Experimental research publications provide figure form resources including graphs, charts, and any type of images to effectively support and convey methods and results. To describe figures, authors add captions, which are often incomplete, and more descriptions reside in body text. This work presents a method to extract figure descriptive text from the body of scientific articles. We adopted ontological semantics to aid concept recognition of figure-related information, which generates human and machine-readable knowledge representations from sentences. Our results show that conceptual models bring an improvement in figure descriptive sentence classification over word-based approaches.

Energy is limited in mobile systems, however, so for this possibility to become a viable opportunity, energy usage must be conservative. The Low-Power Image Recognition Challenge (LPIRC) is the only competition integrating image recognition with low power. LPIRC has been held annually since 2015 as an on-site competition. To encourage innovation, LPIRC has no restriction on hardware or software platforms: the only requirement is that a solution be able to use HTTP to communicate with the referee system to retrieve images and report answers. Each team has 10 minutes to recognize the objects in 5,000 (year 2015) or 20,000 (years 2016 and 2017) images.

Although previous work has shown that member and structural features are important to the future popularity of groups in EBSN, it is not yet clear how different member roles and the interplay between them contribute to group popularity. In this paper, we study a real-world dataset from Meetup --- a popular EBSN platform --- and propose a deep neural network based method to predict the popularity of new Meetup groups. Our method uses group-level features specific to event-based social networks, such as time and location of events in a group, as well as the structural features internal to a group, such as the inferred member roles in a group and social substructures among members. Empirically, our approach reduces the RMSE of the popularity prediction (measured in RSVPs) of a group's future events by up to 12%, against the state-of-the-art baselines.

Motivated by what is required for real-time path planning, the paper starts out by presenting RMPD, a new recursive ''local'' planner founded on the key notion that, unless made necessary by an obstacle, there must be no deviation from the shortest path between any two points, which would normally be a straight line path in the configuration space. Subsequently, we increase the power of RMPD by introducing the notion of cost-awareness into the algorithm to improve the path quality -- this is done by associating obstacle and smoothness costs with the currently selected path points and factoring those costs in choosing the best points for the next iteration. In this manner, the overall strategy in the cost-aware form of RMPD, cRMPD, combines the computational efficiency made possible by the recursive RMPD planner with the cost efficacy of a stochastic trajectory optimizer to rapidly produce high-quality local collision-free paths. Based on the test cases we have run, our experiments show that cRMPD can reduce planning time by up to two orders of magnitude as compared to RRT-Connect, while still maintaining a path length optimality equivalent to that of RRT*.

Given a group of publications, we investigate the prob- lem of identifying the papers with the most impact on others. We refer to these papers as influential in the sense that they introduce new concepts and language that will affect how future articles are written. In this pa- per we propose weighted PageRank algorithm that uses textual information from articles and information from citation graph to rank the impact of publications, then we automatically summarize these publications and ex- tract important keywords. We show that using our algo- rithm outperforms default citation-based techniques in ranking influential papers (those which won best paper award) with no less than 2% in F1-score and NDCG. We also show that our algorithm outperforms previous graph-based keyword extraction techniques with no less than 1.5% in F1-score.

Selection and confounding biases are the two most common impediments to the applicability of causal inference methods in large-scale settings. We generalize the notion of backdoor adjustment to account for both biases and leverage external data that may be available without selection bias (e.g., data from census). We introduce the notion of adjustment pair and present complete graphical conditions for identifying causal effects by adjustment. We further design an algorithm for listing all admissible adjustment pairs in polynomial delay, which is useful for researchers interested in evaluating certain properties of some admissible pairs but not all (common properties include cost, variance, and feasibility to measure). Finally, we describe a statistical estimation procedure that can be performed once a set is known to be admissible, which entails different challenges in terms of finite samples.

In this work we generalize traditional node/link prediction tasks in dynamic heterogeneous networks, to consider joint prediction over larger k-node induced subgraphs. Our key insight is to incorporate the unavoidable dependencies in the training observations of induced subgraphs into both the input features and the model architecture itself via high-order dependencies. The strength of the representation is its invariance to isomorphisms and varying local neighborhood sizes, while still being able to take node/edge labels into account, and facilitating inductive reasoning (i.e., generalization to unseen portions of the network). Empirical results show that our proposed method significantly outperforms other state-of-the-art methods designed for static and/or single node/link prediction tasks. In addition, we show that our method is scalable and learns interpretable parameters.

AI plays an increasingly prominent role in society since decisions that were once made by humans are now delegated to automated systems. These systems are currently in charge of deciding bank loans, criminals' incarceration, and the hiring of new employees, and it's not difficult to envision that they will in the future underpin most of the decisions in society. Despite the high complexity entailed by this task, there is still not much understanding of basic properties of such systems. For instance, we currently cannot detect (neither explain nor correct) whether an AI system can be deemed fair (i.e., is abiding by the decision-constraints agreed by society) or it is reinforcing biases and perpetuating a preceding prejudicial practice. Issues of discrimination have been discussed extensively in political and legal circles, but there exists still not much understanding of the formal conditions that a system must meet to be deemed fair. In this paper, we use the language of structural causality (Pearl, 2000) to fill in this gap. We start by introducing three new fine-grained measures of transmission of change from stimulus to effect, which we called counterfactual direct (Ctf-DE), indirect (Ctf-IE), and spurious (Ctf-SE) effects. We then derive what we call the causal explanation formula, which allows the AI designer to quantitatively evaluate fairness and explain the total observed disparity of decisions through different discriminatory mechanisms. We apply these measures to various discrimination analysis tasks and run extensive simulations, including detection, evaluation, and optimization of decision-making under fairness constraints. We conclude studying the trade-off between different types of fairness criteria (outcome and procedural), and provide a quantitative approach to policy implementation and the design of fair AI systems.

Statistical script learning is an effective way to acquire world knowledge which can be used for commonsense reasoning. Statistical script learning induces this knowledge by observing event sequences generated from texts. The learned model thus can predict subsequent events, given earlier events. Recent approaches rely on learning event embeddings which capture script knowledge. In this work, we suggest a general learning model–Featured Event Embedding Learning (FEEL)–for injecting event embeddings with fine grained information. In addition to capturing the dependencies between subsequent events, our model can take into account higher level abstractions of the input event which help the model generalize better and account for the global context in which the event appears. We evaluated our model over three narrative cloze tasks, and showed that our model is competitive with the most recent state-of-the-art. We also show that our resulting embedding can be used as a strong representation for advanced semantic tasks such as discourse parsing and sentence semantic relatedness.

Task instruction quality is widely presumed to affect outcomes, such as accuracy, throughput, trust, and worker satisfaction. Best practices guides written by experienced requesters share their advice about how to craft task interfaces. However, there is little evidence of how specific task design attributes affect actual outcomes. This paper presents a set of studies that expose the relationship between three sets of measures: (a) workers’ perceptions of task quality, (b) adherence to popular best practices, and (c) actual outcomes when tasks are posted (including accuracy, throughput, trust, and worker satisfaction). These were investigated using collected task interfaces, along with a model task that we systematically mutated to test the effects of specific task design guidelines.