Cross-domain few-shot learning (CD-FSL) has drawn increasing attention for handling large differences between the source and target domains-an important concern in real-world scenarios. To overcome these large differences, recent works have considered exploiting small-scale unlabeled data from the target domain during the pre-training stage. This data enables self-supervised pre-training on the target domain, in addition to supervised pre-training on the source domain. In this paper, we empirically investigate which pre-training is preferred based on domain similarity and few-shot difficulty of the target domain. We discover that the performance gain of self-supervised pre-training over supervised pre-training becomes large when the target domain is dissimilar to the source domain, or the target domain itself has low few-shot difficulty. We further design two pre-training schemes, mixed-supervised and two-stage learning, that improve performance. In this light, we present six findings for CD-FSL, which are supported by extensive experiments and analyses on three source and eight target benchmark datasets with varying levels of domain similarity and few-shot difficulty.

For many of the state-of-the-art computer vision algorithms, image segmentation is an important preprocessing step. As such, several image segmentation algorithms have been proposed, however, with certain reservation due to high computational load and many hand-tuning parameters. Correlation clustering, a graphpartitioning algorithm often used in natural language processing and document clustering, has the potential to perform better than previously proposed image segmentation algorithms. We improve the basic correlation clustering formulation by taking into account higher-order cluster relationships. This improves clustering in the presence of local boundary ambiguities. We first apply the pairwise correlation clustering to image segmentation over a pairwise superpixel graph and then develop higher-order correlation clustering over a hypergraph that considers higher-order relations among superpixels. Fast inference is possible by linear programming relaxation, and also effective parameter learning framework by structured support vector machine is possible. Experimental results on various datasets show that the proposed higher-order correlation clustering outperforms other state-of-the-art image segmentation algorithms.

A Korean research team developed a technology for detecting humans lying on the road in real-time. Therefore, preventing safety accidents in the city and rapidly responding to it will be possible, and a safer society is expected to be accomplished. The Electronics and Telecommunications Research Institute (ETRI) announced that it applied the technology of Visual AI ‘DeepView’ to Daejeon Metropolitan City in earnest to prevent safety accidents in the city and promptly respond to them. DeepView is an AI technology recognizing human behavior. It detects people lying on the road through surveillance cameras in real-time. As it can be applied to preventing safety accidents caused by drinking, fainting, etc. and performing prompt emergency rescue measures. It is expected to become the core technology for making a safe city.

From commerce, finance and agriculture to self-driving cars, personalised healthcare and social media – advancements in artificial intelligence (AI) unlock countless opportunities. New applications promise to improve the quality of people's lives throughout the world, but at the same time, raise a number of societal questions. A joint panel discussion of the German National Academy of Sciences Leopoldina and the Korean Academy of Science and Technology (KAST) explores AI technologies, their benefits and their challenges for society. Virtual panel discussion of the German National Academy of Sciences Leopoldina and the Korean Academy of Science and Technology „Realizing the Promises of Artificial Intelligence" Thursday, 25 November 2021, 8am to 9am (CET) Online Following opening remarks from the President of the Leopoldina, Prof (ETHZ) Dr Gerald Haug and Prof Min-Koo Han, PhD, President of the KAST, legal scholar Prof Ryan Song, PhD, Kyung Hee University, Seoul/South Korea, will provide an introduction into the topic. Subsequently, computer scientist Prof Alice Oh PhD, KAIST School of Computing, Daejeon/ South Korea, and Member of the Leopoldina Prof Dr Alexander Waibel, Karlsruhe Institute of Technology/Germany and Carnegie Mellon University, Pittsburgh/USA, will provide input statements for further discussion.

This undated image, provided by KT Corp., shows its logo. KT Corp., a major South Korean telecom operator, said Sunday it signed an agreement with the country's top science and technology university to establish a research institute for artificial intelligence (AI) and software development. Under the agreement, KT will work with the Korea Advanced Institute of Science and Technology (KAIST) to build the research institute in the central city of Daejeon, 164 kilometers south of Seoul, by the end of this year. KT said the institute will house around 200 KAIST researchers, faculty members and KT employees to develop future technologies, including an AI model that can recognize complex situations based on voice and video recognition. The two will also conduct joint research in developing AI for industrial settings, such as in media, health care and robotics.

