특별 프로그램

특별 프로그램
초청강연
  • 12월 17일 11:30 ~ 12:00 (30분)

    제목: 인공지능 서비스를 위한 ETRI X + (AI) 실행전략
    연사: ETRI 지능화융합연구소 박종현 소장
    개요: AI 시대로의 패러다임 전환을 위해 작년 12월, 과학기술정보통신부를 비롯한 전 부처가 합동하여 「인공지능(AI) 국가전략」을 발표했다. AI 서비스가 확산되면서 산업 생태계, 일자리 구조 등 모든 방면의 변화가 발생할 것이다. AI 서비스 실현을 위한 출연연의 역할은 무엇일까? 국가지능화종합연구기관으로서 AI 서비스 실현을 위한 ETRI 인공지능 실행전략을 개괄적으로 소개하고자 한다. ETRI AI 실행전략의 목표는 대한민국 지능화 실현을 위한 인공지능 혁신플랫폼을 구축하는 것이다. 이를 위한 3대 전략목표와 7대 실행전략을 구체화하였다. ETRI AI 실행전략은 우리나라 AI가 글로벌 경쟁력을 확보하기 위해 ETRI가 그간 개발해온 ICT 및 AI 기술을 바탕으로 어떻게 R&D 기술 생태계를 이끌 것인가에 대한 새로운 포지셔닝 전략을 담은 것이라고 볼 수 있다. 이를 통해 AI 기술이 산업에서 가치를 창출하고 국민의 문제를 해결하기 위해 기술개발 초기 단계부터 생태계 구성원과 훨씬 더 깊게 교류하며 함께 성장할 수 있는 기회마련의 계기가 될 것으로 기대해본다.

튜토리얼
  • 12월 17일 10:20 ~ 11:10 (50분)

    제목: Python-based Machine Learning Algorithmic Thinking
    연사: 국민대학교 김상철 교수
    개요: In this tutorial, the installation procedure of Pycham, Anaconda, and TensorFlow for machine learning will be introduced and debugging technology using Pycharm IDE will be addressed through data visualization examples. Numerical algorithms for machine learning are explained and programmed by the Pycharm debugging process. After looking at the strategy of Curve-Fitting, we look at supervised learning of machine learning through TensorFlow debugging. We will understand the algorithm of gradient descent for analyzing the loss function and explain multivariable regression analysis using the scikit-learn module.

  • 12월 17일 13:30 ~ 14:20 (50분)

    제목: Recent Advances in Deep Image Anomaly Detection
    연사: 경북대학교 김재일 교수
    개요: Novelty detection is the task of identifying whether test data is an outlier from the training data in some aspects. Approaches for the novelty detection exploit explicit representation of the distribution of the training data (i.e., positive samples) to determine outliers in feature space. Unseen samples are compared with the models of normality, which can draw the decision boundary of the training samples. In medical imaging, the novelty detection has gained much attention, due to the lack of sufficient dataset for diseases and the limitations in annotating medical images for supervised learning. Recently, deep auto-encoder methods have been proposed to localize abnormal regions, such as multiple sclerosis and brain tumor, in brain MR images. Weakly-supervised anomaly detection methods using epistemic uncertainty based on Bayesian deep networks also showed better performance in characterizing anomalies under several disease conditions in medical images. In this tutorial, we will review recent advances in the novelty detection by deep learning methods and introduce their challenges.

Top Conference 초청 섹션
  • 12월 16일 15:50 ~ 16:20 (30분)

    제목: Bootstrapping neural processes (NeurIPS 2020)
    연사: KAIST 이주호 교수
    개요: Unlike in the traditional statistical modeling for which a user typically hand-specify a prior, Neural Processes (NPs) implicitly define a broad class of stochastic processes with neural networks. Given a data stream, NP learns a stochastic process that best describes the data. While this “data-driven” way of learning stochastic processes has proven to handle various types of data, NPs still rely on an assumption that uncertainty in stochastic processes is modeled by a single latent variable, which potentially limits the flexibility. To this end, we propose the Boostrapping Neural Process (BNP), a novel extension of the NP family using the bootstrap. The bootstrap is a classical data-driven technique for estimating uncertainty, which allows BNP to learn the stochasticity in NPs without assuming a particular form. We demonstrate the efficacy of BNP on various types of data and its robustness in the presence of model-data mismatch.

  • 12월 16일 16:20 ~ 16:50 (30분)

    제목: Monte Carlo Tree Search in continuous spaces using Voronoi optimistic optimization with regret bounds (AAAI 2020)
    연사: KAIST 김범준 교수
    개요: Many important applications, including robotics, data-center management, and process control, require planning action sequences in domains with continuous state and action spaces and discontinuous objective functions. Monte Carlo tree search (MCTS) is an effective strategy for planning in discrete action spaces. We provide a novel MCTS algorithm (VOOT) for deterministic environments with continuous action spaces, which, in turn, is based on a novel black-box function-optimization algorithm (VOO) to efficiently sample actions. The VOO algorithm uses Voronoi partitioning to guide sampling, and is particularly efficient in high-dimensional spaces. The VOOT algorithm has an instance of VOO at each node in the tree. We provide regret bounds for both algorithms and demonstrate their empirical effectiveness in several high-dimensional problems including two difficult robotics planning problems.

  • 12월 17일(목요일) 14:30 ~ 15:00 (30분)

    제목: Randomized Adversarial Imitation Learning for Autonomous Driving (IJCAI 2019)
    연사: 고려대학교 김중헌 교수
    개요: With the evolution of various advanced driver assistance system (ADAS) platforms, the design of autonomous driving system is becoming more complex and safety-critical. The autonomous driving system simultaneously activates multiple ADAS functions; and thus it is essential to coordinate various ADAS functions. This paper proposes a randomized adversarial imitation learning (RAIL) method that imitates the coordination of autonomous vehicle equipped with advanced sensors. The RAIL policies are trained through derivative-free optimization for the decision maker that coordinates the proper ADAS functions, e.g., smart cruise control and lane keeping system. Especially, the proposed method is also able to deal with the LIDAR data and makes decisions in complex multi-lane highways and multi-agent environments.