56016 research outputs found
Sort by
A Gene-Centric Perspective of Scientific and Technological Innovations
No full textDepartment of Biomedical EngineeringResearch on genes and gene products is a foundation of modern biotechnology, and recognized for its applicability in medicine, agriculture, food industry, energy supply, environmental remediation, and many others. To investigate a macroscopic and gene-centric perspective of scientific discoveries and technological innovations, we employed a large-scale curation of research papers and patents. As a raw data to represent the scientific impact of each gene, we collected the entire set of research articles available on PubMed, that have the names of genes or gene products in their title or abstract. The more dedicated literature source, the United States Patent and Trademark Office (USPTO) patent publication, was retrieved as a source data to represent the counterpart in technological innovations. In parallel with this literature curation, the gene symbols were collected from curated subset of UniProt consortium database maintained by European Bioinformatics Institute, and then clustered into the non-overlapping standardized sets to eliminate the overwhelming duplicates and possible contamination by common acronyms. Based on the annual counts of papers or patents whose titles/abstracts include a given gene, we show the overall trends of genetic research since the launch of the Human Genome Project. The gene citation fluctuated more in inventive activity compared to those in the research, while both sides were largely contributed by medicinal discipline. The volume of publications mentioning genes has been increased while the debut of new genes on titles and abstracts has been deflated. In contrast, new combinations of previously-studied genes kept actively explored, and their frequently adopted genes informed biotechnology innovations rather than sheerly top-studied genes.clos
A Page Scheduler using Machine Learning for Hybrid Memory Systems
Get PDFDepartment of Computer Science and EngineeringAs the demand for machine learning and big data workloads grows, the memory becomes important in application performance. The main memory is extended by using hybrid memory systems that include different types of memory components. Data placement across multiple memory components has a significant impact on application performance. We propose SMA, an RNN-based page scheduler to learn page access patterns and ensure that pages to be accessed by applications in the future are prepared in a fast memory in advance. This paper utilizes the existing observation that there is a set of pages that are important to application performance. A single RNN model manages all important pages for efficient page management, and the remaining pages are managed using a history-based method. This work reduces training time and memory usage compared to the existing state-of-the-art machine learning based page scheduler while providing higher accuracy. It also shows that a single RNN model can learn general page accesses patterns by achieving similar accuracy with the existing page scheduler for applications that were not included in the training dataset of the model.ope
A STUDY ON THE AI PREDICTION MODEL OF CURVED BLOCK PRODUCTION PLANNING IN SHIPYARD WITH PRE-PROCESSING
No full textGraduate School of Artificial Intelligenceclos
CNCToolDQN : Deep Q-learning on the anomaly detection and remaining useful life estimation for tool monitoring
No full textDepartment of Industrial EngineeringMonitoring tool health in the computer numerical control (CNC) manufacturing process has been considered an important technique because the overall quality and yield of machined workpiece, as well as the continuity of the manufacturing process, strongly depend on tool conditions. At present, previous approaches for data-driven tool monitoring have been divided into supervised remaining useful life (RUL) estimation and unsupervised anomaly detection (AD). However, both methods have their own limitations, conducting one of them, but not both leads to unsatisfactory performance. Also, considering their different objectives and the main goal of tool monitoring (i.e., predicting proper timing of tool replacements), they should be conducted simultaneously. To address this issue, we propose the CNCToolDQN framework, which is designed to perform the functions of AD and RUL simultaneously. Specifically, the Deep-Q-network agent is trained to identify anomalies using tool load and age data, which are normally collectable from any modern CNC manufacturing process. Then, its predicted output is used to estimate RUL score, which is based on the history of predicted anomalies that occurred during the tool???s lifetime. We conducted an experiment using three real-world tool monitoring datasets. As a result, we obtained the best AD performance against existing models such as Anomaly Transformer and USAD (unsupervised anomaly detection on multivariate time series). We expect that this framework would contribute to fill the gap between AD and RUL, as well as perform the prognostics of tool monitoring, to alert users for tool replacements to prevent shutdown and maximize yield.clos
3?????? ????????? ???????????? ??????????????? ?????? ????????? ?????? ???????????????
