Chung Cheng University Institutional Repository
Not a member yet
889 research outputs found
Sort by
利用化妝攻擊對人臉辨識系統生成對抗範例;Generating Adversarial Examples by Makeup Attacks on Face Recognition
No full text[[abstract]]機器學習的發展非常迅速，在電腦視覺和自然語言處理方面都有很好的成果。有許多深度學習技術被運用在人類日常生活當中例如自動駕駛汽車和人臉辨識系統。如今，人類日常生活漸漸的依賴於深度神經網路可能會導致嚴重的後果，因此神經網路的安全性就變的很重要。因此展現深度神經網路其實有明顯的弱點，我們提出了一種基於生成對抗網絡的方法來生成可以欺騙人臉識別系統的臉部化妝圖片。我們在人類無法察覺的異常化妝照片結果中隱藏了攻擊的擾動信息。實驗結果顯示，我們不但可以生成高畫質的臉部化妝圖像，並且我們的攻擊結果在人臉識別系統中具有很高的攻擊成功率。
Machine Learning has developed rapidly, and has achieve great success in computer vision and natural language processing. Many machine learning technologies are used in human daily life, such as self-driving car and face recognition system. Nowadays, human are really reliance on deep neural networks (DNNs), and if the DNNs has been attacked it will cause terrible results. In order to show the vulnerable of DNNs we propose a method based on Generative Adversarial Networks (GANs) to generate face makeup image that can fooling face recognition system. We hide the perturbation information of attack in the results of the makeup photos that undetectable to human. The experiment results show that we can generate high quality face makeup image and our attack results have high success rate on face recognition system
利用生成對抗網路產生音樂節奏遊戲譜面;Generation of Music Game Beatmap via Generative Adversarial Network
No full text[[abstract]]音樂節奏遊戲是目前非常熱門的遊戲，但設計遊戲譜面過於耗時，此外，現存生成譜面的方法也有許多限制，於是我們提出了基於生成對抗網路的方法產生音樂節奏遊戲譜面。首先，音樂被分離為人聲與配樂兩部分，這使我們產生的譜面更貼近真實的譜面。我們的模型包含兩種生成對抗網路的概念：為了加入音樂資訊而採用的Conditional Generative Adversarial Nets (CGANs)以及使模型更好收斂的Improved Wasserstein GAN (WGAN-GP)。我們比較了此方法以及其他條件下的實驗。我們也設計了問卷來衡量此方法的實驗結果，而此方法的實驗結果與真實的譜面得到的分數非常相近，也就是說，我們的實驗結果與真實譜面非常相似。
Music games are very popular now, but designing beatmaps usually takes too much time. In addition, there is some limitation of existing methods to generate beatmaps. In this thesis, beatmap generation method is proposed based on Generative Adversarial Networks (GANs). Audio is firstly separated into the vocal and instrument parts to make this method close to beatmap design philosophy of designers. Our model consists of Conditional Generative Adversarial Nets (CGANs) and Improved Wasserstein GAN (WGAN-GP) for considering audio information and fast convergency of model training. Our results are compared with different methods. Besides, we conduct a subjective evaluation of our results and the real beatmaps. Our results are very competitive to the real beatmaps which means our results are close to the real beatmaps
行動邊緣計算於直播串流之快取管理;Live Stream Cache Management in Mobile Edge Computing
No full text[[abstract]]基於4G網路的發展及智慧型行動裝置的普及，使用者所使用的功能十分多元其中以視訊串流為大宗。根據CISCO [1] 的分析，到2021年視訊串流將占據網路流量高達82% 成為熱門的網路應用之一。視訊串流可分為直播及點播，通常都是在網際網路上透過內容傳遞網路進行傳遞，但在行動網路上由於其核心網路為各家電信業者所打造的，因此無法在上面佈建內容傳遞網路。因此需要從網際網路上取得視訊內容導入核心網路在轉送給使用者，會造成相當大的延遲及對核心網路的傳輸造成負擔。ETSI為了應對上述情形提出了行動邊緣計算，透過高度分散式的計算環境結合雲端運算與網路功能虛擬化將運算與儲存資源佈建於行動網路的邊緣。使用者便可以就近使用運算資源，以降低核心網路的負擔。由於網路傳輸會隨著當前環境而變化，直播串流使用MPEG-DASH作為編碼，再加上為了提高使用者體驗品質，因此提高使用者體驗品質的MEC快取機制顯得十分重要。本篇論文提出基於行動邊緣計算的視訊直播應用行動網路架構，透過SDN技術的佈建將使用者發出的請求導向視訊直播快取應用，在根據各視訊直播快取應用的快取需求進行群播。透過使用者體驗品質評估快取機制效能分析其在不同情境下的效能，將本論文所提出的快取機制以使用者體驗品質將其與popularity-based的快取機制進行比較，可以得知在不同情境的比較下，本論文所提出之快取機制效能皆優於popularity-based的快取機制，可以有效提高視訊串流觀看者的使用者體驗品質。
