Artificial Intelligence and Machine Learning
[Prof. Kee-Eung Kim] The ultimate goal of artificial intelligence (AI), which is in essence building intelligent systems, requires computational and mathematical frameworks for intelligent behaviors. At the core of these frameworks lies the principle of rationality, and the decision theory provides a classical but effective tool for building intelligent systems as well as understanding the behavior of humans and animals. My research group at KAIST has been devoted to designing and developing decision-theoretic representations and algorithms for AI. Research Results My research group is working on representations and algorithms for decision theoretic planning problems, including large scale Markov decision processes (MDPs) and partially observable MDPs (POMDPs). Decision theory is one of the most important approaches to understanding and implementing the rational and intelligent behavior. 1. MDPs and POMDPs The classical MDP/POMDP representation has been a useful tool for modeling intelligent behaviors for decades. However, we often encounter the need to extend the standard representation to effectively capture real-world problems. For example, how can we naturally specify the constraints on the properties of desired solutions, or handle the uncertainty in the parameters of the model? As we extend the standard model, how can we address the computational challenges in designing efficient algorithms? We have addressed some of these theoretical issues and presented at top-tier AI conferences. 2. Machine learning of behavioral data The standard way of machine learning for the intelligent behavior is the reinforcement learning (RL). However, the standard setting in RL algorithms assumes prescribed reward functions, which is not an easy task to specify them in practice. Inferring the reward function from the behavior data is referred to as the inverse reinforcement learning (IRL), and its significance has emerged from the connection among RL and other disciplines such as neurophysiology, behavioral neuroscience, and economics. IRL is an important problem for understanding human and animal behaviors, and we have presented our algorithms at a top-tier machine learning conference and journal. 3. Applications Besides theoretical research on representations and algorithms, our group also worked on applications to demonstrate the usefulness of the decision-theoretic AI approach. Our work on applications include spoken dialogue systems and brain-computer interface systems. Awards 1. 2nd place in the POMDP track of the International Probabilistic Planning Competition (IPPC), 2011 2. Best poster award at the Pacific-Rim Conference on Artificial Intelligence (PRICAI), 2010 References Published papers 1. Jaedeug Choi and Kee-Eung Kim,“MAP Inference for Bayesian Inverse Reinforcement Learning”In: Proceedings of Neural Information Processing Systems (NIPS 2011). [Accepted] 2. Jaeyoung Park, Kee-Eung Kim, and Yoon-Kyu Song,“A POMDP-based Optimal Control of P300-based Brain-Computer Interfaces”In: Proceedings of the AAAI Conference on Artificial Intelligence (AAAI) NECTAR Track. 2011. [8th Korean AAAI paper] 3. Dongho Kim, Jaesong Lee, Kee-Eung Kim, and Pascal Poupart,“Point-Based Value Iteration for Constrained POMDPs”In: Proceedings of the International Joint Conference on Artificial Intelligence (IJCAI). 2011. [5th Korean IJCAI paper] 4. Eunsoo Oh and Kee-Eung Kim,“A Geometric Traversal Algorithm for Reward-Uncertain MDPs”In: Proceedings of the Conference on Uncertainty in Artificial Intelligence (UAI). 2011. [1st Korean UAI paper] 5. Pascal Poupart, Kee-Eung Kim, and Dongho Kim,“Closing the Gap: Towards Provably Optimal POMDP Solutions”In: Proceedings of the International Conference on Automated Planning and Scheduling (ICAPS). 2011. [1st Korean ICAPS paper] 6. Jaedeug Choi and Kee-Eung Kim,“Inverse Reinforcement Learning in Partially Observable Environments”Journal of Machine Learning Research (JMLR), 12. 2011. [2nd Korean JMLR paper] 7. Dongho Kim, Jin Hyung Kim, and Kee-Eung Kim,“Robust Performance Evaluation of POMDP-Based Dialogue Systems”IEEE Transactions on Audio, Speech, and Language Processing (TASLP), 19(4). 2011. 8. Youngwook Kim and Kee-Eung Kim,“Point-Based Bounded Policy Iteration for Decentralized POMDPs”In: Proceedings of Pacific-Rim Conference on Artificial Intelligence (PRICAI) / Lecture Notes in Computer Science (LNCS) 6230. 2010. [Best poster award] 9. Jaeyoung Park, Kee-Eung Kim, and Sungho Jo,“A POMDP Approach to P300-Based Brain-Computer Interfaces”In: Proceedings of the ACM International Conference on Intelligent User Interfaces (IUI). 2010. 10. U.S. Patent Application 20110152710,“Adaptive Brain-Computer Interface Device”...Read more
Developing a hyperspectral 3D imaging system and s..
