Oral 1

​좌장: 김현우 교수(고려대)

Doctoral Consortium

​좌장: 권준석 교수(중앙대)

Oral 2

​좌장: 김원준 교수(건국대)

Industry 1

​좌장: 곽수하 교수(POSTECH)

Invited Talk 1

좌장: 이경무 교수(서울대)

Learning to See by Looking at Noise

Abstract: 

The importance of data in modern computer vision is hard to overstate. The ImageNet dataset, with its millions of labelled images, is widely thought to have spurred the era of deep learning, and since then the scale of vision datasets has been increasing at a rapid pace. These datasets come with costs: curation is expensive, and they inherit human biases. To counter these costs, interest has surged in learning with unlabeled images as it avoids the curation efforts, or using simulated environments, but content creation is also labor intensive. In this talk I will describe our work in trying to reduce the need for data. We will start by trying to get rid of the annotation effort an explore several self-supervised tasks using multimodal data such as auditory and tactile information. Finally, we will go a step further and ask if we can do away with real image datasets entirely, instead learning from noise processes. Noise processes produce images that are reminiscent of abstract art, where images contain textures and shapes, but there are no recognizable objects. Our findings show that good performance on real images can be achieved even with training images that are far from realistic.

Invited Talk 2

​좌장: 권인소 교수(KAIST)

Perception and Action

Abstract:

I believe that the primary challenge for creating AI is to first master the link between sensory perception and motor control which, over the course of biological evolution, has provided the substrate for the development of capabilities such as language and abstract thought. Recently, we have seen major progress in core computer vision problems such as recognition and reconstruction, and the time is ripe to bootstrap these to advance core robotics problems such as locomotion, navigation, and manipulation. In my talk I will present various results along these directions.

Oral 3

​좌장: 김광인 교수(UNIST)

Industry 2

​좌장: 최종현 교수(연세대)

Oral 4

​좌장: 조민수 교수(POSTECH)

Oral 5

​​좌장: 황원준 교수(아주대) 

Invited Talk 3

​좌장: 손광훈 교수(연세대)

Robust Deep Networks through Invariance and Adaptation

Abstract: 

While we have seen immense progress in machine learning, a critical shortcoming of current methods lies in handling distribution shift between training and deployment. Distribution shift is pervasive in real-world problems ranging from natural variation in the distribution over locations or domains, to shift in the distribution arising from different decision making policies, to shifts over time as the world changes. In this talk, I'll start by discussing our efforts in benchmarking machine learning methods under multiple natural occurrences of distribution shift, including domain shift, subpopulation shift, and gradual shift over time. Then, I will talk about two new algorithms that can mitigate certain forms of covariate shift. The first leverages domain information to learn domain invariant functions, without using any explicit regularizers. This leads to significant and consistent gains on a variety of natural distribution shift problems. The second instead leverages unlabeled target distribution data to learn a diverse set of functions. In doing so, it is able to address major limitations of prior robustness works: it doesn’t require labeled data from the test distribution to tune hyperparameters, and it can handle an extreme version of spurious correlations where there is a perfect correlation between the spurious attribute and label. I'll conclude by discussing important open questions for future work. 

Oral 6

​좌장: 임성훈 교수(DGIST)

Industry 3

​좌장: 김태현 교수(한양대)

Oral 7

​좌장: 심현정 교수(KAIST)