A5101550276- Advanced Data Storage Technologies
- Parallel Computing and Optimization Techniques
- Distributed systems and fault tolerance
- Advanced Image Fusion Techniques
- Cloud Computing and Resource Management
- Image and Signal Denoising Methods
- Advanced Image Processing Techniques
- Embedded Systems Design Techniques
- Advanced Vision and Imaging
- Interconnection Networks and Systems
- Data Management and Algorithms
- Satellite Image Processing and Photogrammetry
- Optimization and Search Problems
- Cellular Automata and Applications
- Image Enhancement Techniques
- Caching and Content Delivery
- Robotics and Sensor-Based Localization
- Image Processing Techniques and Applications
- Remote-Sensing Image Classification
- Advanced Image and Video Retrieval Techniques
- Video Surveillance and Tracking Methods
- Calibration and Measurement Techniques
- Simulation and Modeling Applications
- Generative Adversarial Networks and Image Synthesis
- Virtual Reality Applications and Impacts
Korea Aerospace Research Institute
2015-2023
Samsung (South Korea)
2016-2020
University of Colorado Boulder
2009-2013
Sungkyunkwan University
2010
Intel (United States)
2006-2008
Research Triangle Park Foundation
2003-2004
IBM (United States)
2002
Korea Atomic Energy Research Institute
2001
University of Southern California
1995-1999
Engineering Systems (United States)
1995-1999
Data-intensive queries are common in business intelligence, data warehousing and analytics applications. Typically, processing a query involves full inspection of large in-storage sets by CPUs. An intuitive way to speed up such is reduce the volume transferred over storage network host system. This can be achieved filtering out extraneous within storage, motivating form near-data processing. work presents Biscuit, novel framework designed for modern solid-state drives. It allows programmers...
This paper presents YourSQL , a database system that accelerates data-intensive queries with the help of additional in-storage computing capabilities. realizes very early filtering data by offloading scanning query to user-programmable solid-state drives. We implement our on recent branch MariaDB (a variant MySQL). In order quantify performance gains YourSQL, we evaluate SQL varying complexities. Our result shows reduces execution time whole TPC-H 3.6×, compared vanilla system. Moreover,...
Data-intensive queries are common in business intelligence, data warehousing and analytics applications. Typically, processing a query involves full inspection of large in-storage sets by CPUs. An intuitive way to speed up such is reduce the volume transferred over storage network host system. This can be achieved filtering out extraneous within storage, motivating form near-data processing. work presents Biscuit, novel framework designed for modern solid-state drives. It allows programmers...
In a memory hierarchy, there are various classes of systems depending on the access latency. A typical hierarchy consists CPU cache, DRAM, and an SSD or HDD. The DRAM has latency 100ns, while flash about 50μs [1]. Recently, new non-volatile memories with latencies less than 10μs, including PRAM, MRAM, ReRAM [2], getting attention for business-critical such as big-data analysis storage caches. To meet low requirements, type NAND flash, Z-NAND, read time (t <inf...
Multispectral satellite imaging sensors acquire various spectral band images and have a unique spectroscopic property in each band. Unfortunately, image artifacts from sensor noise often affect the quality of scenes negative impact on applications for imagery. Recently, deep learning approaches been extensively explored to remove Most denoising methods, however, follow supervised scheme, which requires matched noisy clean pairs that are difficult collect real situations. In this article, we...
Cache replacement algorithms originally developed in the context of simple uniprocessor systems aim to reduce miss count. However, modern systems, cache misses have different costs. The cost may be latency, penalty, power consumption, bandwidth or any other ad-hoc numerical property attached a miss. In many practical situations, it is desirable inject into policy. this paper, we propose several extensions LRU which account for nonuniform These implementations, yet they are very effective...
Performance critical transaction and storage systems require fast persistence of write data. Typically, a non-volatile RAM (NVRAM) is employed on the datapath to permanent storage, temporarily quickly store data before system acknowledges request. NVRAM commonly implemented with battery-backed DRAM. Unfortunately, DRAM small costly, occupies precious DIMM slot. In this paper, we make case for dual, byte- block-addressable solid-state drive (2B-SSD), novel NAND flash SSD architecture designed...
Recent advances in deep learning have shown impressive performances for pan-sharpening. Pan-sharpening is the task of enhancing spatial resolution a multi-spectral (MS) image by exploiting high-frequency information its corresponding panchromatic (PAN) image. Many deep-learning-based pan-sharpening methods been developed recently, surpassing traditional approaches. However, most them are trained lower scales using misaligned PAN-MS training pairs, which has led to undesired artifacts and...