[Prof. Min H. Kim] 3D imaging techniques have been broadly used in manufacturing, entertainment and military industries. However, current 3D imaging systems have been limited to capturing and representing only trichromatic 3D objects. This research project extends the spectral dimension of 3D imaging techniques beyond the trichromatic spectrum. It is the first approach to build a complete 3D scanning system that measures the hyperspectral reflectance of solid objects. This research project includes the design and building of a 3D imaging system, the development of 3D imaging algorithms, and several 3D software applications to visualize such hyperspectral 3D image data. The research outcome of this project could be broadly adapted to physically meaningful measurements of hyperspectral material appearance of 3D solid objects in natural science and bio-medical engineering. Schematic overview of the 3D imaging spectroscopy system (ACM Trans. on Graphics, 31(4), 38:1-11, SIGGRAPH 2012) Research Funding ㆍ This work was supported by the National Research Foundation (NRF) of Korea, Samsung Electronics and Microsoft Research Asia additionally. Research Results ㆍ This imaging system was introduced for the first time in ACM Transactions on Graphics, 2012, the top journal in computer graphics. The associated visualization software application was published in ACM Journal on Computing and Cultural Heritage, 2014 such that this work was received a best paper award in 2012 at the International Symposium on Virtual Reality, Archaeology and Cultural Heritage (VAST 2012). References ㆍ M. H. Kim, T. A. Harvey ,D. S. Kittle, H. Rushmeier, J. Dorsey, R. O. Prum, D. J. Brady (2012), "3D Imaging Spectroscopy for Measuring Hyperspectral Patterns on Solid Objects," ACM Transactions on Graphics (Proc. SIGGRAPH 2012), 31(4), July 2012, pp. 38:1-11 (IF=3.361) ㆍ M. H. Kim, H. Rushmeier, J. ffrench, I. Passeri, D. Tidmarsh (2014), "Hyper3D: 3D Graphics Software for Examining Cultural Artifacts," ACM Journal on Computing and Cultural Heritage, 7(3), February, pp. 1:1-19...Read more
Analysis on Media Characteristics of Twitter
[Prof. Sue Bok Moon] Via quantitative analysis with the not-sampled, but complete data, we show media-like characteristics of Twitter. Online communication has emerged as a new form of media, and our work is one of the first to demonstrate a quantitative approach for new media research. Research Results Online Social Network (OSN) services are creating a sea of new products and services based on the social networks. OSN services, such as Facebook and Twitter, threaten the established ones, such as Microsoft, Google and Apple. Twitter, a microblogging service, stands out from other OSN services in that its relations are directional and most user profiles and tweets are public. For three months from June 2009, we collected profiles of all Twitter users. Our analysis shows that Twitter has characteristics of not only a social network but also of media. The reciprocity typically over 80％ in social networks is low (less than 50％). Most tweets are of news nature. Retweets expand readership by 100 or more easily. ※Excellency and Expected effects No further explanation is necessary for the influential power of Twitter, as demonstrated in the recent Seoul mayor election. However, in 2009 when we started our research on Twitter, domestic awareness was low and nobody could have imagined the explosive power in the MENA (Middle East North Africa) situations this spring. Our paper, “What is Twitter, a Social Network or a News Media?” published in the World-Wide Web (WWW) conference on April 2010, investigates social phenomena via Twitter and provides statistical approaches on the complete data of Twitter. Our data crawling and analysis methodologies have become a de-facto standard in follow-up research. The Library of Congress in US has decided to archive the entire Twitter data, thus acknowledging the historic significance of massive-scale communication records among hundreds of million users for the first time in human history. Twitter provides an epoch-making data that allows people to collect information and observe its diffusion in an unprecedentedly massive scale and is changing the face of research methodologies in computer science, sociology, politics, business administration, cognitive science and more. Our paper, published in the WWW conference in 2010, is cited more than 2000 times (via Google Scholar) as of July 2014 and we expect the number to increase even more in the coming years. This work has been covered by domestic and foreign press such as ChosunIlbo, HankookIlbo, MIT Technology Review Blog, ReadWriteWeb, the Observer, PC World, Mashable Op-Ed. Computer science graduate courses at Georgia Institute of Technology, UC Santa Barbara and Cornell have included our paper in their lecture material. Non-computer science departments, such as MIT Business School, Bowdoin College sociology department, and the University of California at San Diego Visual Arts Department have also included our work. From 2009 I have given more than 10 invited talks on this paper, including Boston Univ., Northwestern Univ., UC Santa Barbara, Microsoft Research Bangalore, Microsoft Research Redmond, Duke Univ., North Carolina State Univ. This work provides data collection and analysis methodologies for new media studies and falls to the category of fundamental science. It is hard to file for a patent or apply immediate productization from this work but this work reveals basic characteristics of new media, of which knowledge is imperative to designing derivative products or services. Our investigation of human society via online service provides a new direction for interdisciplinary research between computer science, humanities, and social sciences. KAIST established the Web Science and Technology division under the World-Class University program sponsored by the Ministry of Education, Science and Technology, for this kind of interdisciplinary research. In Korea with its world-best high-speed Internet infrastructure and less-than-1％ illiteracy and highly educated work-force, interdisciplinary research could be the guiding light to take our IT industry to the next level. We hope our social network of interdisciplinary collaborators initiated by this Twitter research continues to expand and become a catalyst for explosive growth in our academic excellence. Reference Published papers 1. Haewoon Kwak, Changhyun Lee, Hyunwoo Chun, Sue Moon,“What is Twitter, a Social Network or a News Media?” Proceedings of the World-Wide Web, April 2010, Raleigh, North Carolina.(Acceptance rate: 14％ , Citation: 313) Funding Sources 1. Collect, Analyze and Share for Future Internet : High-Precision Measurement and Analysis Research, 2008.3.1.∼2013.2.28., MKE...Read more
