教学课程资源库 更多>>

关键词： Artificial intelligence   Information technology   Computer science  

摘要： Co-creative collaboration is a creative collaboration of two or more agents working together on a shared creative product. Effective co-creative collaboration is a combination of interactions and contributions of the task collaborators. In human-human collaboration, gestures, verbal communications, and emotional responses are among the general communication strategies that shape the interactions between the collaborators and enable negotiation of the contributions. Emotional feedback allows human collaborators to passively communicate their stance about the experience and convey their perception of the process without distracting the flow of the task. In human-human co-creative collaboration, participants interact and contribute to the task based on their perception of the collaboration over time. However, perceiving the cognitive state of the user to determine the dynamics of collaboration and decide what the agent should contribute to the artifact are two primary challenges of building effective co-creative Artificial Intelligence systems (Abdellahi et all, 2020). In response to these two challenges, the following thesis statement is presented: Using knowledge of human emotion in human-AI co-creative collaboration can improve the user satisfaction of the collaboration experience and quality of the collaboration outcome. This thesis focuses on establishing dynamics of co-creative collaboration between a human and a co-creative AI agent through an emotion-based interaction model for co-creativity. After the introduction in chapter 1 and reviewing the related background in chapter 2, the 3rd chapter of this dissertation presents an interaction model for Arny V1, a co-creative system designed to explore the research questions of this thesis. Arny V1 interaction model was studied as part of an exploratory study with the Wizard of Oz setup that is discussed in chapter 4 of this document. The modified version of the model, Arny V2, was then deployed and analyzed to confi

关键词： Deep learning   Optimization   Neural networks   Artificial intelligence   Computer science  

摘要： Machine learning using deep neural networks -- deep learning -- has been extremely successful at learning solutions to a very broad suite of difficult problems across a wide range of domains spanning computer vision, game play, natural language processing and understanding, and even fundamental science. Despite this success, we still do not have a detailed, predictive understanding of how deep neural networks work, and what makes them so effective at learning and generalization. In this thesis we study the loss landscapes of deep neural networks using the lens of high-dimensional geometry. We approach the problem of understanding deep neural networks experimentally, similarly to the methods used in the natural sciences. We first discuss a phenomenological approach to modeling the large scale structure of deep neural network loss landscapes using high-dimensional geometry. Using this model, we then continue to investigate the diversity of functions neural networks learn and how it relates to the underlying geometric structure of the solution manifold. We focus on deep ensembles, robustness, and on approximate Bayesian techniques. Finally, we switch gears and investigate the role of nonlinearity in deep learning. We study deep neural networks within the Neural Tangent Kernel framework and empirically establish the role of nonlinearity for the training dynamics of finite-size networks. Using the concept of the nonlinear advantage, we empirically demonstrate the importance of nonlinearity in the very early phases of training, and its waning role farther into optimization.

关键词： Computer science   Human mechanics   Cameras   Accuracy   Deep learning   Datasets   Neural networks   Legs   Anger   Feature selection   Hands   Methods   Algorithms   Walking   Emotions   Animation   Artificial intelligence   Information technology  

摘要： This research project develops a new deep neural network model for real-time human movement prediction based on human habits and emotion through the layering of neural networks, computer vision algorithms, and mathematical matching with a dynamic database. By combining multiple residual neural networks using different layering algorithms, the new model can increase prediction accuracies with reduced errors because of how each neural network adjusts to fill in the gaps left behind be each other to average out a proper evaluation of human motion. Specifically, the new model contains the following components: ResNet50 and altered ResNet34 on ImageNet for motion in all three dimensions, FURIA algorithm on Dlib 68 facial landmarks for emotion, and a nearest neighbor neural network for prediction based off locomotion. The result of this combination was that the prediction would take 0.05 milliseconds after emotion is selected to initiate its prediction path. The prediction module follows motion at first but with reduced accuracy until emotion is given to give a proper prediction with visualized projection. As its database of motions grows, the accuracy grows as well which leads to near real-time movement prediction. In the experiments, the new model outperforms existing models in both prediction accuracy and training speed due to this dynamic database. While other models require retraining of the neural networks to adjust to new testing data, this model relies on only database motion additions which greatly speeds up overall training and testing. The possible usages of this model are, but not limited to, health caregiving, culprit movements during police engagements, and child safety monitoring.

