del Rey
Audiences: Everyone Is Invited
Contact: Peter Zamar
Event Link: http://nlg.isi.edu/nl-seminar/
This event is open to all eligible individuals. USC Viterbi operates all of its activities consistent with the University's Notice of Non-Discrimination. Eligibility is not determined based on race, sex, ethnicity, sexual orientation, or any other prohibited factor.