Scalable Big Graph Mining
[Prof. U Kang] Scalable big graph mining using distributed systems opens new opportunities for the discovery of interesting patterns and anomalies on very large graphs which could not be analyzed before. Article: Graphs are ubiquitous: computer networks, social networks, mobile call networks, biological networks, citation networks, protein regulation networks, and the World Wide Web, to name a few. Spurred by the lower cost of disk storage, the success of social networking sites (e.g. Facebook, Twitter, and Google＋) and Web 2.0 applications, and the high availability of data sources, graph data are being generated at an unparalleled rate. They are now measured in terabytes and heading toward petabytes, with more than billions of nodes and edges. Mining such big graphs helps us find patterns and anomalies which lead to many interesting applications including fraud detection, cyber security, social network analysis, etc. The research team at Prof. U Kang in Department of Computer Science is working on scalable big graph mining which includes two components: scalable algorithms and discoveries on real world graphs. Scalable Algorithms. Traditional graph algorithms assume the input graph fits in the memory or disks of a single machine. However, the recent growth of the sizes in graphs violates this assumption. Since single machine algorithms are not tractable for handling big graphs, the research team at Prof. U Kang is working on designing and developing scalable algorithms and distributed platforms for mining and managing big graphs. Prof. U Kang is the main author of the award-winning Pegasus graph mining software which includes various large scale graph mining algorithms including PageRank, Random Walk with Restart (RWR), diameter estimation, connected components, eigensolver, and tensor analysis. Discoveries on Real World Graphs. The developed scalable algorithms lead to the analysis of large real world graphs, and interesting discoveries of patterns and anomalies which could not be found before. One of the most interesting discoveries is the seven-degrees of separation in one of the largest publicly available Web graphs with ∼7 billion edges. The discovery suggests that the so-called ＇small world' phenomenon exists in the Web, and the distance between any two Web pages is much smaller than people's expectations. Another interesting discovery is the existence of suspicious adult advertisers in the Twitter who-follows-whom social network at 2009 with 3 billion edges. In the scatter plot of degree vs. triangles of Twitter accounts, some famous U.S. politicians have mildly connected followers, while adult advertisers have tightly connected followers, creating many triangles. The reason is that adult accounts are often from the same provider, and the accounts follow each other to boost their rankings, thereby creating many cliques containing triangles. Image caption: Anomaly detection in graph: the degree vs. triangles in the Twitter who-follows-whom social network at 2009. Some famous U.S. politicians have mildly connected followers, while adult advertisers have tightly connected followers, creating many triangles. The reason is that adult accounts are often from the same provider, and the accounts follow each other to boost their rankings, thereby creating many cliques containing triangles. [U KANG AND CHRISTOS FALOUTSOS, BIG GRAPH MINING: ALGORITHMS AND DISCOVERIES, SIGKDD EXPLORATION VOLUME 14 ISSUE 2, DECEMBER 2012....Read more
Smartphone-based Interaction Sensing to Innovate O..
[Prof. Junehwa Song] Advancing our smartphones as conversational interaction sensing platforms to promote social health in everyday life Article: A team of researchers in Dept. of Computer Science, KAIST are leading new mobile initiatives to build social context platforms and create life-immersive social services. The key inspiration is to advance our commodity mobile devices such as smartphones towards a face-to-face social gateway – to acquire the real-time social contexts and create proactive services naturally overlaid on our everyday face-to-face social lives. Why social? Because it is an integral basis which constitutes many of our daily activities, such as family life, team-oriented works, hanging out with friends, and so on. They believe that stretching out for diverse social contexts around the user will realize the foundation of future mobile systems which achieve holistic understanding on the user’s need and useful life-immersive services. The quest begins with a question, “What would be the primitive context in our face-to-face social interaction?” As the initial and natural step, the research team focused on conversation, which would be the most prevalent tool to communicate with people just in front of us. A number of new technical challenges arises to recognize delicate conversational contexts with commodity mobile devices. In 2013, they reached the first milestone, SocioPhone, a mobile face-to-face interaction platform with highlights meta-linguistic contexts in conversations. SocioPhone is described in a paper published in ACM MobiSys 20131 (International Conference on Mobile Systems, Applications, and Services) by a team of a KAIST graduate Youngki Lee, who is now an assistant professor of Information Systems, Singapore Management University (SMU), and 8 others. SocioPhone abstracts on-going conversations as a sequence of turns and pauses. SocioPhone provides applications with a set of intuitive APIs to monitor rich meta-linguistic context on the fly, without requiring computation-intensive semantic inference on conversation contents. At its core, the SocioPhone runtime monitors conversational turn-centric contexts in a highly-efficient and precise manner based on a new socially-leveraged collaborative sensing scheme. In addition to the paper publication, a live demonstration of SocioPhone won the Best Demonstration Award in ACM HotMobile 20132 (International Workshop on Mobile Computing Systems and Applications), which was conducted by a KAIST graduate student Chulhong Min and 14 others. Continuing the efforts, the research team is now creating compelling applications which are indispensable for a new platform to find its unique values. Notably, they proposed TalkBetter, a prominent initiative for everyday clinical care applications. The