Based on the development of 4G networks and the popularization of smart mobile devices, the function used by users are very diverse.Among them,the numbers of video streaming are quite huge. According to the analysis of CISCO [1], video streaming will account for up to 82% of network traffic by 2021 becoming one of the popular applications. Video streaming can be divided into live broadcast and video on-demand. It’s usually delivered over the Internet through a content delivery network. However, on the mobile network, because its core network is built by various telecom operators, it is impossible to build a content delivery network on it. Therefore, it is necessary to obtain video content from the Internet and import it into the core network and forward it to the user,which causes considerable delay and burdens on the transmission of the core network. In response to the above situation, ETSI proposes mobile edge computing to deploy computing and storage resources at the edge of mobile networks through a highly decentralized computing environment combined with cloud computing and network function virtualization. Users can use computing resources nearby to reduce the burden on the core network. Since the network transmission changes with the current environment, live streaming uses MPEG-DASH for encoded.In order to improve the user experience quality, MEC cache mechanism that improves the user experience quality is very important. This paper proposes a mobile video application mobile network architecture based on mobile edge computing. Through the deployment of SDN technology, the request sent by the user is directed to the live streaming cache application, The broadcast is performed according to the cache requirement of each video live cache application. Analyze its performance in different situation through the user experience quality assessment cache mechanism performance.Use user experience quality to compare the cache mechanism proposed in this paper with the popularity-based cache mechanism. As can be seen that under different situation, the performance of the cache mechanism proposed in this paper is superior to the popularity-based cache mechanism, which can effectively improve the user experience quality of video stream viewers
加速以序列比對為基礎之三代定序錯誤校正法;Acceleration of Alignment-based Error Correction for Third-generation Sequencing
No full text[[abstract]]第三代定序技術所定序出來的序列較長，而且定序速度較快，但是序列的錯誤率也比較高。因此在基因體組裝之前，通常會將序列上之定序錯誤進行校。其中一種錯誤校正策略是以序列比對為基礎之方法，此方法基於較耗時之動態規劃(Dynamic Programming)技巧來做序列比對。本篇論文中使用一位元平行化(Bit Parallelism)演算法來加速動態規劃，並與傳統之頻帶動態規劃(Banded DP)加速進行比較．此外，我們亦調整該位元平行演算法以適應三代定序特有之錯誤型態。實驗結果指出，採用位元平行化之動態規劃，確實比傳統頻帶動態規劃速度更快，但是校正準確率卻些微下降。進一步觀察顯示位元平行化動態規劃，在連續重複區間(Tandem Repeats)經常校正失敗，需要為此區域設計特殊方法才能改善。
The 3rd generation sequencing can produce long reads with fast turnaround time yet also with high error rate. Consequently, errors on the sequencing reads are usually corrected before genome assembly. One of the strategies of error correction is alignment-based method, which requires time-consuming alignment among reads based on dynamic programming (DP). In this thesis, we implement a bit-parallelism algorithm to accelerate DP and compare with traditional banded DP speedup. In addition, the bit-parallelism algorithm is fine tuned for correcting errors specific in third-generation sequencing. The results showed that, though bit-parallelism DP is faster than banded DP, the accuracy is unexpectedly decreased. Further investigation indicated that bit-parallelism DP performs worse in tandem repeat regions, which requires specific algorithms for better accuracy
支援高速移動之重構邊緣佈署EPC;Refactoring Edge-based EPC for High Speed Mobility
