教学课程资源库 更多>>

关键词： Computer science  

摘要： With the extensive use of biometrics for authenticating users, the need to ensure privacy of biometric data is greater than ever before. Biometric authentication systems are vulnerable to attacks and the loss of biometric data will lead to loss of privacy of an individual. Multibiometrics refers to the use of multiple biometric modalities simultaneously in order to perform matching. In this work, we introduce a multibiometric fusion technique which can be used to ensure that the original raw biometric data are unlikely to be compromised and, at the same time, recognition can be performed. The face and the iris biometric modalities are fused at the feature-level to produce discriminative embeddings that can be used for recognition. The original face or the iris cannot be retrieved from the combined representation, thus preserving the privacy of the individual. We present the results of this approach, provide analysis, discuss the challenges, and list possible future directions.

关键词： Computer science   Health care management  

摘要： The main aim of this thesis is to develop a library for integrating fully homomorphic encryption-based computations on a standard database. The fully homomorphic encryption is an encryption scheme that allows functions to be performed directly on encrypted data without the requirement of decrypting the data and yields the same results as if the functions were run on the plaintext. This implementation is a promising solution for preserving the privacy of the health care system, where millions of patients’ data are stored. The personal health care tools gather medical data and store it in a database. Upon importing this library into the database, the data that is being entered into the database is encrypted and the computations can be performed on the encrypted data without decrypting.

关键词： Computer science   Standard deviation   Simulation   Homeostasis   Bladder   Success   Mutation   Optimization   Neural networks   Sensors   Robots   Design   Light   Artificial intelligence   Robotics  

摘要： The ongoing deep learning revolution has allowed computers to outclass humans in various games and perceive features imperceptible to humans during classification tasks. Current machine learning techniques have clearly distinguished themselves in specialized tasks. However, we have yet to see robots capable of performing multiple tasks at an expert level. Most work in this field is focused on the development of more sophisticated learning algorithms for a robot’s controller given a largely static and presupposed robotic design. By focusing on the development of robotic bodies, rather than neural controllers, I have discovered that robots can be designed such that they overcome many of the current pitfalls encountered by neural controllers in multitask settings. Through this discovery, I also present novel metrics to explicitly measure the learning ability of a robotic design and its resistance to common problems such as catastrophic interference. Traditionally, the physical robot design requires human engineers to plan every aspect of the system, which is expensive and often relies on human intuition. In contrast, within the field of evolutionary robotics, evolutionary algorithms are used to automatically create optimized designs, however, such designs are often still limited in their ability to perform in a multitask setting. The metrics created and presented here give a novel path to automated design that allow evolved robots to synergize with their controller to improve the computational efficiency of their learning while overcoming catastrophic interference. Overall, this dissertation intimates the ability to automatically design robots that are more general purpose than current robots and that can perform various tasks while requiring less computation.

关键词： Agronomy   Computer science   Plant sciences  

摘要： Crop modeling is a process of translating quantitative knowledge on crop growth into a computer program that simulates the growth in silico. From a software engineering perspective, crop modeling has suffered from a legacy built decades ago when early crop models appeared. Many crop models have been developed in imperative programming approaches striving for high performance, but frequently suffered from error-prone code and technical debts left behind. In this study, we propose a new declarative modeling framework named Cropbox written in Julia programming language to support developing crop models in a concise form equipped with useful abstractions commonly required in modeling. With an insight that a crop model is essentially an integrated network of generalized state variables, the framework provides various primitives for representing variables and systems as well as functions essential to modeling workflow. The modeling workflow based on Cropbox was used to create and illustrate its applications in crop modeling at different levels of organization and complexity. The first application was phenology modeling for building an ensemble model to predict flowering time based on various existing approaches. Extracting common patterns in the models and developing reusable interface for simulation, visualization, and calibration motivated an idea of a new modeling framework. The second application was a coupled gas-exchange model which combines two models for biochemical photosynthesis and empirical stomatal conductance with an additional link to an energy balance equation. The model implemented in Cropbox framework provided the same functionality as an existing model written in C++ with less code and a more flexible interface in terms of parameter management and output visualization. To demonstrate the capability of the model, we evaluated two stomatal conductance modeling approaches and applied them to replicate the observed behavior of transgenic plants from the lit

