Conferences, Lectures, & Seminars
Events for July
-
NL Seminar-Grapheme-to-Phoneme Models for (Almost) Any Language
Fri, Jul 08, 2016 @ 03:00 PM - 04:00 PM
Information Sciences Institute
Conferences, Lectures, & Seminars
Speaker: Aliya Deri, USC/ISI
Talk Title: Grapheme-to-Phoneme Models for (Almost) Any Language
Series: Natural Language Seminar
Abstract: Grapheme-to-phoneme (g2p) models are rarely available in low-resource languages, as the creation of training and evaluation data is expensive and time-consuming. We use Wiktionary to obtain more than 650k word-pronunciation pairs in more than 500 languages. We then develop phoneme and language distance metrics based on phonological and linguistic knowledge; applying those, we adapt g2p models for high-resource languages to create models for related low-resource languages. We provide results for models for 229 adapted languages.
Biography: Aliya Deri is a PhD candidate in Computer Science at USC, advised by Professor Kevin Knight.
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. -
AI SEMINAR
Fri, Jul 15, 2016 @ 11:00 AM - 12:00 PM
Information Sciences Institute
Conferences, Lectures, & Seminars
Speaker: Keith Burghardt,
Talk Title: Understanding the Dynamics of Collective Decisions
Series: AI Seminar
Abstract: In a microscopic setting, humans behave in rich and unexpected ways. In a macroscopic setting, however, distinctive patterns of group behavior emerge, leading researchers to search for an underlying mechanism. The aim of this talk is to analyze the macroscopic patterns of collective decisions in order to discern how group opinions form at the microscopic level. To do so, we first explore the mechanisms that underlie competing ideas using agent-based models (ABMs). We find that simple rules can accurately reconstruct the macroscopic patterns we see in data as diverse as elections and juries. Next, we model collective decision-making in online question and answer boards, and find that a simple individual-level model can capture important features of user behavior, and heuristics appears to predict dynamics with increasing accuracy as the number of answers grow, suggesting information overload undermines collective decisions. Overall, these models reveal two important findings. First, our work begins to untangle how human psychology affects macroscopic behavior: stubbornness, in which users increasingly stick to their latest opinion, is a necessary component in our ABMs to produce the macroscopic patterns we see in many diverse datasets. Furthermore, our models suggest limitations of the wisdom of crowds, in which groups produce better decisions than individuals, in common systems may be undermined by influence and information overload.
Biography: Keith Burghardt is a summer research assistant at ISI under Kristina Lerman studying the collective decisions among users of question and answer boards. He holds a Bachelors of Science in Physics at the University of Maryland, and was recently awarded a Doctorate in Philosophy in Physics from the same university under Michelle Girvan and William Rand. His research has focused on applying statistical mechanics, and other physics-based approaches toward understanding collective social phenomena, including work as diverse as jury decisions, epidemics, and online environments.
No Webcast available
Host: Kristina Lerman
Location: Information Science Institute (ISI) - 6th fl Large CR (Room 689)
Audiences: Everyone Is Invited
Contact: Alma Nava / Information Sciences Institute
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-Commonsense Knowledge Base Completion
Fri, Jul 15, 2016 @ 03:00 PM - 04:00 PM
Information Sciences Institute
Conferences, Lectures, & Seminars
Speaker: Xiang Li, University of Massachusetts Amherst
Talk Title: Commonsense Knowledge Base Completion
Series: Natural Language Seminar
Abstract: We enrich a curated resource of commonsense knowledge by formulating the problem as one of knowledge base completion KBC. Most work in KBC focuses on knowledge bases like Freebase that relate entities drawn from a fixed set. However, the tuples in Concept Net Speer and Havasi define relations between an unbounded set of phrases. We develop neural network models for scoring tuples on arbitrary phrases and evaluate them by their ability to distinguish true held out tuples from false ones. We find strong performance from a bilinear model using a simple additive architecture to model phrases. We manually evaluate our trained models ability to assign quality scores to novel tuples finding that it can propose tuples at the same quality level as medium confidence tuples from Concept Net.
Biography: Xiang Li is asummer intern under the supervision of Prof Kevin Knight and Prof Daniel Marcu. She is also going to be a PhD student at the University of Massachusetts Amherst in Andrew McCallums research group in this coming Fall. She got her BS at the East China Normal University Shanghai China and got her M.S at the University of Chicago. Her research interest mainly focused on natural language processing and machine learning. This work is done when she was in Chicago working with Prof Kevin Gimpel at TTIC Toyota Technological Institute at Chicago.
Host: Xing Shi and Kevin Knight
More Info: http://nlg.isi.edu/nl-seminar/
Location: Information Science Institute (ISI) - 6th Floor -CR # 689; ISI-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. -
NL Seminar-LSTM's for OCR
Fri, Jul 22, 2016 @ 03:00 PM - 04:00 PM
Information Sciences Institute
Conferences, Lectures, & Seminars
Speaker: Stephen Rawls and Huaigu Cao , USC/ISI
Talk Title: LSTM's for OCR
Series: Natural Language Seminar
Abstract: We present ongoing research into OCR for both machine print and handwriting recognition. We utilize a neural network along with LSTM's to perform OCR directly from pixel intensity. We are exploring a few novel improvements, including using a CNN for feature extraction prior to the LSTM, and combining reinforcement learning into our training to directly optimize word error rate in our test-time decoding procedure, which utilizes a (non-differentiable) language-model based decoding of the LSTM output. Finally, we present the design of the OCR system we used to win a pilot project with the US Census for recognizing handwritten first and last names.
Biography: Stephen Rawls is a research programmer and a PhD student at USC/ISI advised by Dr. Prem Natarajan. He works in the Computer Vision group at ISI on face recognition and OCR, among other projects
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. -
NL Seminar- Let's not be clever: simple pre- and post-processing tricks in machine translation
Fri, Jul 29, 2016 @ 03:00 AM - 04:00 PM
Information Sciences Institute
Conferences, Lectures, & Seminars
Speaker: Sebastian Mielke, USC/ISI Summer Intern
Talk Title: Let's not be clever: simple pre- and post-processing tricks in machine translation
Series: Natural Language Seminar
Abstract: Today's machine translation system are highly complex and extending them often means leaving highly sophisticated solutions and established algorithms behind. Therefore it is attractive to try to extend the process outside of the translation system: in pre- and post-processing steps. I will show a pre-processing step for helping to translate tweets and a post-processing step that helps "guess" the translations of unknown and thus untranslated words in arbitrary sentences using dictionaries and other resources.
Biography: Sebastian is currently pursuing a CS masters degree in Dresden, Germany with Prof. Heiko Vogler, taking a break from studying to work on low-resource machine translation with Prof. Kevin Knight and Prof. Daniel Marcu as an ISI intern in 2016.
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.
