동국대학교 인공지능학과

커뮤니티

공지사항

인공지능학과 해외연사 초청 온라인 세미나 안내

페이지 정보

작성자 관리자 조회80회 작성일 20-09-14 10:21

본문


2020 2
학기

동국대학교 일반대학원 인공지능학과 해외연사 초청 온라인 세미나 일정


 9/16() 10:00

 

 - 연사 : Prof. Kenneth D. Forbus

     https://www.mccormick.northwestern.edu/research-faculty/directory/profiles/forbus-ken.html

 - 연사 소속 : Northwestern University

 - 주제 : Qualitative and analogical reasoning for human-centered artificial intelligence

 - 내용 : Human-centered artificial intelligence requires creating systems that interact with people on their terms, and learn in similar ways. This talk will argue that qualitative representations and analogical reasoning are important tools for human-centered AI. Qualitative representations provide natural representations for continuous phenomena, providing a bridge between perception and cognition. Analogy provides a reasoning method that operates with sparse data and performs data-efficient learning, instead of requiring massive amounts of data. These ideas will be illustrated through three lines of research: (1) recognizing human behaviors from 3D camera data, (2) moral decision-making, and (3) theory of mind reasoning.
 

 

 - Webex 정보

 - 미팅 링크 : https://dongguk.webex.com/dongguk/j.php?MTID=m381e9df570763d6d5c94e068f7f2bd0b

 - 미팅 번호 : 170 700 7250

 - 비밀번호 : wtGGTtRm863

 - 호스트 키 : 112412
 

 9/25() 10:00

 

 - 연사 : 윤승현 박사

     https://david-yoon.github.io/

 - 연사 소속 : Adobe Research

 - 주제 : Recent Advancements in NLP for Question Answering System, Language Model,

           and Evaluation Metric

 - 내용 : Question answering system has long been considered a primary objective of artificial

intelligence. The advancement of the QA system has attracted huge interests from the academic and industry community these days. In this talk, I'll first present the recent development of natural language processing technologies, including question answering (QA). Despite the rapid advances in the QA system, the method for understanding long texts calls for further research. I'll introduce techniques for dealing with long text; hierarchical architecture, auxiliary topic-clustering technique, and other latest approaches.

I'll then present recent language models that provide textual representations and show significant performance improvement in many NLP tasks, including the QA system. Finally, I'll discuss challenges for developing automatic evaluation metrics for natural language generation tasks such as generative QA, dialogue model, and multimodal captioning.

 - 장소 : 동국대학교 원흥관 3I.SPACE (비대면일 경우 Webex로 진행)
 

 

 - Webex 정보

 - 미팅 링크 : https://dongguk.webex.com/dongguk/j.php?MTID=m3848412150fdf38349dc736a95846645

 - 미팅 번호 : 170 832 2615

 - 비밀번호 : umJPD7bSc78

 - 호스트 키 : 257530
 

10/14() 10:00 

 

 - 연사 : Prof. Yolanda Gil

     https://viterbi.usc.edu/directory/faculty/Gil/Yolanda

 - 연사 소속 : University of Southern California

 - 주제 : Core Knowledge Technologies for Publishing the Scientific Record on the Web

 - 내용 : Are we far from a day when each scientific article will be properly linked to hypotheses, models, software, provenance, workflows, and other key scientific entities on the Web? This talk will describe our work on seven ontologies that we have developed to describe complementary aspects of scientific work, and that interlinked together present a path towards publishing the scientific record on the Web. The Linked Earth Ontology extends existing standards, and was developed collaboratively and entirely online by scientists. The OntoSoft ontology describes scientific software artifacts with information relevant to scientists. The W3C PROV-O ontology represents provenance of scientific data, whether observable or derived through computation. The P-PLAN ontology extends PROV-O to describe high-level general plans, and the OPMW-PROV ontology extends both to describe abstract computational workflows linked to their executions. The DISK Hypothesis ontology describes hypothesis statements, their supporting evidence, and their revisions as new data is analyzed. The Software Description Ontology for Models characterizes the development of models so they can be understood and compared. These seven ontologies provide essential capabilities, but much work remains to be done to capture more comprehensively the scientific record. Our vision is that AI research tools will then be able to access this information to reason about the state of the art and ultimately to generate new findings, and fundamentally change how we do science.
 

 

 - Webex 정보

 - 미팅 링크 : https://dongguk.webex.com/dongguk/j.php?MTID=m25cf59064c2a180ccaecb8fb8b48d66f

 - 미팅 번호 : 170 393 7440

 - 비밀번호 : eaPDqUPr624

 - 호스트 키 : 898019
 

 10/30() 13:30

 

 - 연사 : 최승진 교수

     https://baroaiacademy.app/about

 - 연사 소속 : CTO, 바로AI

 - 주제 : Learning (How) to Learn with Gradients

 - 내용 : Learning-to-learn implies inferring a learning strategy from a set of relevant tasks

(corresponding to a set of past experiences) via a meta-learner that a task-specific learner can leverage when learning a new task. Learning-to-learn, also known as meta-learning, jointly learns features, a model, and an algorithm from data, while the standard machine learning or deep learning jointly learns features and a model. Recently meta-learning has emerged as a promising approach to tackle few-shot problems when only a handful of labeled examples are provided. In this talk, I begin with why I am interested in meta-learning, to emphasize its importance. After reviewing a few examples where meta-learning is useful, I introduce a few methods. Of particular interests are meta-learning methods with gradients. With a quick introduction of model-agnostic meta-learning, I will talk about a few gradient-based meta-learning algorithms, including Reptile, iMAML, and my own work (MT-net).

 - 장소 : 미정 (비대면일 경우 Webex로 진행)
 

 

 - Webex 정보

 - 미팅 링크 : https://dongguk.webex.com/dongguk/j.php?MTID=macc3cb84c9c199f886b098f81c44ce66

 - 미팅 번호 : 170 845 1867

 - 비밀번호 : uPPPqq26XJ3

 - 호스트 키 : 173767



 

 11/6() 16:00

 

 - 연사 : Prof. Hyung Jin Chang

    https://hyungjinchang.wordpress.com/

 - 연사 소속 : University of Birmingham

 - 주제 : Human-centred Visual Learning

 - 내용 : One of the ultimate goals of AI research is to make humans’ lives better by creating intelligent systems that are like or better than people. In that sense, humans are the best teachers as well as beneficiaries for AI development. Human-centred visual learning is an approach to develop vision-based algorithms that aim to make systems usable and useful by focusing on humans, especially their needs and requirements. In this talk, I will introduce my recent research on human-centred vision tasks including human attention mimicking visual object tracking, real-time hand pose tracking, and human gaze estimation. In addition, the visual learning of articulated kinematic structures and their uses for higher-level applications such as kinematic structure correspondence matches and personalised assistance will be presented. 



※ 인공지능학과 학생들은 특별한 일이 없다면 참석 바랍니다.

   

 

 


첨부파일

04620 서울시 중구 필동로 1길 30 동국대학교 원흥관 F314호 전화: 02-2290-1408
Copyright(c) 2016 DONGGUK UNIVERSITY. ALL RIGHTS RESERVED.