关键词： Computer science   Artificial intelligence   Robotics  

摘要： Co-robots are robots that cooperate with their human partners to accomplish a task. For co-robots to have a widespread use, they must co-exist and operate safely in human dominated natural environments without colliding with anyone or any obstacle. This requires the co-robots to visually detect people using cameras or other sensors and plan its own path with a new level of intelligence. This research proposes a visual based solution using a Bayesian classifier to predict people’s motion and using a genetic algorithm to plan the path of the co-robot in a crowd while fulfilling the two major tasks—maintaining proper distance from the human partner and people in the field of view while also following the human partner. A three-dimensional model with a two-dimensional plane and one-dimensional time is used for the modeling. The information about the crowd was obtained through video and image sequence. A Bayesian based motion prediction algorithm was implemented for the prediction of people movements, based on which an optimal path for the co-robot was determined using a genetic algorithm. A Windows Presentation Foundation (WPF) application software system was developed to test the implemented solution in four different test environments—no obstacle, static obstacle, moving pedestrian, and real scene simulation. The optimal co-robot paths from the genetic algorithm were recorded, analyzed, and displayed in animation in the WPF application. The results showed that this solution is highly effective in fulfilling the tasks of partner-following (scores 97%-99%) and human collision avoidance (93%-99%). They also showed that the solution is effective in avoiding static obstacle (74%-86%) and keeping comfortable distance from the people in the scene (72%-98%). Overall, the results showed that this hybrid solution using Bayesian based motion prediction and genetic path planning can be an effective solution for co-robot path planning problem.

关键词： Robotics   Mechanical engineering   Industrial engineering   Computer science  

摘要： The number of robotic systems in the world is growing rapidly. However, most industrial robots are isolated in caged environments for the safety of users. There is an urgent need for human-in-the-loop collaborative robotic systems since robots are very good at performing precise and repetitive tasks but lack the cognitive ability and soft skills of humans. To fill this need, a key challenge is how to enable a robot to interpret its human co-worker's motion and intention. This research addresses this challenge by developing a collaborative human-robot interface via innovations in computer vision, robotics, and system integration techniques. Specifically, this work integrates a holistic framework of cameras, motion sensors, and a 7-degree-of-freedom robotic manipulator controlled by vision data processing and motion planning algorithms implemented in the open-source robotics middleware Robot Operating System (ROS).

关键词： Computer science   Datasets   Neural networks   Bias   Classification   Artificial intelligence  

摘要： Many machine learning applications require a high degree of confidence in the correctness of the classification results. Ideally, no decision made by the classification model should be a misclassification. However, no classifier can guarantee to provide a correct classification for all data items. Therefore, it is reasonable to allow the classification model to classify a data item only when the certainty level is high and reject to make a decision if the certainty level is not *** this dissertation, a high confidence data classification method is developed. The method helps the classification model avoid classifying data items that are difficult to classify, which allows reducing the classification error to as low as zero and increasing the confidence in the classification results. In addition, using the new method, new neural network and decision tree classifiers and their training algorithms are developed. These new classifiers provide an extra output to indicate that the data item cannot be classified into a predefined class and is a reject. The claim that the method empowers the classification model to learn to determine the set of difficult to classify data samples automatically during the training process is validated through the classification experiments of diverse datasets. All experiments have evaluated the classifiers and algorithms presented in this dissertation on achieving the objective of the new method. The research presented in this dissertation has brought us one step closer to achieving near zero classification error for increasing our confidence in the correctness of the prediction made by the classification model.

