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...

In this paper, we provide new complexity results for algorithms that learn discrete-variable Bayesian networks from data. Our apply whenever the learning algorithm uses a scoring criterion favors simplest structure which model is able to represent generative distribution exactly. therefore hold consistent and applied sufficiently large dataset. We show identifying high-scoring structures NP-hard, even when any combination of one or more following hold: perfect with respect some DAG...

We develop a semantic parsing framework based on similarity for open domain question answering (QA).We focus single-relation questions and decompose each into an entity mention relation pattern.Using convolutional neural network models, we measure the of mentions with entities in knowledge base (KB) patterns relations KB.We score relational triples KB using these measures select top scoring triple to answer question.When evaluated open-domain QA task, our method achieves higher precision...

While relation extraction has traditionally been viewed as a task relying solely on textual data, recent work shown that by taking input existing facts in the form of entity-relation triples from both knowledge bases and performance can be improved significantly. Following this new paradigm, we propose tensor decomposition approach for base embedding is highly scalable, especially suitable extraction. By leveraging relational domain about entity type information, our learning algorithm...

We present a new methodology for visualizing navigation patterns on Web site. In our approach, we rst partition site users into clusters such that only with similar paths through the are placed same cluster. Then, each cluster, display these within The clustering approach employ is model based (as opposed to distance based) and partitions according order in which they request pages. particular, cluster by learning mixture of rst-order Markov models using ExpectationMaximization algorithm....

We present several methods for mining knowledge from the query logs of MSN search engine. Using logs, we build a time series each word or phrase (e.g., 'Thanksgiving' 'Christmas gifts') where elements are number times that is issued on day. All describe use sequences this form and can be applied to data generally. Our primary goal discovery semantically similar queries do so by identifying with demand patterns. Utilizing best Fourier coefficients energy omitted components, improve upon...

The statistical community has brought logical rigor and mathematical precision to the problem of using data make inferences about a model's parameter values. TETRAD project, related work in computer science statistics, aims apply those standards background knowledge specification. We begin by drawing analogy between estimation model specification search. then describe how structural equation entails familiar constraints on covariance matrix for all admissible values its parameters; we survey...

We formulate necessary and sufficient conditions for an arbitrary discrete probability distribution to factor according undirected graphical model, or a log-linear other more general exponential models. For decomposable models these are equivalent set of conditional independence statements similar the Hammersley–Clifford theorem; however, we show that nondecomposable they not. also can have nonrational maximum likelihood estimates. These results used give several novel characterizations

Content-based recommendation systems can provide recommendations for "cold-start" items which little or no training data is available, but typically have lower accuracy than collaborative filtering systems. Conversely, techniques often accurate recommendations, fail on cold start items. Hybrid schemes attempt to combine these different kinds of information yield better across the board.

This is the Proceedings of Nineteenth Conference on Uncertainty in Artificial Intelligence, which was held Acapulco, Mexico, August 7-10 2003

Learning to capture text-table alignment is essential for tasks like text-to-SQL. A model needs correctly recognize natural language references columns and values ground them in the given database schema. In this paper, we present a novel weakly supervised Structure-Grounded pretraining framework (StruG) text-to-SQL that can effectively learn based on parallel corpus. We identify set of prediction tasks: column grounding, value grounding column-value mapping, leverage pretrain encoder....

Modern systems can augment people's capabilities by using machine-learned models to surface intelligent behaviors. Unfortunately, building these remains challenging and beyond the reach of non-machine learning experts. We describe interactive machine teaching (IMT) its potential simplify creation models. One key characteristics IMT is iterative process in which human-in-the-loop takes role a teacher how perform task. explore alternative theories as theoretical foundations for IMT, intrinsic...

We consider the dispute between causal decision theorists and evidential over Newcomb-like problems. introduce a framework relating causation directed graphs developed by Spirtes et al. (1993) evaluate several arguments in this context. argue that much of debate two camps is misplaced; disputes turn on distinction conditioning an event E as against I which action to bring about E. give essential machinery for calculating effect intervention recent work extends basic account given here case...

Soft keyboards offer touch-capable mobile and tabletop devices many advantages such as multiple language support room for larger displays. On the other hand, because soft lack haptic feedback, users often produce more typing errors. In order to make robust noisy input, researchers have developed key-target resizing algorithms, where underlying target areas keys are dynamically resized based on their probabilities. this paper, we describe how overly aggressive can sometimes prevent from...

The current processes for building machine learning systems require practitioners with deep knowledge of learning. This significantly limits the number that can be created and has led to a mismatch between demand ability organizations build them. We believe in order meet this growing we must increase individuals teach machines. postulate achieve goal by making process teaching machines easy, fast above all, universally accessible. While focuses on creating new algorithms improving accuracy...