关键词： Mathematics   Quantum physics   Research   Commuting   Collaboration   Computer science   Noise   Complexity theory   Information theory   Games   Bias  

摘要： Quantum graph theory, also known as non-commutative graph theory, is an operator space generalization of graph theory. The independence number, and Lovasz theta function were generalized to this setting by Duan, Severini, and Winter [DSW13] and two different version of the chromatic number were introduced by Stahlke [Sta16] and Paulsen [HPP16]. We introduce two new generalizations of the chromatic number to non-commutative graphs and provide an upper bound on the parameter of Stahlke. We provide a generalization of the graph complement and show the chromatic number of the orthogonal complement of a non-commutative graph is bounded below by its theta number. We also provide a generalization of both Sabidussi's Theorem and Hedetniemi's conjecture to non-commutative graphs. The study of non-local games considers scenarios in which separated players collaborate to provide satisfying responses to questions given by a referee. The condition of separating players makes non-local games an excellent setting to gain insight into quantum phenomena such as entanglement and non-locality. Non-local games can also provide protocols known as self-tests. Self-testing allows an experimenter to interact classically with a black box quantum system and certify that a specific entangled state was present and a specific set of measurements were performed. The most studied self-test is the CHSH game which certifies the presence of a single EPR entangled state and the use of anti-commuting Pauli measurements. We introduce an algebraic generalization of CHSH and obtain a self-test for non-Pauli operators resolving an open question posed by Coladangelo and Stark (QIP 2017). Our games also provide a self-test for states other than the maximally entangled state, and hence resolves the open question posed by Cleve and Mittal (ICALP 2012). This thesis is primarily based on two collaborative works written by the author and several coauthors [CMMN20], [KM19].

关键词： Computer science   Artificial intelligence  

摘要： The Particle Swarm Optimization (PSO) is a typical, powerful and commonly used search paradigm in the metaheuristic family, and it has gained prominence because of its ability to tackle many complex optimization problems. Recent research works have shown that the enhanced methodologies for PSO could be applied to a much wider range of optimization applications. In this thesis, the essentials of equilibrium are analyzed based on the balanced search behaviors of diversification and intensification of each search agent from the whole swarm under the solution space. After that, three variants of equilibrium approaches based on the canonical PSO have been presented for tackling various kinds of optimization problems accordingly, namely the Clustering-based Variable sub-Window approach of Particle Swarm Optimization (CVW-PSO), the Leader and Long-Tailed-based Particle Swarm Optimization (LLTPSO) and the Group Theoretic Particle Swarm Optimization (GT-PSO). They are inspired by different theoretical ideas and proposed with corresponding mathematical foundations to support each one of them, i.e., statistical feature extraction and signal processing for CVW-PSO, Lévy flight of random walk for LLT-PSO and group theory for GT-PSO. They combine these theories with classic PSO respectively and formulize the behaviors of each particle and the entire swarm under the new framework to improve the original search mechanism. Their application scenarios include but not limited to the perspectives of data stream mining and monitoring, classification improvement and model reconstruction, combinatorial optimization problem with discrete features, etc. In order to validate and investigate the efforts of these PSO variants, the numerical experiments over various datasets are designed, conducted and compared with other temporal optimization algorithms for the purpose of the comprehensive study. The experimental results show the encouraging performances of those three variants and prove they

关键词： Electrical engineering   Computer science   Energy   Civil engineering  

摘要： Nowadays, the widespread use of distributed generators (DGs) raises significant challenges for the design, planning, and operation of power systems. To avoid the harm caused by excessive DGs, evaluating the reliability and sustainability of the system with high penetration of DGs is essential. The concept of hosting capacity (HC) is used to achieve this purpose. It is to assess the capability of a distribution grid to accommodate DGs without causing damage or updating facilities. To obtain the HC value, traditional HC analysis methods face many problems, including the computational difficulties caused by the large-scale simulations and calculations, lacking the considering temporal correlation from data to data, and the inefficient on real-time analysis. This paper proposes a machine learning-based method, the Spatial-Temporal Long Short-Term Memory (ST-LSTM), to overcome these drawbacks using the traditional HC analysis method. This method will significantly reduce the requirement of calculations and simulations, and obtain HC results in real-time. Using the time-series load profiles and the longest path method, ST-LSTMs can capture the temporal information and spatial information respectively. Moreover, compared with the basic Long Short-Term Memory (LSTM) model, this modified model will improve the performance in the HC analysis by some specific designs, which are the sensitivity gate to consider voltage sensitivity information, the dual forget gates to build spatial and temporal correlation.

关键词： Computer science   Artificial intelligence  

关键词： Computer science   Biomedical engineering   Bioinformatics   Artificial intelligence  

关键词： Agreements   COVID-19 vaccines   Subjectivity   Sentiment analysis   Support vector machines   Computer science   Web studies