No full textDepartment of Electrical Engineeringclos
Development of Hematite Photoanode for Photoelectrochemical Water Oxidation
No full textSchool of Energy and Chemical Engineering (Energy Engineering)clos
Development of Quantum Bead-based Lateral Flow Assay for Detection of Respiratory Syndrome Viruses
No full textSchool of Energy and Chemical Engineering (Energy Engineering)clos
Fundamentals and developments in 4d Ru-based Li-rich layered oxides
No full textSchool of Energy and Chemical Engineering (Energy Engineering (Battery Science and Technology))Layered oxide insertion compounds have been mainly used to host Li ion sources, which are composed of Li ions, transition metals (TMs) and oxygens (for example, LiNiO2, LiCoO2, LiNi1-x-yCoxMnyO2, etc.). In this type of cathode material, oxygens are coordinated by three Li ions andthree TM ions, and forms cubic close-packed (ccp) oxygen arrays. With these oxygen arrays, both TM and Li octahedrons are stacked alternatively, forming a layered structure. This layered structure operates through (de-)intercalation reaction of Li ions, ideally without volume change and degradation during charging/discharging. During the reaction, Li ions and electrons are reversibly extracted/inserted in Li layers, balancing the electroneutrality with change in oxidation states of electrochemically active TMs. In other words, the capacity of LIBs is limited by the number of electrons that oxidizable TMs afford to supply. Unfortunately, however, these conventional layered oxides suffer from large variation of lattice parameter in crystal structure due to O-O repulsive forces at highly charged states, causing structural degradation.
As an alternative to conventional layered oxides, lithium-rich layered oxides have been attracted much attention. These novel layered oxides are synthesized by incorporating excess Li instead of TM ions in the TM layers (for example, zLi2MnO3-(1-z)LiTMO2 (TM=Mn, Co, Ni, etc.)), and delivers abnormal high capacities beyond its theoretical capacity within wide voltage range of 2.0???4.6 V. It is believed that lattice oxygens participate in the redox reaction, forming the oxidized oxygen species (On-???O(n-1)-). Nevertheless, the cumulative contribution from both cationic and anionic reversible redox processes has been overlooked since the oxidized oxygens are intrinsically unstable, giving rise to irreversible loss of lattice oxygens. In the early stages of these novel cathodes, its abnormal capacity was estimated to be caused by irreversible oxygen gas evolution in which oxygen dimerization loses electrons and lattice oxygens escape from the host materials. Recently, however, many researchers have attempted to make anionic redox reversible and to reveal why the extraordinary capacity is offered by Li-rich cathode.
In this thesis, I will take a close look at fundamental redox principles of Li-rich cathode materials for high-energy-density batteries, and its methods to improve high-voltage redox stability. With the help of various characterization tools and computations, I have suggested the redox behaviors of Ru-based Li-rich cathodes within wide voltage ranges.clos
?????? ??????????????? ?????? ?????? ????????? ?????? ???????????? ?????? ?????? ?????????
No full textGraduate School of Artificial IntelligenceDemand-Responsive Transport (DRT) is a dynamically routed bus service that caters multiple-origin-multiple-destination demands. DRT has been an attention-getting method as one of the best alterna- tives to fulfill the grouped passengers. As to running DRT service, DRT service operators must con- sider the number of passengers with respect to distance transports travel called Passenger Occupancy Rate(passenger per kilometer). POR is important in determining potential profits of DRT. Meanwhile, for sustainable management of DRT in big cities, it has been emphasized for the DRT service operators to consider urban compactness. This begs the question of how service operators should balance POR with the needs of their service? To answer this, I propose Demand-Oriented Route Recommender (DORR) to increase POR in demand compact areas and times. DORR is a framework that recommends sustainable DRT routes with high demand hubs surrounded in the compact demand area based on density-based algorithm for spotting profitable transport hubs. DORR utilizes a deep learning model to predict pick- up and drop-off demands to generate DRT routes targeting latent passengers. Not only do the results of our experiment indicate that our framework finds routes with higher PORs than those from conven- tional methods in both individual and grouped requests, but I also discuss valuable insights related to the features of individual request and group request ride-share data.ope
?????? ???????????? ????????? 3?????? ?????? ???????????? ?????? ?????? ?????????
No full textDepartment of Computer Science and Engineeringclos