Search

Dialogue models that automatically generate utterance sentences are being actively studied with the increase of data and the development of deep neural network technology. Research is also being conducted using information from multiple modalities to create dialogues that are more natural and reflect various aspects in the dialogue. In this paper, we focus dialogue strategy, which is important for understanding a natural dialogue flow. We implement a deep-learning based, dialogue strategy classifier based on the data labeled with emotion information from multiple modalities (i.e., text, sound, and image). Results of our experiment indicate that the use of emotion information from multi-modalities can help classify dialogue strategies with approximately 10% performance increase compared to the baseline. Results also show that the emotion information from the sound modality contributes to dialogue strategy classification more than that from the text and image modalities.

Human-computer interaction has attracted much attention from researchers and the public. Detecting user intent from a user utterance is a crucial sub-task for human-computer interaction. For task-oriented dialogue systems, correct detection of user intent makes task completion requested by a user possible. Many traditional methods for intent classification suggest rule-based methods which display inherent limitation caused by handcrafted rules. Recently significant research effort has been spent on deep learning based approaches to overcome the limitation of rule-based methods.In this paper, we conduct experiments to compare the performance among different amounts of context on deep learning based intent classification models. We utilize a Long Short-Term Model (LSTM) network and a Recurrent Convolutional Neural Network (RCNN) for intent classification models. We construct our multi-turn dialogue dataset consisting of intent labeled task-oriented dialogues between a user and a dialogue system in Korean. Both LSTM and RCNN based intent classification models illustrate higher accuracies with more amounts of context.

Accordingly a lot of broadcasting medias pursuing various concepts have been appeared and the major type of contents consumed on the web has been changed to multimodal contents, the attempt to actively utilize multimedia content in artificial intelligence research is also starting. This paper introduces a study that constructs a converged information dataset in an integrated form by analyzing various types of multimodal information on video clips. The constructed dataset was released with various semantic labels for artificial intelligence research about various information classification. The labels and descriptions in this dataset include various context, intention and emotion information describing with vision, speech and language in each video clips. The constructed dataset can be resolved the problem of lack of public data for multimodal interaction research with Korean. It is expected that this dataset can be applied in the constructions of various artificial intelligence services like Korean dialogue processing, visual information extractions and various multimodal data analysis tasks.

String pattern matching with finite automata (FAs) is a well-established method across many areas in computer science. Until now, data dependencies inherent in the pattern matching algorithm have hampered effective parallelization. To overcome the dependency-constraint between subsequent matching steps, simultaneous deterministic finite automata (SFAs) have been recently introduced. Although an SFA facilitates parallel FA matching, SFA construction itself is limited by the exponential state-growth problem, which makes sequential SFA construction intractable for all but the smallest problem sizes.In this paper, we propose several optimizations to leverage parallelism, improve cache and memory utilization and greatly reduce the processing steps required to construct an SFA. We introduce fingerprints and hashing for efficient comparisons of SFA states. Kernels of x86 SIMD-instructions facilitate cache-locality and leverage data-parallelism with the construction of SFA states. Our parallelization for shared-memory multicores employs lock-free synchronization to minimize cache-coherence overhead. Our dynamic work-partitioning scheme employs work-stealing with thread-local work-queues. The structural properties of FAs allow efficient compression of SFA states. Our construction algorithm dynamically switches to in-memory compression of SFA states for problem sizes which approach the main memory size limit of a given system.We evaluate our approach with patterns from the PROSITE protein database. We achieve speedups of up to 312x on a 64-core AMD system and 193x on a 44-core (88 hyperthreads) Intel system. Our SFA construction algorithm shows scalability on both evaluation platforms.

In many agricultural applications, PolSAR data are widely used because they can be decomposed into various scattering components, which can be of help in observing the characteristics of agricultural areas. Recently, studies have been conducted to find suitable polarimetric parameters for specific applications. This paper tried to find appropriate polarimetric parameters for line extraction from agricultural areas as the line features are among the basic features of the surface. Towards this end, various polarimetric parameters were produced using polarimetric decomposition methods. LSD was used to extract lines from each polarimetric parameter image over agricultural areas, without any threshold value selection. The comparison of the line extraction result from each polarimetric parameter with the others was conducted through quantitative evaluation. Through this process, three parameters of the Pauli decomposition is found to be the suitable polarimetric parameters for line extraction from agricultural areas.

The crop boundary data provide the basic information for agricultural management. In this study, line-based boundary enhancement and extraction methods are proposed to delineate detailed paddy boundaries. Line extraction is performed based on the boundary enhancement result and Hough line extraction using RapidEye satellite image. The proposed line-based method is efficient in detecting the paddy boundaries while preserving linearity. In addition, the proposed method adopts an automated boundary detection process except for line editing. Therefore, the proposed method economically provides information on the paddy boundaries that can be utilized as the basic spatial unit for yield analysis, ownership management, and precision farming.