Attempts to train a comprehensive artificial intelligence capable of solving multiple tasks have been impeded by chronic problem called catastrophic forgetting. Although simply replaying all previous data alleviates the problem, it requires large memory and even worse, often infeasible in real world applications where access past is limited. Inspired generative nature hippocampus as short-term system primate brain, we propose Deep Generative Replay, novel framework with cooperative dual...

Co-speech gestures enhance interaction experiences between humans as well and robots. Most existing robots use rule-based speech-gesture association, but this requires human labor prior knowledge of experts to be implemented. We present a learning-based co-speech gesture generation that is learned from 52 h TED talks. The proposed end-to-end neural network model consists an encoder for speech text understanding decoder generate sequence gestures. successfully produces various including...

Deep learning, based on which many modern algorithms operate, is well known to be data-hungry. In particular, the datasets appropriate for intended application are difficult obtain. To cope with this situation, we introduce a new dataset called ETRI-Activity3D, focusing daily activities of elderly in robot-view. The major characteristics as follows: 1) practical action categories that selected from close observation lives elderly; 2) realistic data collection, reflects robot's working...

As-built building information modeling (BIM) currently is regarded as a tool with the potential to manage buildings efficiently in operation and maintenance phases. However, as-built BIM labor-intensive process that requires considerable cost time existing buildings. Although active research on scan-to-BIM automation has addressed this issue, previous studies modeled only major objects such walls, floors, ceilings, consequently requiring other indoor spaces. In addition, there was limitation...

We present a new class of irregular low-density parity-check (LDPC) codes for moderate block lengths (up to few thousand bits) that are well-suited rate-compatible puncturing. The proposed show good performance under puncturing over wide range rates and suitable usage in incremental redundancy hybrid-automatic repeat request (ARQ) systems. In addition, these linear-time encodable with simple shift-register circuits. For length 1200 bits the outperform optimized LDPC extended...

Automatic engagement recognition is a technique that used to measure the level of people in specific task. Although previous research has utilized expensive and intrusive devices such as physiological sensors pressure-sensing chairs, methods using RGB video cameras have become most common because cost efficiency noninvasiveness cameras. are usually based on hand-crafted features statistical temporal dynamics modeling algorithm. This paper proposes data-driven convolutional neural networks...

Variational Autoencoders (VAEs) and other generative models are widely employed in artificial intelligence to synthesize new data. However, current approaches rely on Euclidean geometric assumptions statistical approximations that fail capture the structured emergent nature of data generation. This paper introduces Convergent Fusion Paradigm (CFP) theory, a novel framework redefines generation by integrating dimensional expansion accompanied qualitative transformation. By modifying latent...

To better interact with users, a social robot should understand the users’ behavior, infer intention, and respond appropriately. Machine learning is one way of implementing intelligence. It provides ability to automatically learn improve from experience instead explicitly telling what do. Social skills can also be learned through watching human–human interaction videos. However, datasets are relatively scarce interactions that occur in various situations. Moreover, we aim use service robots...

A character segmentation algorithm for automatic license plate recognition is presented in this paper. Character regions are selected through binarization, connected component analysis, and recognition. blob analysis operation excludes noisy blobs, merges fragmented splits clumped blobs. module achieved an accuracy rate of 97.2%. The the complete system with localization was 90.9%. In depth failure cases also provided better understanding a future development direction.

The purpose of this study was to identify the effects on cognitive function, depression, and activities daily living Silver-Care-Robot Program for institutionalized elders.This a nonequivalent control group pretest-posttest design. participants were 42 elders (17 in experimental 25 group). provided as an intervention which conducted twice week 5 weeks. is integrated entertainment program help mental, emotional, physical health elderly people. Pre-test two groups, and, order examine...

Fully automating machine learning pipelines is one of the key challenges current artificial intelligence research, since practical often requires costly and time-consuming human-powered processes such as model design, algorithm development, hyperparameter tuning. In this paper, we verify that automated architecture search synergizes with effect gradient-based meta learning. We adopt progressive neural \cite{liu:pnas_google:DBLP:journals/corr/abs-1712-00559} to find optimal architectures for...

Autonomous mobile robots have been deployed at hospitals to cope with numerous delivery services. Robots deliver items from point through autonomous navigation. Conventional work individually conducting only one task a time. This paper proposes fleet optimization method maximize the advantages of operating group concurrently by assigning multiple tasks robot. The server selects an appropriate robot for based on cost each minimum is assigned new task. way, conduct tasks. scheduler finds...

Non-verbal behavior is essential for embodied agents like social robots, virtual avatars, and digital humans. Existing authoring approaches including keyframe animation motion capture are too expensive to use when there numerous utterances requiring gestures. Automatic generation methods show promising results, but their output quality not satisfactory yet, it hard modify outputs as a gesture designer wants. We introduce new toolkit, named SGToolkit, which gives higher than automatic more...

We present a new class of irregular low-density parity-check (LDPC) codes for moderate block lengths (up to few thousand bits) that are well-suited rate-compatible puncturing. The proposed show good performance under puncturing over wide range rates and suitable usage in incremental redundancy hybrid-automatic repeat request (ARQ) systems. In addition, these linear-time encodable with simple shift-register circuits. For length 1200 bits the outperform optimized LDPC extended...

Most previous studies on the association between bullying and body weight were performed using North American or White samples from Western countries. The present study is first empirical endeavor to examine whether such an exists in a sample of adolescents East Asia. Specifically, authors examined associations bully/victim experiences among 805 male 835 female middle school students South Korea questionnaire survey. Results showed that overweight more likely both physically bully others be...

This paper presents a recurrent learning-based facial attribute recognition method that mimics human observers' visual fixation. The concentrated views of observer while focusing and exploring parts image over time are generated fed into network. network makes decision concerning attributes based on the features gleaned from observer's fixations. Experiments expression, gender, age datasets show applying fixation to networks improves rates significantly. proposed not only outperforms...

In this paper we study and propose an algorithm to puncture finite length low density parity check (LDPC) codes (Ha, J, et al., 2002). The introduced puncturing criterion results in good performance (for 1024 4096 bits) when compared with both random dedicated LDPC codes, i.e. unpunctured designed for a given rate. comparison also shows that the proposed punctured have better block-error rates than because of longer effective block lengths high-rate puncturing. Although apply idea regular...

This paper presents an elder care robot system for older people. An provides human-robot interaction application and monitoring service. The study was carried out five weeks in nursing homes, with the participation of 17 residents. We have conducted a t-test to compare difference between satisfaction usability evaluation residents staff.

The BCH codes are widely used as Error Correcting Code (ECC) schemes for NAND Flash Memories. There have been strong demands to implement controller having low cost, power and high throughput. We focus on implementation since it has the largest portion in controller. In this paper, we configure 3-stage pipelined decoder: syndrome computation, Berlekamp-Massey algorithm Chien search. decoder supporting 800MB/s using structure early termination method.