教学课程资源库 更多>>

关键词： Computer science   Research   Internships   Photographs   Datasets   Experiments   Adaptation   Methods   Advisors   Ablation   Semantics   Editing   Painting   Artificial intelligence  

关键词： Artificial intelligence   Robotics   Computer science  

摘要： The biological paradigm of learning by trial and error has motivated tremendous success in the field of Machine Learning. While Reinforcement Learning (RL) rests behind a myriad of breakthroughs, its practical application to real-world scenarios remains an open question. This thesis addresses the three challenges of restricted scalability, reduced robustness and limited practical viability through the lens of hierarchies serving as abstractions of composite behavior. Novel evolutionary RL methods present an evolving hierarchy which provisions scalability among members of its population. Novel energy based RL schemes, on the other hand, minimize surprise utilizing low energy configurations among members of the multi agent hierarchy. The framework of energy-based surprise minimization steers practical application of hierarchical RL to the setting of trade execution. The end result of this study is a hierarchical scheme demonstrating trade patterns analogous to humans with this thesis serving as a motivation for application of RL to practical problems.

摘要： Exposing students to computing activities, especially in younger grades, increases the likelihood that students will develop an interest in computer science (CS). This interest may translate into a heightened ability that carries over into elective middle and high school courses, selecting a college major, and eventually a career. Upper elementary students (typically ages 9 to 11) are at a critical time in developing their interests and in receiving messages that inform how they feel about their abilities. Expectancy-value theory makes use of Bandura’s self-efficacy theory—the belief a student has in his or her ability to complete a task successfully—and the notion of outcome expectancy—students’ belief that their behaviors will result in a desired outcome. Together, these help students form proximal and distal goals. These distal goals, be they future academic choices or career options, are influenced by students’ expectations of success and how much they value, or show interest in, the task. Assessing young students’ CS interests, through an attitudinal survey focused on self-efficacy and outcome expectancy items, is relatively novel. Students’ perceptions of themselves—including whether they belong in a class, a major, or a field—and their abilities likely affects to what extent they are successful in regulating their learning in that domain. High self-efficacy students persist when challenges arise and they utilize more effective strategies while learning; moreover, they tend to modify their goals and strategy use through feedback and engage in adaptive help seeking. Because learning is a social process, it is also important to explore how student self-efficacy and outcome expectancy interacts with the dynamics of academic collaboration. This is a particularly interesting area to study in CS education due to the growing popularity of collaborative (pair) programming. As an analytic approach, students' discourse as they problem solve is a rich source of data on c

关键词： Neurons   Neural networks   Maps   Algorithms   Human subjects   Artificial intelligence   Computer science  

摘要： Machine learning is being applied in various critical applications like healthcare. In order to be able to trust a machine learning model and to repair it once it malfunctions, it is important to be able to interpret its decision-making. For example, if a model's performance is poor on a specific subgroup (gender, race, etc), it is important to find out why and fix it. In this thesis, we examine the drawbacks of existing interpretability methods and introduce new ML interpretability algorithms that are designed to tackle some of the shortcomings. Data is the labor that trains machine learning models. It is not possible to interpret an ML model's behavior without going back to the data that trained it in the first place. A fundamental challenge is how to quantify the contribution of each source of data to the model's performance. For example, in healthcare and consumer markets, it has been suggested that individuals should be compensated for the data that they generate, but it is not clear what is an equitable valuation for individual datum. In this thesis, we discuss principled frameworks for equitable valuation of data; that is, given a learning algorithm and a performance metric that quantifies the performance of the resulting model, we try to find the contribution of individual datum. This thesis is divided in 3 sections, machine learning interpretability and fairness, data valuation, and machine learning for healthcare - all linked by the common goal of making the use of machine learning more responsible for the benefit of human beings.

关键词： Artificial intelligence   Computer science  

摘要： Low dimensional embeddings of nodes in graphs using Graph Neural Networks (GNNs) have proved extremely useful in various prediction tasks, from content recommendations to identifying protein-protein interactions. However, Graph Neural Networks have mostly been explored only on homogeneous graphs. A heterogeneous information network (HIN) is a network that consists of different types of nodes in their own feature spaces with different semantic relations between them. These heterogeneous rich semantic properties bring significant challenges in learning low-dimensional embeddings of a node in a HIN. A new GNN model is needed to address these issues, tailored to address a heterogeneous graph's specifics. We introduce Heterogeneous Graph Variational Auto-Encoders to address these issues that maintain uncertainty in node representation and the attention coefficients, which help in learning representations based on multiple contexts and modeling graphs with implicit temporal evolution. We report link-prediction results for our models across three graph data sets and note significant improvements over the current state of the art.

关键词： Computer science   Neural networks  

摘要： Vision-Language Navigation is a highly demanding cognitive task that approached from a Machine Learning perspective, involves training an agent to navigate different scenarios following natural-language instructions. This task gets us one step closer to having smooth human-robot interactions. However, there is still a big gap between human performance and current Vision-Language Navigation models. Instructions usually describe paths making reference to places (i.e., turn right at the end of the kitchen), so understanding the semantics of different rooms is necessary to achieve correct navigation. Nevertheless, this understanding is usually not directly supervised and left to be learned implicitly. In this work, we propose an auxiliary task in which agents need to classify the different types of rooms they navigate and show that by adding this task, models learn how to navigate better and more efficiently, resulting in an increase in most Vision-Language Navigation metrics for seen and unseen scenarios during the training phase.

