The current neural network models for event detection have only considered the sequential representation of sentences. Syntactic representations have not been explored in this area although they provide an effective mechanism to directly link words to their informative context for event detection in the sentences. In this work, we investigate a convolutional neural network based on dependency trees to perform event detection. We propose a novel pooling method that relies on entity mentions to aggregate the convolution vectors. The extensive experiments demonstrate the benefits of the dependency-based convolutional neural networks and the entity mention-based pooling method for event detection. We achieve the state-of-the-art performance on widely used datasets with both perfect and predicted entity mentions.

Unsolicited or unwanted messages is a byproduct of virtually every popular social media website. Spammers have become increasingly proficient at bypassing conventional spam filters, prompting a stronger effort to develop new methods that accurately detect spam while simultaneously acting as a more robust classifier against users that modify their behavior in order to avoid detection. This paper shows the usefulness of a relational model that works in conjunction with an independent model. First, an independent model is built using features that characterize individual comments and users, capturing the cases where spam is obvious. Second, a relational model is built, taking advantage of the interconnected nature of users and their comments. By feeding our initial predictions from the independent model into the relational model, we can start to propagate information about spammers and spam comments to jointly infer the labels of all spam comments at the same time. This allows us to capture the obfuscated spam comments missed by the independent model that are only found by looking at the relational structure of the social network. The results from our experiments demonstrates the viability of our method, and shows that models utilizing the underlying structure of the social network are more effective at detecting spam than ones that do not.

In recent years, deep learning has spread beyond both academia and industry with many exciting real-world applications. The development of deep learning has presented obvious privacy issues. However, there has been lack of scientific study about privacy preservation in deep learning. In this paper, we concentrate on the auto-encoder, a fundamental component in deep learning, and propose the deep private auto-encoder (dPA). Our main idea is to enforce ε-differential privacy by perturbing the objective functions of the traditional deep auto-encoder, rather than its results. We apply the dPA to human behavior prediction in a health social network. Theoretical analysis and thorough experimental evaluations show that the dPA is highly effective and efficient, and it significantly outperforms existing solutions.

In this paper, we focus on a novel knowledge reuse scenario where the knowledge in the source schema needs to be translated to a semantically heterogeneous target schema. We refer to this task as “knowledge translation” (KT). Unlike data translation and transfer learning, KT does not require any data from the source or target schema. We adopt a probabilistic approach to KT by representing the knowledge in the source schema, the mapping between the source and target schemas, and the resulting knowledge in the target schema all as probability distributions, specially using Markov random fields and Markov logic networks. Given the source knowledge and mappings, we use standard learning and inference algorithms for probabilistic graphical models to find an explicit probability distribution in the target schema that minimizes the Kullback-Leibler divergence from the implicit distribution. This gives us a compact probabilistic model that represents knowledge from the source schema as well as possible, respecting the uncertainty in both the source knowledge and the mapping. In experiments on both propositional and relational domains, we find that the knowledge obtained by KT is comparable to other approaches that require data, demonstrating that knowledge can be reused without data.

The biggest limitation of probabilistic graphical models is the complexity of inference, which is often intractable. An appealing alternative is to use tractable probabilistic models, such as arithmetic circuits (ACs) and sum-product networks (SPNs), in which marginal and conditional queries can be answered efficiently. In this paper, we present the first discriminative structure learning algorithm for ACs, DACLearn (Discriminative AC Learner), which optimizes conditional log-likelihood. Based on our experiments, DACLearn learns models that are more accurate and compact than other tractable generative and discriminative baselines.

This paper outlines three projects that lay the foundation for a trans-disciplinary approach to the creation of interactive, multi-sensory devices combining biofeedback, virtual reality, and physical/virtual human-machine interactions. We explore new possibilities for interoperability and enhancing interoception and mindfulness with potential research contributions for novel personal, professional and medical applications.

Statistical relational artificial intelligence combines first-order logic and probability in order to handle the complexity and uncertainty present in many real-world domains. However, many real-world domains also include multiple agents that cooperate or compete according to their diverse goals. In order to handle such domains, an autonomous agent must also consider the actions of other agents. In this paper, we show that existing statistical relational modeling and inference techniques can be readily adapted to certain adversarial or non-cooperative scenarios. We also discuss how learning methods can be adapted to be robust to the behavior of adversaries. Extending and applying these methods to real-world problems will extend the scope and impact of statistical relational artificial intelligence.

This paper describes the Dynamic Execution Order Scheduling (DEOS) system that has been developed to handle highly dynamic and interactive scheduling domains. Unlike typical scheduling problems which have a static task list, DEOS is able to handle dynamic task lists in which tasks are added, deleted and modified “on the fly" DEOS is also able to handle tasks with uncertain and/or probabilistic outcomes. DEOS extends the current scheduling paradigm to allow tasking in dynamic and uncertain environments by viewing the planning and scheduling tasks as being integrated and evolving entities. DEOS has been successfully applied to the domains of Air Campaign Planning (ACP) and Intelligence, Surveillance and Reconnaissance (ISR) management. The paper provides an overview of the dynamic task model and the “penalty box" scheduling algorithm which was developed to provide robust solutions to over constrained scheduling problems. The basic algorithm is described together with extensions to handle flexible time constraints.

In recent years, there has been a growing interest in learning tractable graphical models in which exact inference is efficient. Two main approaches are to restrict the inference complexity directly, as done by low-treewidth graphical models and arithmetic circuits (ACs), or introduce latent variables, as done by mixtures of trees, latent tree models, and sum-product networks (SPNs). In this paper, we combine these approaches to learn a mixtures of ACs (MAC). A mixture can represent many distributions exponentially more compactly than a single AC. By using ACs as mixture components, MAC can represent complex distributions using many fewer components than required by other mixture models. MAC generalizes ACs, mixtures of trees, latent class models, and thin junction trees, and can be seen as a special case of an SPN. Compared to state-of-the-art algorithms for learning SPNs and other tractable models, MAC is consistently more accurate while maintaining tractable inference.

Dependency networks are a compelling alternative to Bayesian networks for learning joint probability distributions from data and using them to compute probabilities. A dependency network consists of a set of conditional probability distributions, each representing the probability of a single variable given its Markov blanket. Running Gibbs sampling with these conditional distributions produces a joint distribution that can be used to answer queries, but suffers from the traditional slowness of sampling-based inference. In this paper, we observe that the mean field update equation can be applied to dependency networks, even though the conditional probability distributions may be inconsistent with each other. In experiments with learning and inference on 12 datasets, we demonstrate that mean field inference in dependency networks offers similar accuracy to Gibbs sampling but with orders of magnitude improvements in speed. Compared to Bayesian networks learned on the same data, dependency networks offer higher accuracy at greater amounts of evidence. Furthermore, mean field inference is consistently more accurate in dependency networks than in Bayesian networks learned on the same data.