No full text[[abstract]]LTE (Long Term Evolution)為目前普遍使用的行動網路技術，它的架構可以分為RAN (Radio Access Network)和EPC (Evolved Packet Core)。近年來，從LTE到LTE-Advanced，主要的發展在於強化RAN的部分，而EPC的部分並沒有明顯的修正。目前EPC的設計是佈署元件在核心網路，用戶設備(User Equipment; UE)的流量傳輸都會匯集到這些元件，在面對未來日益增加的上網裝置，這會造成網路壅塞與效能不彰。一種分散式的解決網路負擔方式是將需要的EPC的功能元件佈署在靠近基站的邊緣端(稱為Edge-based EPC)，以分散核心網路的流量負擔，藉此讓UE可以獲得延遲較短、速度較快的網路服務。目前有些研究是直接使用目前的EPC元件進行邊緣佈署的設計，然而這會造成一些問題，當UE在邊緣網路間移動的話，會面臨這些服務邊緣元件的切換，這個切換的過程涉及許多元件的訊息交換與重新建立連線，這會影響UE的使用體驗。代表目前的EPC架構不適合進行邊緣佈署的探討，因此，我們首先整理各元件在LTE/EPC流程中主要使用到的功能，然後討論在流程中功能序列的對應字串，並列出可能的功能組合，接著我們利用三個指標評估可能的功能組合帶來的效益，包含MER (Message exchange reduction)、MHN (Message handling number)和SSE (Scaling side effect)。根據評估的結果，我們得到一個Refactoring EPC (R-EPC)架構。接著再根據R-EPC元件負責的功能，以決定元件在網路中的佈署位置，我們將此架構稱為edge-based refactoring EPC (abbreviated to E-R-EPC)，在E-R-EPC中，我們將UE常使用的服務觸發流程元件與負責流量傳輸的gateway佈署在邊緣網路，而記錄UE位置的元件則佈署於核心網路，如此一來UE在邊緣網路能使用低延遲流量傳輸服務，若是在邊緣網路間移動時，由於不會更換記錄位置的元件，所以能減少需要更新位置的訊息成本。此外，我們還探討UE在邊緣網路間移動，面臨Session Anchor (PGW)更換並保持session連續性的議題，並設計了一套新的handover過程，我們稱為inter-edge handover。在這個過程中，我們設計提前幫助UE取得新的邊緣網路IP，並且利用封包封裝的機制將session導入至下一個UE移動到的邊緣網路底下，以達到session不會中斷的設計。接著，我們還將E-R-EPC與其他架構進行signaling cost與queuing delay的效能比較，並呈現不錯的表現，尤其是在應對UE於邊緣網路間頻繁地移動情況(例如: 高速移動)。另外，我們評估了gateway元件佈署在邊緣的效益，結果顯示此設計可以有效地降低傳遞data traffic需要的時間，提供UE更快速的服務體驗，證明了我們的設計可以作為基於邊緣網路架構設計的參考。
LTE (Long Term Evolution) is the commonly used mobile network technology, and its architecture can be divided into RAN (Radio Access Network) and EPC (Evolved Packet Core). In recent years, the main development from LTE to LTE-Advanced is to strengthen the RAN technology, but the EPC has no significant modifications. The current design of EPC is to deploy the EPC components in a core network, and the data traffic from UEs (User Equipments) is converged to these components. It will cause network congestion and inefficiencies with the increasing number of internet devices in the future.One distributed way to solve the network burden is to deploy the on demand EPC components close to the edge of network (near the base station; known as Edge-based EPC). In this way, the traffic load at the core network can be distributed, thereby UE will obtain better quality of Internet service.At present, some researches directly use the the current EPC components for edge deployment. However, it causes some problems. When an UE moves between two edge networks, it will face the inter-edge procedures to change of these serving edge components. The procedure involves many number of message exchanges between components for the re-establishment of the connection, which will affect the UE's experience. It represents that the current EPC architecture is not suitable for the edge deployment. Therefore, we start by sorting out the functions that each component used in the LTE/EPC procedures, and then to list the corresponding strings for the sequence of functions in all the procedures. From the corresponding strings, we evaluate the benefits of the possible merged functions by three indicators, including MER (Message exchange reduction), MHN (Message handling number), and SSE (Scaling side effect). Based on the evaluation, we derive a Refactoring EPC (R-EPC) architecture. Then, we determine the deployment of the R-EPC components based on their responsibilties, and call this architecture edge-based refactoring EPC (abbreviated to E-R-EPC). In E-R-EPC, we deploy the commonly use of service triggering related components and the gateway responsible for data traffic to edge network, while deploy the components that record the location of UE to core network; thus, the UE can use the low-latency data traffic services in edge network, since the served recording location components are not changed, it can reduce the signaling cost to update the location information. In addition, we discuss the issue of changing Session Anchor (PGW) and maintaining session continuity when the UE moves between two edge networks, and we propose a new handover procedure called inter-edge handover. In this procedure, we design to assist the UE to obtain the IP