关键词： Computer engineering   Computer science   Electrical engineering   Artificial intelligence  

摘要： Identifying human pose over time provides critical information for understanding human behavior and their physical interaction with the environment surrounding them. In the past few decades, human pose estimation topic has witnessed groundbreaking success in the computer vision field thanks to the powerful deep learning models. These models are trained using several thousands of labeled sample images if not more. Such extensive data requirement poses a fundamental problem for domains (i.e. Small Data domains), in which data collection or labeling is expensive or limited due to privacy or security concerns such as healthcare. In this thesis, we present a data-efficient learning pipeline to address small data problem in a healthcare-related human pose estimation application. In particular, we infer Spatio-temporal human poses to analyze typical vs. atypical behaviors in children with Autism spectrum disorder (ASD). To mitigate data limitation, we propose two thrusts in our learning pipeline. Thrust I is a data-efficient machine learning approach, where a pre-trained (on adult pose images) pose estimation model with deep structure is fine-tuned on a small set of children pose videos, provided to us by our collaborators. We implement a non-linear particle filter interpolation to deal with any missing body keypoints in the estimated poses and employ a novel PoTion (pose motion) based temporal aggregation technique to evaluate poses over time. Thrust II is a synthetic data augmentation technique, where we build a framework to create synthetic 3D humans with articulated bodies in order to render more pose images/videos in our application contexts. We use a novel 3D registration approach based on RANSAC and implement iterative closest point (ICP) to obtain 3D meshes from the scanned point clouds from a kid mannequin and several animal toys, which are then rigged and articulated in the Blender to generate our posable avatars. We then infuse these avatars in various synthetic

关键词： Computer science   Artificial intelligence   Accuracy   Deep learning   Datasets   Forecasting   Optimization   Neural networks   Support vector machines   Probability   Algorithms   Queries   Time series   Performance evaluation   Probability distribution  

摘要： Resource provisioning is essential to optimize cloud operating costs and the performance of cloud applications. Understanding job arrival rates is critical for predicting future workloads to determine the proper amount of resources for provisioning. However, due to the dynamic patterns of cloud workloads, developing a model to accurately forecast job arrival rates is a challenging task. Previously, various prediction models, including Long-Short-Term-Memory (LSTM), have been employed to address the cloud workload prediction problem. Unfortunately, the current state-of-the-art LSTM model leverages recurrences to make a prediction, resulting in increased complexity and degraded computational efficiency as input sequences grow longer. To achieve both higher prediction accuracy and better computational efficiency, this work presents a novel time-series forecasting model for cloud resource provisioning, called WGAN-gp (Wasserstein Generative Adversarial Network with gradient penalty) Transformer. WGAN-gp Transformer is inspired by Transformer network and improved WGAN (Wasserstein Generative Adversarial Networks). Our proposed method adopts a Transformer network as the generator and a multi-layer perceptron network as a critic to improve the overall forecasting performance. WGAN-gp also employs MADGRAD (Momentumized, Adaptive, Dual Averaged Gradient Method for Stochastic Optimization) as the model’s optimizer due its ability to converge faster and generalize better. Extensive experiments on the various real-world cloud workload datasets show improved performance and efficiency of our method. In particular, WGAN-gp Transformer shows 5× faster inference time with up to 5.1% higher prediction accuracy than the state-of-the-art workload prediction technique. Such faster inference time and higher prediction accuracy can be effectively used by cloud resource provisioning and autoscaling mechanisms. We then apply our model to cloud autoscaling and evaluate it on Google Cloud Pl

关键词： Computer science   Artificial intelligence  

摘要： The extensive use of machine learning models for making major decisions such as employment, credit card approval, or juridical decisions has resulted in rising concerns over the widespread existent of bias and discrimination. Several types of bias have been detected in literature so far and they are widely classified into three categories of dataset bias, algorithm bias, and interaction bias. Decisions made by biased machine learning models result in damaging consequences for one orseveral groups of people. In the present research, we focused on using statistical measurement methods to quantify the inherent dataset bias by examining several real-world datasets. The effects of bias on a machine learning model (Random Forest classifier) were studied to understand if in any way bias affects this ML model prediction result. We also investigated the effects of bias on different protected groups with various correlation levels to the class labels. We are curious to know how bias affects different groups of people in a dataset. The answers to these questions will be presented by a series of experiments.

关键词： Computer science   Artificial intelligence  

摘要： Deer Vehicle Collisions (DVCs) are a global problem that is not only resulting in serious injuries to humans but also results in loss of human and deer lives. Deer are more active and less attentive during the mating and hunting seasons. Roadside deer activity such as feeding and strolling along the roadside has a significant correlation with DVCs. To mitigate DVCs, several strategies were used that include vegetation management, fences, underpasses and overpasses, population reduction, warning signs and animal detection systems (ADS). These strategies vary in their efficacy. These strategies may help to reduce DVCs. However, they are not always easily feasible due to false alarms, high cost, unsuitable terrain, land ownership, and other factors. Thus, DVCs are increasing due to the increase in number of vehicles and the absence of intelligent highway safety and alert systems. Detecting deer in real-time on our roads is a challenging problem. Thus, this research work proposed a deer detection and movement DDM technique to automate DVCs mitigation system. The DDM combines computer vision, artificial intelligent methods with deep learning techniques. DDM includes two main deep learning algorithms 1)one stage deep learning algorithm based on Yolov5 that generates a detection model(DM) to detect deer and 2) deep learning algorithm developed by python toolkit DeepLabCut to generate movement model(MM) for detecting the movement of the deer. The proposed method can detect deer with 99.7% precision and succeeds to ascertain if the deer is moving or static with an inference speed of 0.29s. The proposed method can detect deer with 99.7% precision and using DeepLabCut toolkit on the detected deer we can ascertain if the deer is moving or static with an inference speed of 0.29s.