1,721,003 research outputs found
An Unsupervised Approach to Detect Spam Campaigns that Use Botnets on Twitter
No full textIn recent years, Twitter has seen a proliferation of automated accounts or bots that send spam, offer clickbait, compromise security using malware, and attempt to skew public opinion. Previous research estimates that around 9% to 17% of Twitter accounts are bots contributing to between 16% to 56% of tweets on the medium. Our research introduces an unsupervised approach to detect Twitter spam campaigns in real-time. The bot groups we detect tweet duplicate content with shortened embedded URLs over extended periods of time. Our experiments with the detection protocol reveal that bots consistently account for 10% to 50% of tweets generated from 7 popular URL shortening services on Twitter. More importantly, we discover that bots using shortened URLs are connected to large scale spam campaigns that control thousands of domains. We present two use cases of our detection protocol: one as a filtering tool for sentiment analysis during 2014 #UmbrellaRevolution event, the other as a measurement tool to track political bot activities during 2018 #ReleaseTheMemo event. We also document two distinct mechanisms used to control bot groups. Our detection system runs 24/7 and actively collects bots involved in spam campaigns. As of November 2017, we have identified 200,379 unique bot accounts. We make our database of detected bots available for query through a REST API so others can filter out bots to get high quality Twitter datasets for analysis. We report bot accounts and suspicious domains to URL shortening services and Twitter, and our efforts have impacted those companies to suspend abused URLs and update their anti-spam policy
Predicting protein-protein interactions from primary structure
No full textOne of the key challenges in the post-genomic era is to understand protein-protein interactions on a large scale. Given the primary structures of proteins and ligands, along with other information, how well can we computationally predict protein-protein interaction networks? We train Naive Bayes classifiers to classify positive and negative examples of protein-ligand interactions. Such a predictive model can screen large numbers of potential ligands, saving laboratory time and costs.
We demonstrate our approach in predicting interactions between SH3 domains and proline-rich ligands. Using laboratory data, we construct positive and negative examples, learn Naive Bayes models of ligand binding specificity of 8 diverse SH3 domains and visualize our models using an information theory-based measure to reveal potential binding sites. We use our classifiers to screen PxxP ligands from SwissProt for the given SH3 domains and demonstrate improvements over existing predictors. For validating our method, we use our technique to predict a computational interaction network and intersect it with an experimental yeast 2-hybrid network, using the methodology and data from Tong et al. [TDN +02]. Our technique produces comparable results to Tong et al., even without incorporating their consensus sequences
Using reinforcement learning to control advanced life support systems
No full textThis thesis deals with the application of reinforcement learning techniques to the control of a closed life support system simulator, such as could be used on a long duration space mission. We apply reinforcement learning to two different aspects of the simulator, control of recycling subsystems, and control of crop planting schedules. Comparisons are made between distributed and centralized controllers, generalized and non-generalized RL, and between different approaches to the construction of the state table and the design of reward functions. Distributed controllers prove to be superior to centralized controllers both in terms of speed and performance of the controller. Generalization helps to speed convergence, but the performance of the policy derived is dependent on the shape of the reward function
Multi-threat Sustainability Assessment of Bridges and Bridge Networks
No full textSustainable performance of critical infrastructure like bridges under both service loads and extreme events is of growing importance to the society. This performance of highway networks and their constituent bridges can be mapped to sustainability indicators like cost, embodied energy, carbon dioxide (CO2) emissions or resource utilization. With heightened load demands due to overweight trucks and natural hazards on the aging bridge infrastructure, an integrated multi-threat sustainability framework for bridges and bridge networks is essential. The existing sustainability
quanti cation methods and approaches lack a joint sustainability assessment considering bridge vulnerabilities to hazards as well as truck loads, especially using a probabilistic approach. The primary goal of this study is the development of probabilistic frameworks to assist and evaluate sustainability on two levels: bridge and bridge network.