address for new edge network in advance, and use the mechanism of packet encapsulation to guide the onging session to the new edge network; therefore, the session will not be interrupted.Further, we compare the signaling cost and queuing delay of E-R-EPC with other architectures, the results of the proposed one show good performance, especially in dealing with frequent movements of UE between two edge networks (eg: high-speed mobility); Moreover, we estimate the effectiveness of deploying gateway to edge network, and we recognize that the design is successfully in reducing the time to deliver the data traffic, which can provide the fast services to UE. It proves E-R-EPC is a suitable reference for edge-designed networks
拓樸最佳化應用於戟齒輪加工機結構設計與開發;Structural Design and Development of bevel gear cutting machine using the Topology optimization method
No full text[[abstract]]機台進行輕量化拓樸設計時，往往需要不斷的嘗試及修正才能決定該鑄件所需之重量，此為相當耗時之過程。本研究將發展最佳配重之等校剛性曲線擬合技術，基於拓樸最佳化之密度函數法概念，鑄件結構拓樸設計區域之材料係數為設計變量，在機台結構細部設計前先進行最佳配重比分析，以預估結構靜剛性與自然模態頻率；各鑄件配重結果為拓樸最佳化分析之拘束條件，以此流程進行機台設計，各鑄件只需進行一次拓樸最佳化分析，將有效提升開發設計效率。首先對各鑄件進行不同質量分佈之拓樸最佳化分析，評估鑄件體積與剛性變化趨勢，對此趨勢進行曲線擬合定義出關係式後，將此關係式代入整機有限元素模型進行參數最佳化分析，該鑄件結構為實心，以各鑄件之體積定義為變數，整機模態頻率閃避機台加工頻率且同時達到靜剛性之需求為目標，透過此分析可以得到機台之最佳配重比，根據此配重比再進行拓樸最佳化分析，得到機台之最佳結構設計。研究結果將可直接嘉惠於國內工具機設計。
The required mass of casting for machine tool was determined by a lot of attempts and corrections under the lightweight topology optimization design. The above iterative analysis is a time-consuming process. This study will develop the technique of the curve-fitting between the equivalent stiffness and the optimal weight ratio. The design variables are the material property of the topological design area based on the density function method of topology optimization, i.e. density and young’s modulus. The analysis for the optimal weight ratio between each casting component is utilized to predict the static stiffness and the natural frequency before the detailed rib design of each casting. The results of weight arrangement between each component are the constraints of the topology optimization analysis. Following this process only needs one topology optimization analysis for each casting component. The design efficiency will be improved effectively.First, the topology optimization analyses under different weight arrangement for each casting are utilized to evaluate the variant trend between different volume and corresponding equivalent stiffness. The function between the volume and equivalent stiffness was defined by curve fitting. The function was put into the finite element modeling in order to execute the parametric optimization analysis. The internal structure of casting is solid. The variable is the volume of casting. The objective is to let the natural frequency of the machine tool against the processing frequency and achieve the required static stiffness. The optimal weight ratio of each casting was acquired to proceed topology optimization analysis of rib structure. The research result will directly benefit the domestic machine tool design
戟齒輪切削刀具研磨方法之研究;
No full text[[abstract]]本論文研究戟齒輪切削刀具之數學模式，參考格里森(Gleason)公司設計的面銑式刀盤系統設計刀具參數，其中刀刃型式主要分為條狀刀刃(Stick blade)與成形刀刃(Profile blade)，本論文研究其研磨方法，並且做分析及改善。本論文使用Mathematica軟體推導戟齒輪刀具數學模式，將磨刀座標系上圓弧刃口線之固定位置切點作為參考點，壓力角由刃口線之切線與垂直軸夾角定義，經由計算轉換至切削座標系並求得實際刃口線，將實際刃口線投影至切削方向之平面上，計算參考點上切線與垂直軸之夾角為切削座標系上之壓力角，由於切削座標系和磨刀座標系之壓力角會因為經計算過後而有差異，因此本論文推導兩者間之關係式，以便未來設計刀具時在切削座標系上定義壓力角參數。戟齒輪條狀刀具研磨方法以兩面研磨及三面研磨為主，由於此兩種方法研磨前需將條狀刀具從刀盤上拆卸，並安裝在磨刀專用機上進行研磨，本論文研究條狀刀具單面研磨方法，此方法可應用在戟齒輪切齒機進行線上重磨動作，節省組裝及校正時間，本論文研究其數學模式，計算重磨前後之刃口線形及位置是否一致，以及探討刀刃切削傾角(Hook angle)是否合乎需求，並在加工模擬軟體上進行應用。本論文建立磨刀與切齒專用機台三維模型及加工座標系統，利用VERICUT進行加工模擬，規劃修砂及研磨路徑並導入加工模擬軟體內，觀察加工過程是否發生干涉現象，驗證研磨方法之可行性，最後分析及探討研磨方法產生的問題，並提出改善辦法。
