A5053373401- Bayesian Modeling and Causal Inference
- Machine Learning and Algorithms
- Adversarial Robustness in Machine Learning
- Topic Modeling
- Machine Learning and Data Classification
- Semantic Web and Ontologies
- Network Security and Intrusion Detection
- Spam and Phishing Detection
- Advanced Malware Detection Techniques
- Data Quality and Management
- Anomaly Detection Techniques and Applications
- Explainable Artificial Intelligence (XAI)
- Natural Language Processing Techniques
- Data Mining Algorithms and Applications
- Biomedical Text Mining and Ontologies
- Neural Networks and Applications
- Hate Speech and Cyberbullying Detection
- AI-based Problem Solving and Planning
- Data Management and Algorithms
- Domain Adaptation and Few-Shot Learning
- Advanced Database Systems and Queries
- Misinformation and Its Impacts
- Data Stream Mining Techniques
- Service-Oriented Architecture and Web Services
- Complex Network Analysis Techniques
University of Oregon
2015-2024
University of Washington
2005-2012
Seattle University
2005-2009
We propose an efficient method to generate white-box adversarial examples trick a character-level neural classifier. find that only few manipulations are needed greatly decrease the accuracy. Our relies on atomic flip operation, which swaps one token for another, based gradients of one-hot input vectors. Due efficiency our method, we can perform training makes model more robust attacks at test time. With use semantics-preserving constraints, demonstrate HotFlip be adapted attack word-level...
Many classification tasks, such as spam filtering, intrusion detection, and terrorism are complicated by an adversary who wishes to avoid detection. Previous work on adversarial has made the unrealistic assumption that attacker perfect knowledge of classifier [2]. In this paper, we introduce reverse engineering (ACRE) learning problem, task sufficient information about a construct attacks. We present efficient algorithms for linear classifiers with either continuous or Boolean features...
Most subfields of computer science have an interface layer via which applications communicate with the infrastructure, and this is key to their success (e.g., Internet in networking, relational model databases, etc.). So far has been missing AI. First-order logic probabilistic graphical models each some necessary features, but a viable requires combining both. Markov powerful new language that accomplishes by attaching weights first-order formulas treating them as templates for features...
The study and understanding of human behaviour is relevant to computer science, artificial intelligence, neural computation, cognitive philosophy, psychology, several other areas. Presupposing cognition as basis behaviour, among the most prominent tools in modelling are computational-logic systems, connectionist models cognition, uncertainty. Recent studies psychology have produced a number reasoning, learning, language that underpinned by computation. In addition, efforts science research...
Naive Bayes models have been widely used for clustering and classification. However, they are seldom general probabilistic learning inference (i.e., estimating computing arbitrary joint, conditional marginal distributions). In this paper we show that, a wide range of benchmark datasets, naive learned using EM accuracy time comparable to Bayesian networks with context-specific independence. Most significantly, is orders magnitude faster than network Gibbs sampling belief propagation. This...
Evaluating on adversarial examples has become a standard procedure to measure robustness of deep learning models. Due the difficulty creating white-box for discrete text input, most analyses NLP models have been done through black-box examples. We investigate character-level neural machine translation (NMT), and contrast adversaries with novel adversary, which employs differentiable string-edit operations rank changes. propose two types attacks aim remove or change word in translation,...
We propose an efficient method to generate white-box adversarial examples trick a character-level neural classifier. find that only few manipulations are needed greatly decrease the accuracy. Our relies on atomic flip operation, which swaps one token for another, based gradients of one-hot input vectors. Due efficiency our method, we can perform training makes model more robust attacks at test time. With use semantics-preserving constraints, demonstrate HotFlip be adapted attack word-level...
Traditional Markov network structure learning algorithms perform a search for globally useful features. However, these are often slow and prone to finding local optima due the large space of possible structures. Ravikumar et al. recently proposed alternative idea applying L1 logistic regression learn set pair wise features each variable, which then combined into global model. This paper presents DTSL algorithm, uses probabilistic decision trees as Our approach has two significant advantages:...
Supervised stance classification, in such domains as Congressional debates and online forums, has been a topic of interest the past decade.Approaches have evolved from text classification to structured output prediction, including collective sequence labeling.In this work, we investigate stances on Twitter, using hinge-loss Markov random fields (HL-MRFs).Given graph all posts, users, their relationships, constrain predicted post labels latent user correspond with network structure.We focus...
A number of text mining and information extraction projects such as Text Runner NELL seek to automatically build knowledge bases from the rapidly growing amount on web. In order scale size web, these often employ ad hoc heuristics reason about uncertain contradictory rather than reasoning jointly all candidate facts. this paper, we present a Markov logic-based system for cleaning an extracted base. This allows scalable take advantage joint probabilistic inference, or, conversely, logic be...
Markov logic can be used as a general framework for joining logical and statistical AI.
Determining a computer’s identity is challenge of critical importance to users wishing ensure that they are interacting with the correct system; it also extremely valuable forensics investigators. However, even hosts contain trusted computing hardware establish can be defeated by relay and impersonation attacks. In this paper, we consider how leverage virtually ubiquitous USB interface uniquely identify computers based on characteristics their hardware, firmware, software stacks. We collect...
An increasing number of machine learning applications involve detecting the malicious behavior an attacker who wishes to avoid detection. In such domains, attackers modify their evade classifier while accomplishing goals as efficiently possible. The typically do not know exact parameters, but they may be able it by observing classifier's on test instances that construct. For example, spammers learn most effective ways spams sending emails accounts control. This problem setting has been...
Rumor stance classification is the task of determining towards a rumor in text. This first step effective tracking on social media which an increasingly important task. In this work, we analyze Twitter users' toward rumorous tweet, users could support, deny, query, or comment upon rumor. We propose deep attentional CNN-LSTM approach, takes sequence tweets thread conversation as input. use neighboring timeline context vectors to capture temporal dynamism evolution. addition, extra features...