Bridge sustainability is dependent on contributions from different life-cycle phases like construction, operation, maintenance, failure and demolition. Unlike past life-cycle studies, the probabilistic life-cycle sustainability analysis (LCS-A) framework proposed in this study considers life-cycle phase interactions as well as integrates
post-repair performance of bridge components while exploring the distribution of sustainability costs to provide a more holistic view of an bridges sustainability. Interactions
in the life-cycle emerge due to interventions such as maintenance activities which primarily benefit service load performance, but also enhance hazard performance, a previously unexplored secondary interaction effect from life-cycle studies. For bridges subjected to multiple hazard events in their lifetime, the impact of post-repair modification of bridge component behavior in LCS-A is significant and cannot be neglected. This study gathers the insights from bridge level LCS-A and integrates it into developing a probabilistic sustainability evaluation framework for highway networks
as well. The inclusion of bridge failures due to both spatial variation of hazard
as well as truck presence is a key advancement proposed through this research. The traffic simulation of overweight trucks on a bridge in the network is developed using the average daily truck traffic on the bridge and using extreme value theory to predict percentage of overweight trucks. The variation in hazard occurrence and intensity is captured by using a probabilistic suite of scenarios for the network. Advances are also made in the methodology to evaluate traffic emissions by incorporation of traffic flow modeling and fuel congestion into the network assessment.
The LCS-A and network frameworks developed in this study are capable of handling various sources of uncertainties, with propagation of uncertainties facilitated by use of surrogate models when predicting bridge failures. In addition to developing probabilistic distributions of sustainability metrics, this study also recommends using probabilistic sensitivity analysis to understand how the uncertainty in the sustainability
indicator is influenced by uncertainties in the input parameters. Such a
detailed sensitivity analysis highlights opportunities for reducing uncertainties in sustainability outcomes by focusing on reducing the uncertainties in the most important input parameters. The sustainability framework developed as part of this thesis also can be used to de-aggregate bridge or network sustainability into contributions from
their constituent components. A new sustainability informed component importance measure (SCIM) is proposed in this study that leverages the probabilistic nature of the sustainability contributions from individual components and maps it to system level sustainability consequences. The SCIMs proposed in this study is developed by
adopting a flexible system failure based on a user-defined threshold on the system level sustainability indicator. The temporal evolution of component importance due to aging related deterioration or potential change in user-de fined thresholds are also incorporated into the SCIMs. The probabilistic frameworks developed in this thesis can support owners in their efforts to improve bridge or network sustainability,such as evaluating the impact of interventions or repairs on performance of bridge for future hazards or service loads and making upgrade decisions to minimize impacts
of bridge failures to the surrounding natural and built environment. Moreover, stakeholders may pose sustainability objectives or intervention schedules for preferred risk threshold given new insight on the full probability distribution of sustainability outcomes
Analyzing robustness of models of chaotic dynamical systems learned from data with Echo state networks
No full textLarge scale engineering as well as natural systems, such as weather, often have high-dimensional state spaces and exhibit chaotic dynamics. To model the behavior of such systems, sets of coupled Partial Differential Equations (PDEs) are formulated and solved using high-performance computing systems. More recently, significant attention has grown toward the use of Artificial Intelligence (AI) and Machine Learning (ML) techniques, in particular, using data-driven modeling to learn fast and accurate surrogate process models trained on high-resolution data obtained from simulations, or observations of chaotic systems. Echo state networks (ESN), a family of recurrent neural network algorithms, have emerged as one of the most promising techniques to learn predictive models of chaotic dynamical systems directly from data.
In spite of their success in learning chaotic dynamical systems from data, there are many open questions. Some of them are practical engineering concerns such as: how to choose training parameters (reservoir size, spectral radius, length of training sequence) for specific problems, how robust the learned models are to variations in data, and in training parameters (initialization of random weights, reservoir size, spectral radius). Others are open theoretical questions such as: why do ESNs work at all, in particular, which aspects of the underlying dynamical systems are captured by the learned reservoirs, and which factors determine the prediction horizon of the learned models. In this thesis, we study these practical and theoretical questions in the context of two models of chaotic dynamical systems, Lorenz63 and Lorenz96, which are prototypes of more complex weather models.