The aim of this thesis attempts to explore the mathematical model of hypoid gear, which reference to the cutter design parameters of face milling system of Gleason. The blade type is mainly divided into Stick blade and Profile blade. This research involved the grinding method, comprised of analysis and improvement.In this thesis, we use Mathematica to derive the mathematical model of the hypoid gear cutter. Taking the fixed position of the cutting edge line on the sharpener coordinate system as the reference point. The pressure angle is defined by the angle between the tangent to the cutting edge line and the vertical axis. Transfer to the cutting coordinate system through calculation and find the actual cutting edge line. To project the actual edge line onto the plane of the cutting direction. Calculate the angle between the tangent and the vertical axis at the reference point as the pressure angle on the cutting coordinate system. Due to the pressure angle of the cutting coordinate system and the sharpening coordinate system will be different after calculation. The research centers on an empirical study deriving the relationship from the pressure angles on the different coordinate system, that in order to define the pressure angle parameter on the cutting coordinate system when designing the blade in the future.The differentiation of grinding method of stick blade into two-face and three-face grinding. Since the methods need to remove the blade from the cutter before grinding, and install it on the machine for sharpening. The research centers on an empirical study involving single-face grinding from grinding method of blades that it can be applied on the resharpening action of the gear cutting machine to save processing time. The analysis of mechanical process is conducted through shape of the resharpening edge and hook angle of the blade. Finally, we apply on the software program of simulating machining in order to indicate feasibility of the grinding method.To conclude, this thesis builds a three-dimensional model and machining coordinate system for grinding and cutting machines, and uses VERICUT for machining simulation, which plans the machining path and imports it into the software program of simulating machining to observe whether the interference occurs during the machining process and verify the feasibility of the grinding method
人類安全下之人口販運研究 ─以東南亞移工在臺灣之處境為例;A Research of the Human Security Framework for Human Trafficking—The Situation of Southeast Asia Migrant Workers in Taiwan
No full text[[abstract]]人類安全強調以人為中心，重視個人並以人類福祉為重，具有普世性、互相聯繫性、預防勝於解決、以人為中心四項特點。臺灣自1989年開放外籍移工來臺後，人數直線上升，至2017年已破67萬人，而行蹤不明的外勞人數也逐年上升中。本研究探討其中東南亞移工遭人口販運與剝削之情形，以此對比臺灣人類安全發展不足之處。本研究首先蒐集人口販運相關文獻與國際組織統計資料，對當前國際上與臺灣的人口販運情況進行初步分析，並結合深度訪談仲介業者與外籍移工，依據其提供的實務經驗輔以政府數據，聚焦於討論外籍移工在臺灣艱難處境，以及被剝削之各種可能成因，歸納出幾項建議，以供政府未來研擬政策之參考。
Human security emphasizes people-centeredness, values individuals and human well-being. There are four characteristics of human security, universal, interconnected, prevention is better than solution, and people-centric. Taiwan implemented Southeast Asia migrant workers can coming to Taiwan policy in 1989, the number of migrant workers increased. In 2017, there were over than 670,000 people. However, the number of migrant workers with unknown whereabouts has gradually increased.This study explores the situation of trafficking and exploitation of Southeast Asia migrant workers in Taiwan, and compares the lack of human security development in Taiwan. First step, the study collects relevant literature on human trafficking and statistics of international organizations, and conducts a preliminary analysis of the current international trafficking situation in Taiwan. Combining interviews in some depth with agent and migrant workers, based on the practical experience provided by the agency and government data, focusing on discussing the difficult situation of migrant workers in Taiwan and the various possible causes of exploitation.Finally, there were several recommendations are summarized to provide a reference for the government research policy in the future