paper describing TalkBetter has won the Best Paper Award in ACM CSCW 20143 (International Conference on Computer Supported Cooperative Work and Social Computing), which was authored by an interdisciplinary team of a KAIST graduate Inseok Hwang who is now a research assistant professor of Center for Mobile SW Platform in KAIST, Prof. Dongsun Yim in Dept. of Communication Disorders, Ewha Womans University, and 5 others. They introduced a mobile in-situ intervention service to expedite everyday family-driven care for children with speech impediments. With close collaboration with the speech-language pathologists and the patients regularly taking speech-language care, TalkBetter has been designed and developed as an integrated therapeutic service which facilitates the parents to acquire clinically desirable conversation habits throughout natural conversations, not only depending on simple a-priori prescription as it has been done thus far. Today, many people observe growing concerns about the anti-social effects of smartphones, attempting to suppress the smartphone use in face-to-face social situations. These researches advocate an antithesis of such technophobic skepticisms. They are pursuing the pro-social potential of pervasive mobile devices around us, and realizing the first-of-its-kind mobile systems to enhance our everyday social life experience. These researches were supported by National Research Foundation funded by Ministry of Science, ICT, and Future Planning (MSIP) and IT R&D Program of MSIP/KEIT of the Korean Government. 1. Lee, Y., Min, C., Hwang, C., Lee, J., Hwang, I., Ju, Y., Yoo, C., Moon, M., Lee, U., Song, J. SocioPhone: Everyday Face-To-Face Interaction Monitoring Platform Using Multi-Phone Sensor Fusion. ACM MobiSys 2013, Taipei, Taiwan, June 2013. 2. Min, C., Hwang, I., Lee, J., Hwang, C., Yoo, C., Moon, M., Park, T., Lee, C., Lee, H., Kim, Y., Ju, Y., Lee, Y., Lee, U., Song, J. Demo: Bringing In-situ Awareness to Mobile Systems: Everyday Interaction Monitoring and Its Applications. ACM HotMobile 2013 (Demo), Jekyll Island, GA, USA, February 2013. 3. Hwang, I., Yoo, C., Hwang, C., Yim, D., Lee, Y., Min, C., Kim, J., Song, J. TalkBetter: Family-driven Mobile Intervention Care for Children with Language Delay. ACM CSCW 2014, Baltimore, MD, USA, February 2014. Image caption: Real-time meta-linguistic monitoring and feedback by the smartphone for the on-going conversation between the mother and her son. [PHOTO COURTESY OF THE RESEARCHERS] Image caption: Face-to-face conversation situations and socially-leveraged collaborative sensing scheme for conversational turn monitoring [PHOTO COURTESY OF THE RESEARCHERS]...Read more
Virtual Reality-based Dental Training Simulator
[Prof. Jinah Park] Dental caries and calculus are prevalent diseases among people. For better treatment of these diseases, dental students need to receive training efficiently and their skills must be assessed by standardized methods. Our research allows students to assess their skill based on the training results and provide an effective practice to master their skills. A novel collision model is developed to deal with various interactions between different shaped instruments and the oral cavity. The instrument is represented by a distance field and a set of points, and the tooth is represented as a distance field. The distance field of the tooth is updated by tracing the plastic deformation of tooth while the collision detection and the reflected force are efficiently computed from the distance field of the tooth and the set of points. Research Results 1. A collision model for real-time simulation of tooth preparation Since instruments for tooth cutting are small and sharp compared to the tooth, it is necessary to not only elaborate but also refine the collision model that handles the plastic deformation of the tooth caused by cutting instruments. To address this issue, we propose a real-time collision model in which the tooth is represented as a volume model with a distance field. This volume model is shared by not only a procedure to visualize the shape deformation of the tooth but also a procedure to detect collision and compute feedback force for haptic rendering. Our method can handle more than 10,000 volume elements within 1kHz, there is no limit on the shape of the instruments. 2. Adhesion attenuation model for dental calculus removal In contrast to the hand scaling, the ultrasonic scaling delivers ultrasonic wave energy to the dental calculus to remove it by reducing the adhesion between dental calculus and tooth surface. To simulate this characteristic, we propose an adhesion attenuation model that reduces the dental calculus adhesion by the amount of wave energy inversely proportional to the distance squared from collided voxel. Our method enables the users to use the ultrasonic scaler in the advisable way. 3. VR-based dental simulator Dental simulation ran on a personal computer with Core2Duo 3GHz central processing unit, NVIDIA Quadro FX 3700 graphics card, and 4 GB Ram. A Phantom Desktop was used to display the haptic feedback. Moreover, an immersive workbench, SenseGraphics Display 300, was used to align the hand-eye coordination. Our simulator provides different training scenarios: impacted wisdom tooth surgery, dental cavity preparation, ultrasonic scaling as shown in the Figure 1-3. Excellency and Expected effects Most of the previously developed dental simulations consider only a spherically shaped tool due to its simplicity in collision detection. Our collision model can handle any arbitrarily shaped instruments with multiple contacts between the instrument and the tooth. Moreover, synchronization between a visual model and a collision model is not required because our method resolves volume cut, multiple contacts, and feedback force within a haptic rate. The collision model can be applied to not only a bone surgery simulation but also a virtual sculpting simulation....Read more
Internet of things and system software