关键词： Artificial intelligence   Remote sensing   Geographic information science   Computer science  

摘要： Many regions around the world suffer from a lack of authoritatively-collected data on factors critical to understanding human well-being. This challenges our ability to understand the progress society is making towards reducing poverty, improving lifespans, or otherwise improving livelihoods. A growing body of research is exploring how deep learning algorithms can be used to produce novel estimates of sparse development data, and how access to such data can impact development efforts. This dissertation contributes to this literature in three parts. First, using Landsat 8 satellite imagery and data from the Armed Conflict Location & Event Data Project, convolutional neural networks are trained to predict locations where conflict is likely to result in fatalities for one year. Second, building on the findings in chapter 1, this dissertation explores the potential to extend predictions to a time series using both yearly and six month intervals. Finally, chapter 3 introduces GeoQuery, a dynamic web application which utilizes a High Performance Computing cluster and novel parallel geospatial data processing methods to overcome challenges associated with integrating, and distributing geospatial data within research communities.

关键词： Artificial intelligence   Computer science   Bioinformatics   Analytical chemistry   Biochemistry  

摘要： Identifying specific cell types is a significant step for studying diseases and potentially leading to better diagnosis, drug discovery, and prognosis. High-throughput single-cell RNA-Seq (scRNA-seq) technologies have advanced in recent years, enabling researchers to investigate cells individually and understand their biological mechanisms. Computational techniques such as clustering, which are categorized in the form of unsupervised learning methods, are the most suitable approach in scRNA-seq data analysis when the cell types have not been characterized. These techniques can be used to identify a group of genes that belong to a specific cell type based on their similar gene expression patterns. However, due to the sparsity and high-dimensional nature of scRNA-seq data, classical clustering methods are not efficient. Therefore, the use of non-linear dimensionality reduction techniques to improve clustering results is crucial. We introduce a pipeline to identify representative clusters of different cell types by combining non-linear dimensionality reduction techniques such as modified locally linear embedding (MLLE) and clustering algorithms. We assess the impact of different dimensionality reduction techniques combined with the clustering of thirteen publicly available scRNA-seq datasets of different tissues, sizes, and technologies. We evaluate the intra- and inter-cluster performance based on the Silhouette score before performing a biological assessment. We further performed gene enrichment analysis across biological databases to evaluate the proposed method's performance. As such, our results show that MLLE combined with independent component analysis yields overall the best performance relative to the existing unsupervised methods across different experiments.

关键词： Internet of Things   User interface   Unmanned aerial vehicles   Software   Sensors   Aerospace engineering   Computer science   Engineering   Robotics   Transportation  

摘要： A new application for Unmanned Aerial Vehicles comes to light daily to solve some of modern society’s problems. One of the mentioned predicaments is the possibility for optimization in agricultural processes. Due to this, a new area arose in the last years of the twentieth century, and it is in constant progression called Precision Farming. Nowadays, a division of this field growth is relative to Unmanned Aerial Vehicles applications. Most traditional methods employed by farmers are ineffective and do not aid in the progression and solution of these issues. However, there are some fields that have the possibility to enhance many agriculture methods, such fields are Cyber-Physical Systems and Cloud Computing. Given its capabilities like aerial surveillance and mapping, Cyber- Physical Systems like Unmanned Aerial Vehicles are being used to monitor vast crops, to gather insightful data thatwould take a lot more time if being collected by hand. However, these systems typically lack computing power and storage capacity, meaning that much of its gathered data cannot be stored and further analyzed locally. That is the obstacle that Cloud Computing can solve. With the possibility to offload computing power by sending the collected data to a cloud, it is possible to leverage the enormous computing power and storage capabilities of remote data-centers to gather and analyze these datasets. This dissertation proposes an architecture for this use case by leveraging the advantages of Cloud Computing to aid the obstacles of Unmanned Aerial Vehicles. Moreover, this dissertation is a collaboration with an on-going Horizon 2020 European project that deals with precision farming and agriculture enhanced by Cyber-Physical Systems.