運用監督式學習技術建構植牙手術成敗之預測模型;Constructing the Forecasting Models for Dental Implant Results with a Supervised Learning Approach
No full text[[abstract]]假牙?復治療一直是牙科治療針對患者缺牙重建相當重要的一環，隨著時代變遷及科技進步，以鈦金屬牙根取代傳統式假牙製作的人工植牙手術更是成為全口全建的首選。但是因為植牙手術牽涉範圍相當廣泛，舉凡病人身體狀況、手術醫師個人經驗及植牙系統的選擇等，都是治療必須考量之項目。只是因為植牙治療金額相較於傳統式假牙高出許多，讓許多醫療人員趨之若鶩，但卻未仔細分析手術未來影響因素而導致醫療糾紛的發生。 目標：在過去文獻中探討植體成敗因素多數僅就患者系統性疾病、手術方式及?復型式等單一因素進行統計學上相關顯著分析;本研究將建立預警機制以減少失敗機會，且進一步納入醫師個別因素及?復體固定方式。 方法：本研究收集嘉義某區域教學醫院接受人工植牙治療患者臨床資料，共有患者年齡、性別、缺牙原因、系統性疾病、抽煙、喝酒、嚼檳榔、醫師科別、醫師執業年數、植牙位置、骨頭硬度、骨脊增生術、上顎竇增高術、植體系統、植體長度、植體寬度、?復物形式、支台體角度及義齒固定方式，總計八大類別20個變項，再以決策樹、支援向量機及邏輯斯迴歸等監督式學習技術分成植體早期成敗預測及晚期成敗預測兩個面向進行分析。研究結果：以單一分類器比較預測率方面，不論植體早期成敗或是晚期成敗方面，C4.5(J48)決策樹預測效能最高，分別為62.6%及69.8%；並從分析中發現植體系統不分植體早期成敗及晚期成敗皆有影響。另外又以CART分類迴歸樹建立早期及晚期風險預測模型，建立影響植體早期及晚期成功失敗的關聯規則。結論：本研究希望能藉由研究之結果建立植體早期及晚期成敗預測模型，幫助臨床醫師遇到臨床相關情形時，可以選用預測結果較佳之植體系統及?復治療，以減少治療失敗發生糾紛之目的。
Prosthodontic treatment hasalways been a major part of dental treatment for rehabilitation of patient with edentulous ridge. The traditional dentures were replaced with dental implant surgery insertion due to scientific and technological progressing. Dental implant surgery involves range to be quite extensive, include physical condition of patient, personal habit, surgical experience , and the choice of implant system etc, so all of the projects that must be considered. Because charge of dental implant is better than other treatment , many dentists preferred to perform implant surgery without thinking about the complications which cause the emergence of the medical dispute. Purpose:In the past literature, most of the exploratory factors were statistically significant only for single disease such as systemic disease, surgical procedure and prosthetic type. We wants toestablish early warning mechanism including factors of dentist and fixation of denture to reduce the chance of failure.Methods:We collected clinical data of patients undergoing dental implant treatment in a local hospital of Chiayi, including age, sex, causes of tooth loss, systemic diseases, smoking, drinking, chewing betel nuts, department of dental division, physician practice years, implant position , Bone density, bony augmentation, maxillary sinus augmentation, implant systems, implant length, implant width, prosthetic form, the angle of the abutment and fixation of denture.A total of eight categories of 20 variables, Decision tree, support vector machine and logistic regression were used to analyze the early success or failure prediction and the late success or failure prediction. Results: C4.5 (J48) decision tree had the highest predictive efficiency of 62.6% and 69.8%respectively, regardless of the early or late successes and failures in a single classifier.In addition, the early and late risk prediction models were established based on the classification regression tree (CART) , and the association rules were established to affect the early and late success of the implant. Conclusion:Ｗe hope to establish the prediction models of early and late success or failure of the implant by using the results of the study, and help the clinicians to choose the optimal implant system and the prosthetic therapy to reduce failure of treatment