[Prof. Dae Young Kim] The term Internet of Things was firstly introduced in 1999 by Kevin Ashton at the Auto-ID Labs, MIT, which is the primary research partner of GS1. GS1 is a global organization which provides various types of codes such as bar code, RFID code, QR code for thing identification, and also standardizing system infrastructure for global business and application. Among seven Auto-ID Labs (MIT, Cambridge, ETH Zurich, Keio, Fudan, Adelaide, KAIST) over the world, Auto-ID Lab, KAIST have studied in IoT field since 2002, with the RFID and wireless sensor network technology. And since 2005, we started to develop various IoT technologies that are specialized to GS1 standard. We are currently working on following projects; Oliot (GS1 based IoT Infrastructure Platform), SNAIL (6LoWPAN based IoT Connectivity Platform), SeaHaven (Visual Sensor Network Cloud Platform), IoT-App Ecosystem (Ecosystem for Mobile Versatile Applications), GPGPU HPC Cloud (Cloud Computing for HPC with GPGPU). [Research Results] The IoT is the vision that aims to give every day object virtual personality. The rationale behind this is to let them have global identification, computation and communication capabilities. That is, our everyday things become intelligent and are able to provide us with any information about themselves. As a result, vast opportunities to create entirely new dimension of services appear. In this regard, we have developed the following five technologies to realize the IoT vision. (1) SNAIL: We enabled communication among things over Internet to achieve 6LoWPAN IoT network. SNAIL (Sensor Network for an All-IP World) is a solution for IoT network, which is a tiny IPv6-based sensor networking platform including a complete architecture of a lightweight TCP/IP stack supporting IPv6 adaptation, ad-hoc routing, header compression, and bootstrapping as well as four important technologies, mobility, web enablement, time synchronization, and security. (2) Oliot: Oliot is aiming an international standard based IoT Infrastructure Platform, by extending the code system of GS1 and their standard architecture to support various IoT connectivity and protocols such as bar code, RFID, ZigBee, 6LoWPAN, etc. Oliot also aims a complete implementation of GS1/EPCglobal standard. (3) IoT-App Ecosystem: IoT-App Ecosystem is a new ecosystem for mobile software, which enables easier interaction between mobile application and various smart things. This is currently being implemented for Android and supports current Android development environment. (4) SeaHaven: SeaHaven is a portable and secure multi-tenant visual sensor networks cloud platform which covers visual sensor node operating systems, visual sensor streaming service, visual big data processing service, and user applications. The major goal of this project is to make the machine understand the context of the scene by leveraging visual sensors all over the world which are very beyond human’ visual sensory. (5) GPGPU HPC Cloud: This project is about Cloud computing for HPC with general purpose graphical processing unit (GPGPU). Using GPGPU on Cloud will reduce cost and power usage than using only CPUs. We implemented platform with OpenStack, KVM, and API forwarding technique. [Excellency and Expected effects] (1) SNAIL: We make an effort to support secure, dynamic, global, and easy access to everyday objects using IPv6 address. Our IoT platform SNAIL is evaluated as a promising IoT platform and new version SNAIL 2.0 will come out soon. (2) Oliot: The entire source code was opened to public and is designated to be utilized on various project such as EU IoT6 Smart Building Project, KAIST Dr. M Project, c-ITRC Food Safety System, Stanford Civil Engineering Project, Korea University Hospital Project, and Smart Consumer Electronics. The roadmap and vision of Oliot project is described in oliot.org. (3) IoT-App Ecosystem: With this work, mobile application developers can more easily implement applications which interacts with pervasive smart things. Also, it aids the developers to publish their business logic to customers. (4) SeaHaven: To make the machine understand the context of an event is a quite challenging job to handle. Variety and sparse distribution of sensors must be the most helping key in resolving this issue. SeaHaven provides a universal interface for sensors to cover variety and heterogeneity of sensor devices and a very scalable service architecture to easily scale out the system over the cloud infrastructure. (5) GPGPU HPC Cloud: To use GPGPU on Cloud, previous approach only supports one to one mapping virtual machine and GPGPU. With API forwarding technique and Kepler architecture GPGPU, our platform supports scalable use of GPUs. And with efficient GPU resource scheduling algorithms, our platform can maximize resource utilization while providing SLA for HPC users....Read more
Investigating Reliable Computing Systems in Nano-s..
[Prof. Soontae Kim] With technology scaling, feature sizes, and supply and sub-threshold voltages are decreasing for high performance, high transistor density, and low power consumption. At the same time, microprocessors and memory systems are integrating more transistors to extract more performance. Unfortunately, these trends make computing systems more susceptible to various errors. Transient errors occur when energetic neutrons coming from deep space and alpha particles from packaging materials hit transistors, which change the state of memory bits or the output of combinational logic temporarily. In addition, permanent errors due to process variations and wear-out in interconnect and transistors increase over time, which in turn decrease yield and lifetime of the computing systems. Therefore, it is essential to provide reliable computing on top of unreliable systems for the continued success of computing. We investigate low-cost processor architectures, memory systems and software to combat against those various errors in nano-scale era. [Research Highlight] Access-time Variation Insensitive Level-1 Caches  Ever-scaling process technology increases variations in transistors. The process variations cause large fluctuations in the access times of SRAM cells. Caches made of those SRAM cells cannot be accessed within the target clock cycle time, which reduces yield of processors. To combat these access time failures in caches, many schemes have been proposed, which are, however, limited in their coverage and do not scale well at high failure rates. We proposed a new level-1 cache architecture (AVICA) employing asymmetric pipelining and pseudo multi-banking. Asymmetric pipelining eliminates all access time failures in L1 caches. Pseudo multi-banking minimizes the performance impact of asymmetric pipelining. For further performance improvement, architectural techniques were proposed. [Reference] 1. Seokin Hong and Soontae Kim. AVICA: An Access-time Variation Insensitive L1 Cache Architecture. Design Automation and Test in Europe Conference (DATE’13), March 18∼22, 2013, Grenoble, France (Best Paper Award)....Read more
OncoSearch： A web tool that searches biomedical li..