We show that the predictive performance of the learned models is highly sensitive to initial conditions — i.e., for different training sequences all of the same lengths but with different initial states, there is considerable variation in prediction horizon from
0.1 MTU to 3.8 MTU in Lorenz63 and from 0.4 MTU to 2.8 MTU in Lorenz96. We also show that variations in the initialization of (random) input weights and (random) reservoir weights at the start of the training phase yields models with varying prediction horizon for the very same training sequence. We discuss the implications of these findings in the construction of robust ESN models for Lorenz systems. To help explain the observed variations in predictive performance with initial conditions, and to understand when and why ESNs work, we use dimensionality reduction and clustering algorithms to visualize the evolution of high-dimensional reservoir states during training and prediction. Our main finding is that, in a well-trained model, reservoir states mirror the dynamics of the chaotic system from which the data is derived. In particular, we can infer the number of dynamical components from the
non-linear clustering of the reservoir states. In the context of Lorenz63, we show that the sensitivity to initial conditions stems from the locations of the initial condition relative to the two components of the underlying system
On accelerating the searches for compilation sequences in an adaptive compiler
No full textRecent research show that adaptive compiler can produce consistent improvement over a traditional fixed-sequence compiler by conducting feedback-directed searches for good compilation sequences for specific programs, machines and performance objectives. However, such improvement is usually achieved at very high search cost. This thesis proposes two approaches to accelerate the searches for a good compilation sequence in an adaptive compiler. First, a local search algorithm, Greedy Neighbor Exploration algorithm (GNE), is proposed. It uses optimistic greedy construction and cleanup procedures to generate a richer set of meaningful variations by randomized insertion and removal of transformations. Experimental results on a range of standard benchmark suites show that GNE finds better compilation sequences in less than a quarter of the evaluations required by current search algorithms, such as genetic and hill climbing algorithms. Second, code normalization techniques are developed to hash programs and detect equivalent code. This can avoid unnecessary runs of programs
Comparing vector-based and ACT-R memory models using large-scale datasets: User-customized hashtag and tag prediction on Twitter and StackOverflow
No full textThe growth of social media and user-created content on online sites provides unique opportunities to study models of declarative memory. The tasks of choosing a hashtag for a tweet and tagging a post on StackOverflow were framed as declarative memory retrieval problems. Two state-of-the-art cognitively-plausible declarative memory models were evaluated on how accurately they predict a user’s chosen tags: an ACT-R based Bayesian model and a random permutation vector-based model. Millions of posts and tweets were collected, and both declarative memory models were used to predict Twitter hashtags and StackOverflow tags. The results show that past user behavior of tag use is a strong predictor of future behavior. Furthermore, past behavior was successfully incorporated into the random permutation model that previously used only context. Also, ACT-R’s attentional weight term was linked to a common entropy-weighting natural language processing method used to attenuate low-predictor words. Word order was not found to be strong predictor of tag use, and the random permutation model performed comparably to the Bayesian model without including word order. This shows that the strength of the random permutation model is not in the ability to represent word order, but rather in the way in which context information is successfully compressed. Finally, model accuracy was moderate to high for the tasks, which supports the theory that choosing tags on StackOverflow and Twitter is primarily a declarative memory retrieval process. The results of the large-scale exploration show how the architecture of the two memory models can be modified to significantly improve accuracy, and may suggest task-independent general modifications that can help improve model fit to human data in a much wider range of domains
Going Beyond Counting First Authors in Author Co-citation Analysis
Get PDFThe present study examines one of the fundamental aspects of author co-citation analysis (ACA) - the way co-citation
counts are defined. Co-citation counting provides the data on which all subsequent statistical analyses and mappings
are based, and we compare ACA results based on two different types of co-citation counting - the traditional type that
only counts the first one among a cited work's authors on the one hand and a non-traditional type that takes into
account the first 5 authors of a cited work on the other hand. Results indicate that the picture produced through this non-traditional author co-citation counting contains more coherent author groups and is therefore considerably clearer. However, this picture represents fewer specialties in the research field being studied than that produced through the traditional first-author co-citation counting when the same number of top-ranked authors is selected and analyzed. Reasons for these effects are discussed
Predicting wind induced damage to residential structures: a machine learning approach
No full textHurricane winds can cause significant physical damage to residential properties. Pre-storm prediction of wind damage risk allows residents and city emergency officials to plan actions to reduce loss of life and property. In this thesis, I have developed a data-driven machine learning framework to estimate the probability of structural damage risk to a home subject to hurricane force winds. The modeling framework maps a set of predictor variables with the potential to explain structural damage to actual observations of homes damaged by hurricane winds. Widely used wind damage prediction models are parametric and are based on the physics of a struc- ture responding to a wind load. Using a wind damage dataset gathered from about 700,000 residential buildings after Hurricane Ike in 2008 over Harris County, I have built a hybrid machine learning model that combines classification trees and logistic regression. My model is 23.7% more accurate than the physics-based approach at pre- dicting expected damage at the one-kilometer square block level. I demonstrate the robustness of model by using it to predict wind damage to homes in Harris County for simulated hurricanes of category 1 through 5 on the Saffir-Simpson scale. My model produces more accurate pre-storm predictions of wind damage risk which will enable communities to respond to hurricane threats more effectively
- …