Cache replacement algorithms originally developed in the context of uniprocessors executing one instruction at a time implicitly assume that all cache misses have same cost. However, modern systems, some are more expensive than others. The cost may be latency, penalty, power consumption, bandwidth or any other ad hoc numerical property attached to miss. We call class designed minimize nonuniform miss function "cost-sensitive algorithms". In this paper, we first introduce and analyze an...
Since incoming light to an unmanned aerial vehicle (UAV) platform can be scattered by haze and dust in the atmosphere, acquired image loses original color brightness of subject. Enhancement hazy images is important task improving visibility various UAV images. This paper presents a spatially-adaptive dehazing algorithm that merges histograms with consideration wavelength-dependent atmospheric turbidity. Based on wavelength-adaptive acquisition model, proposed consists three steps: (i)...
3D NAND flash memory has enhanced its areal density by more than 50% per year virtue of the aggressive development WL stacking technology for recent three consecutive years [1]–[3]. Also storage market still requires bits diverse digital applications. [4]
In very high-resolution (VHR) push-broom-type satellite sensor data, both destriping and denoising methods have become chronic problems attracted major research advances in the remote sensing fields. Since estimation of original image from a noisy input is an ill-posed problem, simple noise removal algorithm cannot preserve radiometric integrity data. To solve these problems, we present novel method to correct VHR data acquired by combining wavelet-Fourier multiscale non-local means (NLM)...
Article Free Access Share on Optimal replacements in caches with two miss costs Authors: Jaeheon Jeong Department of Electrical Engineering - Systems, University Southern California, Los Angeles, CA CAView Profile , Michel Dubois Authors Info & Claims SPAA '99: Proceedings the eleventh annual ACM symposium Parallel algorithms and architecturesJune 1999 Pages 155–164https://doi.org/10.1145/305619.305636Online:01 June 1999Publication History 17citation446DownloadsMetricsTotal Citations17Total...
Satellite Synthetic Aperture Radar (SAR) images are immensely valuable because they can be obtained regardless of weather and time conditions. However, SAR have fatal noise less contextual information, thus making it harder interpretable. So, translation to Electro-Optical (EO) is highly required for easier interpretation. In this paper, we propose a novel coarse-to-fine context-aware SAR-to-EO image (CFCA-SET) framework misalignment-resistant loss the misaligned pairs SAR-EO images. With...
Hardware emulation using FPGAs is an intermediate approach between software simulation and hardware prototyping. The RPM-2 multiprocessor emulator uses this to achieve much greater flexibility observability at less cost than typical prototypes.
Recent advances in Field-Programmable Gate Arrays (FPGA) and programmable interconnects have made it possible to build efficient hardware emulation engines. In addition, improvements Computer-Aided Design (CAD) tools, mainly synthesis greatly simplify the design of large circuits. The RPM (Rapid Prototype Engine for Multiprocessors) Project leverages these two technological advances. Its goal is develop a common platform multiprocessor systems with different architectures.
We present a visual odometry algorithm for trinocular systems with divergent views and minimal overlap. Whereas the bundle adjustment is preferred method multi-view problems, it infeasible if number of features in images-such as HD videos-is large. propose divide conquer approach, which reduces problem to five monocular one each individual camera sequence two more using matched temporally from consecutive images center left right cameras, respectively. Unlike method, whose computational...
Multi-spectral satellite imaging sensors acquire various spectral band images such as red (R), green (G), blue (B), near-infrared (N), etc. Thanks to the unique spectroscopic property of each with respective objects on ground, multi-spectral imagery can be used for geological survey applications. Unfortunately, image artifacts from sensor noises often affect quality scenes and have negative impacts applications imagery. Recently, deep learning approaches been extensively explored removal in...
When track the recognized rigid target object for Position Based Visual Servo(PBVS) system on generic complex environment, existing optical flow based tracking approach has possibility to occur outlier due its aperture problem which is occurred from drifting points. In this paper, we propose method how remove points using Kalman filter. We compare measurement result with estimation filter and then our proposed detects Then detected outliers are removed. Finally, calculate pose without...
Classic cache replacement policies assume that miss costs are uniform. However, the correlation between rate and performance is not as straightforward it used to be. Ultimately, true cost measure of a should be penalty, i.e. actual processing bandwidth lost because miss. It known that, contrary loads, penalty stores mostly hidden in modern processors. To take advantage this observation, we propose simple schemes replace load misses by store misses. We extend classic algorithms such LRU...
For real time weather forecasting, it is necessary to search most similar map very fast among a large amount of data accumulated so far. Recently, deep learning used for more accurate forecasting. However, takes huge training model in order process number previous maps. In this paper, we shall present distributed parallel algorithms neural network based on CNN and environment with GPUs various models extract from CNN. each case single multi nodes, compare the performance our algorithm...