[Prof. Jong Cheol Park] OncoSearch (http://oncosearch.biopathway.org) is a web tool that allows the user to query into biomedical literature for information on cancer-related genes and shows the results for further insights into oncogenesis, with an aim to catalyze and accelerate the ongoing cancer research. [Article] Automatic identification of gene-cancer relations from a very large volume of biomedical text is an important task for cancer research since changes in genes are known to be the main cause of oncogenesis and a huge amount of information on such genes is archived in biomedical literature databases. To identify such relations, it is essential to understand, as much as possible, how a gene affects a cancer and to distinguish oncogenes (genes that cause cancers), tumor suppressor genes (genes that protect cells from cancers), and biomarkers (genes that indicate normal or cancerous states) since this will speed up the development of treatment and diagnosis methods for cancer. Although genes may sometimes be explicitly claimed as oncogenes or tumor suppressor genes in the biomedical text, it is more often the case that information on gene-cancer relations is conveyed only implicitly with detailed descriptions about gene and cancer properties. Consider the example of the sentence below. WWOX overexpression induced apoptosis and suppressed prostate cancer growth in vitro and in vivo [PMID:17704139]. While the gene WWOX is a well-known tumor suppressor, the sentence above does not contain an explicit reference to the gene as such. Instead, the sentence gives information that helps to classify the gene WWOX as a tumor suppressor of prostate cancer through the following inference: 1) WWOX expression level is increased, 2) prostate cancer regresses when WWOX expression increases, and 3) there is causality between the change in WWOX and the change in prostate cancer. By combining the three pieces of information above, one may classify the gene WWOX as a tumor suppressor gene. Although a single sentence with such implicit information may not provide enough evidence to confirm a particular gene's class, collecting a large amount of such information in the literature would certainly help to substantiate such a conclusion. Prof. Jong C. Park’s research team at KAIST developed OncoSearch, a web tool that allows the user to query into the biomedical literature for free-text information on cancer-related genes and provides the results for further insights into oncogenesis, or the process by which normal cells are transformed into cancer cells. In particular, OncoSearch can classify genes into either oncogenes, tumor suppressor genes, or biomarkers by taking into account implicit information as well as explicit information on their roles. The tool characterizes gene-cancer relations described in biomedical text with 1) how a gene changes, 2) how a cancer changes, and 3) the causality between the gene and the cancer, and the tool infers the respective roles of genes in cancers. Through this classification, the research team showed that the tool can correctly pick out oncogenes and tumor suppressor genes already registered as such in biology databases. The research team also showed that only small portions, or 6.87％ and 3.76％, respectively, of the oncogenes and tumor suppressor genes in one of the de facto standard gene databases, or UniProtKB, are registered in the list of oncogenes and tumor suppressor genes published by Vogelstein and colleagues (Science, 2013). This indicates either 1) that the process of identifying new oncogenes or tumor suppressor genes is still at an early stage or 2) that the exact definitions of oncogene and tumor suppressor gene are highly dependent upon each biology database. OncoSearch is, thus, expected to catalyze much further research in oncology since the tool can collect and infer information about novel oncogenes, tumor suppressor genes, and biomarkers from the rapidly growing body of the literature that does not necessarily contain explicit expressions such as oncogene and tumor suppressor gene....Read more
High-physicality touch interfaces
[Prof. Geehyuk Lee] Touch interfaces are now a de facto standard for information appliances. They enabled more direct and natural interaction than traditional computer interfaces, but are still in their early stage as the types of operations are quite limited compared with diverse surface operations that we perform in the real world. The limitations of the current touch interfaces stem mainly from their inability to discriminate touches of different degrees, for example, hover, light touch and heavy touch. We are interested in new possibilities that may be enabled when touch surfaces become more physically sensitive, for instance, when they can sense the approach or the pressure of the fingers as well as their touch. We are conducting a series of interaction design studies with high-fidelity prototyping, and are publishing some of early results in leading HCI conferences. Research Results Touch with Hover We developed a world-best class optical hover tracking touchpad technology and are exploring the potential of using it. Using this technology, we can provide a touch-screen-like interaction in the TV environment. For instance, the shadow of the user's fingers touches the screen, presses a button, and flicks a cover-flow-like list, while the fingers stay and move on a touchpad. In order to explore this concept, we developed a hover-tracking optical touchpad, and designed a TV application to demonstrate possible new interaction techniques. Through a prototyping study, we could correct some of our false expectations, and verify its potential as a viable option for a TV remote interface. (left) RemoteTouch concept and device, (right) Hover-tracking long touchpad Touch with Force The same finger movement on an object may be associated with different intentions depending on its normal and tangential forces. For instance, the same finger movement can be intended to turn a single page or to turn multiple pages, or to slide on a page, but a touch screen cannot differentiate these gestures. We solved this problem with a touch screen that can sense not only touch positions but also touch forces. We implemented a prototype device that can sense the normal and the tangential component of the forces on the screen, designed Force Gestures, which differ in terms of both touch movements and force patterns, and conducted an experiment to verify the feasibility of this approach. (left) example force gestures, (right) example multi-point tangential force interaction Excellency and Expected