SK Telecom said on Thursday it has launched its 5G edge cloud service, called SKT 5GX Edge, embedded with Amazon Web Services (AWS) Wavelength in South Korea. The launch of the service will allow customers to build ultra-low latency mobile apps, the telco said, in areas such as machine learning, Internet of Things, gaming, and streaming. Use of the service will allow apps that are accessing the cloud to bypass the internet and regional websites, and quickly reach SK Telecom's data centre. The reduced step will allow customers to enjoy the full benefits offered by 5G network's low latency and bandwidth, the telco said. The first AWS Wavelength Zone has been launched in the city of Daejeon. It will expand to Seoul and other regions next year.

With its COVID-19 outbreak seemingly contained, South Korea may offer the rest of the world a glimpse of what society could look like after the pandemic ends -- and it may include robotic baristas. According to Reuters, a cafe in Daejeon, South Korea, is now using robots to prepare drinks and deliver them to customers. Proponents say the robots could encourage "distancing in daily life." The barista system consists of a robotic arm that prepares 60 different beverages and wheeled bots that deliver the drinks to customers. The system can communicate with other devices, contains self-driving tech to determine the best route around people and tables and communicates with customers via voice controls.

A smart factory in South Korea has developed an artificial intelligence (AI) solution to help people stay socially distanced and keep the coronavirus contained as the country reopens. A cafe in Daejeon, South Korea is using a robot barista to serve people lattes and make it easier for consumers to stay the recommended six feet apart, according to a report in Reuters on Monday (May 25). The robot was developed in collaboration with smart factory Vision Semicon and a state-run science institute, according to Lee Dong-bae, director of research at Vision Semicon. "Our system needs no input from people from order to delivery, and tables were sparsely arranged to ensure smooth movements of the robots, which fits well with the current'untact' and distancing campaign," he said. The robot can make 60 different types of coffee and serves orders to consumers at their tables.

The new robot barista at the cafe in Daejeon, South Korea, is courteous and swift as it seamlessly makes its way toward customers. "Here is your Rooibos almond tea latte, please enjoy. It's even better if you stir it," it says, as a customer reaches for her drink on a tray installed within the large, gleaming white capsule-shaped computer. After managing to contain an outbreak of the new coronavirus that infected more than 11,000 people and killed 267, South Korea is slowly transitioning from intensive social distancing rules toward what the government calls "distancing in daily life." Robots could help people observe social distancing in public, said Lee Dong-bae, director of research at Vision Semicon, a smart factory solution provider that developed the barista robot together with a state-run science institute.

ADDITIONAL SHARED DECODER ON SIAMESE MUL TI-VIEW ENCODERS FOR LEARNING ACOUSTIC WORD EMBEDDINGS Myunghun Jung, Hyungjun Lim, Jahyun Goo, Y oungmoon Jung, and Hoirin Kim School of Electrical Engineering, KAIST, Daejeon, Republic of Korea ABSTRACT Acoustic word embeddings -- fixed-dimensional vector representations of arbitrary-length words -- have attracted increasing interest in query-by-example spoken term detection. Recently, on the fact that the orthography of text labels partly reflects the phonetic similarity between the words' pronunciation, a multi-view approach has been introduced that jointly learns acoustic and text embeddings. It showed that it is possible to learn discriminative embeddings by designing the objective which takes text labels as well as word segments. In this paper, we propose a network architecture that expands the multi-view approach by combining the Siamese multi-view encoders with a shared decoder network to maximize the effect of the relationship between acoustic and text em-beddings in embedding space. Discriminatively trained with multi-view triplet loss and decoding loss, our proposed approach achieves better performance on acoustic word discrimination task with the WSJ dataset, resulting in 11.1% relative improvement in average precision. Index T erms -- acoustic word embedding, query-by- example spoken term detection, multi-view learning, Siamese network, encoder-decoder 1. INTRODUCTION Query-by-example spoken term detection (QbE-STD) is the task of retrieving a spoken query from a set of speech utterances. Amazon Echo, Google Home, Apple Siri), the QbE-STD has drawn interest as a technique that can be applied to wake-up or command word detection, search engine, etc.