effects A complete report of our research on the RemoteTouch concept was presented at ACM CHI 2011 [C1]. The CHI paper soon attracted the attention of technology reporters and a news article about the paper appeared in DiscoveryNews [W1] and MSNBC.com [W2]. The optical hover-tracking technology was also implemented into form factors of a standard laptop touchpad and long shaped palm rest-length touchpad, and presented at ACM UIST 2011 [C2] and ACM CHI 2013 [C3], as a demo and a poster, respectively. The long touchpad was also introduced by American technology blogs, NewScientist Blog [W3] and Gizmodo [W4]. Regarding Force-sensing touch technology, a complete report is presented at ACM UIST 2011 [C4] and ACM CHI 2013 [C5], which is one of the most competitive venues for UI researchers. Published at ACM CHI and ACM UIST, the results of the current research were verified to be original and academically important. The research is also expected to have an impact on ICT industry since their main application targets are smart phones and smart TVs. After the introduction of iPhone, mobile phone manufacturers realized the importance of UI technologies and are experimenting with many new UI technologies to impress the market. High physicality touch screen or touchpad as demonstrated by our research results is certainly one of them. We were invited to present our research on touch interfaces by major companies such as LG and Samsung this year. We could also attract research funds from an industry source [F1] and a government source [F2] for the current research. The aforementioned news articles [W1, W2, W3] also reflect the public interests in the current research. References C1. Sangwon Choi, Jaehyun Han, Geehyuk Lee, Narae Lee, and Woohun Lee, RemoteTouch: Touch-Screen-like Interaction in the TV Viewing Environment, CHI 2011 (paper). C2. Sangwon Choi, Jaehyun Han, Sunjun Kim, Seongkook Heo, and Geehyuk Lee, ThickPad: A Hover-Tracking Touchpad for a Laptop, ACM UIST 2011 (demo). C3. Jiseong Gu, Seongkook Heo, Jaehyun Han, Sunjun Kim, and Geehyuk Lee, LongPad: a touchpad using the entire area below the keyboard of a laptop computer, ACM CHI 2013 (paper) C4. Seongkook Heo and Geehyuk Lee, Force gestures: augmenting touch screen gestures with normal and tangential forces, ACM UIST 2011 (paper). C5. Seongkook Heo and Geehyuk Lee, Indirect shear force estimation for multi-point shear force operations, ACM CHI 2013 (paper) W1. http://news.discovery.com/tech/shadow-remote- touchscreen-110519.html W2. http://www.msnbc.msn.com/id/43095028 W3. http://www.newscientist.com/blogs/onepercent/2013/01/trackpad-ignores-accidental-to.html?cmpid=RSS％7CNSNS％7C2012-GLOBAL％7Conline-news W4. http://gizmodo.com/5982160/intelligent-keyboard-wide-touchpad-is-smart-enough-to-ignore-your-palms Funding Sources 1. Implementation of USN Sensor Platform and Network Systems, Funded by National Research Laboratory (NRL) Program of NRF, 2007-current 2. u-Agriculture, Funded by ITRC (Information Technology Research Center) Program of MKE, 2007-current...Read more
Prof. KyuYoung Whang Receives Contributions Award..
Prof. Kyu-Young Whang, Distinguished Professor of Computer Science at KAIST, was the recipient of the 2014 ACM SIGMOND Contributions Award. Founded in 1947, the Association for Computing Machinery (ACM) is the world’s largest educational and scientific computing society, delivering resources that advance computing as a science and profession. SIGMOD is the Association for Computing Machinery’s Special Interest Group on Management of Data, which specializes in large-scale data management problems and databases. Since 1992, the ACM SIGMOND has presented the contributions award to one scientist who has made significant contributions to the field of database systems through research funding, education, and professional services. So far, 23 people including Professor Whang have received the award. Professor Whang was recognized for his key role in the growth of international conferences and journals in the field of database such as The VLDB Journal (The International Journal on Very Large Data Bases), VLDB Endowment Inc., IEEE Technical Committee on Data Engineering, and Database Systems for Advanced Applications (DASFAA). IEEE stands for the Institute of Electrical and Electronics Engineering. For the full list of ACM SIGMOND Contributions Award recipients, please go to http://www.sigmod.org/sigmod-awards/sigmod-awards#contributions...Read more
Prof. Min H. Kim is appointed as an Associate Edit..
Prof. Min H. Kim is appointed as an Associate Editor of ACM Transactions on Graphics (TOG). The Association for Computing Machinery (ACM) was founded in 1947 and has served as the world’s most prestigious scientific and educational computing society along with the Institute of Electrical and Electronics Engineers (IEEE). The roles and responsibilities of an Associate Editor include selecting appropriate referees to perform reviews on submitted manuscripts and preparing reports for the main findings of the review process. The manuscripts selected for publication are presented at the world’s largest Computer Graphics conference, ACM SIGGRAPH. Professor Kim has published numerous papers in the areas of computer graphics research, with emphases in the areas of 3D imaging spectroscopy and visual perception. He regards his appointment to TOC as a great opportunity and looks forward to making further outstanding contributions to advance research in computing....Read more
Eugen Wüster Prize Award
[Prof. KeySun Choi] At the closing ceremony of the International Conference in Terminology and Knowledge Engineering (TKE) 2014 hosted by the German Institute for Standardization in Berlin, Germany, Professor Key-Sun Choi of Department of Computer Science and Korea Terminology Research Center for Language and Knowledge Engineering (KORTERM) at KAIST has been awarded the Eugen Wüster Prize for his long-time international achievements in the field of terminology science as the Secretary of ISO/TC37/SC4 for language resource management and the Vice-President of Infoterm since 2002. The Prize named in honour of Eugen Wüster (1898-1977), commonly known as the “Father of Terminology Science”, is being awarded every three years from 1997, to recognize scholars of outstanding achievements in the field of terminology science and other related studies. So far, a total number of 10 scholars around the world, including Prof. Choi, have received the Award. For further information, please visit http://www.infoterm.info/activities/news/2014/2014_07_06.php....Read more
Developing a Text Mining Search Engine for Cancer ..
[Prof. Jong Cheol Park] The Ministry of Science, ICT, and Future Planning announced on May 22nd that the research teams led by Professor Jong Cheol Park of KAIST and Professor Hyunju Lee of GIST have developed OncoSearch, a text mining search engine that searches Medline abstracts for sentences describing gene expression changes in cancers. This federally funded research project was carried out on an interdisciplinary research effort converging linguistics, computer engineering, biology, and medicines. In order to identify genes that cause cancers and to understand how such genes affect cancers, abnormal gene expressions in cancers are actively studied. To facilitate the studies, OncoSearch utilizes powerful text mining techniques to extract the relevant information from the large amount of information available in the biomedical literature. OncoSearch allows the user to efficiently search for genes that affect particular types of cancers, compare expression levels of a gene across various types of cancers, and explore a graph to find interactions between genes in a type of cancer. “OncoSearch is a novel tool that automatically collects information on cancer related genes using the latest text mining techniques, and we expect its active use will help the ongoing cancer research efforts,” said Professor Park. The research results were published in the online version of Nucleic Acids Research, on May 9th, 2014....Read more
Prof. Kyu-Young Whang Recognized for his Distingui..
Prof. Kyu-Young Whang has received the Distinguished Contributions Award at the Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD) which was held in Taiwan on May 13-16th. At this respected data mining conference in the Pacific-Asia region, Professor Whang was recognized for his influential work in the research field as a life member on the PAKDD Steering Committee. The award was given to 6 recipients including Professor Whang, and he is the first Korean person to be a recipient. In 2011, Professor Whang also received the Outstanding Contributions Award at another prestigious conference called Database Systems for Advanced Applications (DASFAA). The database and data mining research in the Asia-Pacific region, regarded as primitive 20 years ago, have come a long way to be on comparable footings with those in North America/Europe, thanks to relentless efforts and contributions made by countlessly many researchers including Professor Whang. Recently, database research in the Asia-Pacific region is much heightened as, among the three major academic organizations in the database area, Prof. Whang is leading IEEE TCDE, Prof. Beng Chin Ooi from Singapore The VLDB Endowment, and Prof. Don Kossmann from Switzerland ACM SIGMOD....Read more
Prof. Kwangjo Kim to Represent Korea in IFIP TC11
Prof. Kwangjo Kim was appointed to represent Korea in International Federation for Information Processing (IFIP)’s Technical Committee-11 (TC-11) for his internationally recognized contributions to the advancement of security and cryptology research in the past 30 years. Since its establishment in 1960, IFIP has been the leading multinational, apolitical organization in Information & Communications Technologies and Sciences. It is recognized by many international organizations including the United Nations for bringing IT Societies from 56 countries together on important topics in informatics. In particular, TC-11 is involved in research for increased trustworthiness in information processing, such as developing a common frame of reference for security and privacy protection. “Through this IFIP role, I would like to present Korea’s outstanding information and security technology and show that we are one of the world’s leading nations in the security technology,” said Professor Kim. His previous roles include Director of International Association for Cryptologic Research, AsiaCrypt Steering Committee Chair, President of Korea Institute of Information Security & Cryptology, and visiting professor at MIT and KUSTAR. This year, he is serving as the General Chair of Cryptographic Hardware and Embedded Systems (CHES) 2014, to be held in Busan this coming September....Read more
The Godfather of the Internet in Korea
[Prof. Kilnam Chon] While it is a commonly known fact that the United States is the birthplace of the Internet, many people are surpised to learn that Korea was the second country to successfully establish the Internet. The historic setting was May 15, 1982 inside a research laboratory located in Gu-mi, Korea. Dr. Chon and his team watched as their computer printed out “＄ rlogin snucom” on the screen, indicating a successful remote login to a computer at Seoul National University. Hugging each other and crying shouts of joy, the research team led by Dr. Chon had officially opened a new chapter in the history of the Internet. From that day on, Chon has remained the godfather of the Internet in Korea. Dr. Chon received a PhD degree in computer science from University of California, Los Angeles in 1974, and a BS degree in engineering science from Osaka University. He joined the Korea Institute of Electronics Technology in 1979 to work on computer system development, and moved to Korea Advanced Institute of Science and Technology in 1982 as a professor in the Computer Science Department. In a recent interview with a Korean newspaper, Dr. Chon explained that he came to Korea with a determination to make a difference in the country with his technical expertise. With a full support from the Korean government, Dr. Chon and his fellow researchers pursued the challenging goal of advancing the computing technology in Korea and succeeded in making Korea to be the world’s second to establish the Internet. During his time as a professor in KAIST Compuster Science Depeartment, Dr. Chon was well known for his demand for perfection from his students. He would not only help his students fulfill their full intellectual potential but also emphasize the importance of regular physical exercise. He sharply points out the fact that the students who succeeded the most after graduating from his laboratory are the ones who exercised the most rigorously under his advice. After his retirement, Dr. Chon, the godfather of the Internet in Korea, is currently working as a professor at Keio University with the goal of making its computer science program to be the best in Japan....Read more