An Attention Factor Graph Model for Tweet Entity Linking

The ability to reason over partially observable networks of interacting states is a fundamental competency in probabilistic robotics. While the well-known factor graph and Gaussian process models provide flexible and computationally efficient solutions for this inference problem in the special cases in which all of the hidden states are either finite-dimensional parameters or real-valued functions, respectively, in many cases we are interested in reasoning about heterogeneous networks whose hidden states are comprised of both finite-dimensional parameters and functions. To that end, in this paper we propose a novel probabilistic generative model that incorporates both factor graphs and Gaussian processes to model these heterogeneous systems. Our model improves upon prior approaches to inference within these networks by removing the assumption of any specific set of conditional independences amongst the modeled states, thereby significantly expanding the class of systems that can be represented. Furthermore, we show that inference within this model can always be performed by means of a two-stage procedure involving inference within a factor graph followed by inference over a Gaussian process; by exploiting fast inference methods for the individual factor graph and Gaussian process models to solve each of these subproblems in succession, we thus obtain a general framework for computationally efficient inference over heterogeneous finite-/infinite-dimensional systems.

Realizing accurate detection of an unknown radio transmitter (URT) has become a challenging problem due to its unknown parameter information. A method based on received signal strength difference (RSSD) fingerprint positioning technique and using factor graph (FG) for 2-D scenario has been developed. However, the URT positioning under 3-D scenario is more difficult with the large number of unknown parameters and has greater practical significance. In addition, the previous RSSD-based FG model is not accurate enough to express the relationship between the RSSD and corresponding location coordinates since the RSSD variances of reference points are different in practice. This paper proposes a more accurate 3-D FG model to reduce the influence of difference of RSSD measurement variances on positioning accuracy effectively by utilizing weighted least square (WLS). With the proposed RSSD-based 3-D WLSFG model and sum-product rule, positioning process of the proposed 3-D RSSD-WLSFG algorithm is derived. To verify the feasibility of the proposed method, we also explores the effects of different signal receiver numbers and grid distances on positioning accuracy. The simulation experiment results show that the proposed algorithm can obtain the best positioning performance compared with the conventional K nearest neighbor (KNN) algorithm and RSSD-FG algorithm under different grid distances and signal receiver numbers.

A large number of stochastic networks including loss networks and certain queueing networks have product-form steady-state probabilities. However, for most practical networks, evaluating the system performance is a difficult task due to the presence of a normalization constant. We propose a new framework based on probabilistic graphical models to tackle this task. Specifically, we use factor graphs to model the stationary distribution of a network. For networks with arbitrary topology, we can apply efficient message-passing algorithms like the sum-product algorithm to compute the exact or approximate marginal distributions of all state variables and related performance measures such as blocking probabilities. Through extensive numerical experiments, we show that the sum-product algorithm returns very accurate blocking probabilities and greatly outperforms the reduced load approximation for loss networks with a variety of topologies. The factor graph model also provides a promising approach for analyzing product-form queueing networks.

Globally referenced and accurate positioning is of great significance for the application of intelligent vehicular systems. The visual and inertial measurement unit (IMU) integrated navigation system (VINS) is widely used in many intelligent positioning-based services. However, VINS are prone to drift over time and their performance is significantly compromised in urban canyons since the numerous outlier visual features caused by moving objects and unstable illuminations. Similarly, the global navigation satellite system (GNSS) is reliable in open areas but struggles with signal obstructions and reflections in urban canyons. To exploit the complementariness of the VINS and GNSS, this paper proposed a sliding window factor graph optimization (FGO)-based GNSS/Visual/IMU/Map (GVIM) integration, which exploits the full suit raw measurements from GNSS, IMU, and visual, together with the prior lightweight map. Unlike the existing tightly-coupled GNSS/Visual/IMU integration schemes, the potential of the carrier-phase measurement is fully explored using the recently developed window carrier phase (WCP) model. To further mitigate the impacts of the GNSS outliers, a novel sliding window (SW)-based map matching model is proposed to correct the states using the lightweight OpenStreetMap (OSM). Different from conventional filtering-based map matching, the states within the sliding window are associated with the lane information from the OSM which effectively exploits the measurement redundancy provided by the factor graph model. The experiments are conducted in the urban canyons of Hong Kong, utilizing a challenging dataset characterized by poor satellite visibility and numerous visual feature outliers, to verify the performance of the proposed method.

In the research, we present a novel nonlinear state estimation method to solve the problems of multi-sensor fusion application in navigation systems. In a full Bayesian framework, the muti-sensor fusion is performed by estimating the maximum a posterior over the joint probability distribution function (PDF) of all state variables. In order to exploit the full sparsity of the system, the joint PDF is represented by a factor graph model. Since the outliers which represent corrupted observations could affect the accuracy of state estimation, a variational approximation scheme is applied for robust multi-sensor fusion. The proposed method is experimentally verified using the multi-sensor data that recoded by an integrated navigation system. The simulation results demonstrate that the proposed method provides a comparable performance to the traditional muti-sensor fusion method.

<p>Location awareness has been widely used for cooperative localization of target in wireless sensor networks (WSN). But as the number of agent node increases, cooperative localization based on nonparametric belief propagation (BP) causes high communication and computation burden. In addition, high localization accuracy is also the goal of this paper. To this end, this paper based on parameterized BP strategy proposed a distributed cubature Kalman filter (DCKF) algorithm named BP-DCKF. Firstly, basing on joint posterior probability density function of all nodes, this paper constructs a factor graph (FG) model, then the edge posterior distribution of the nodes is obtained by BP strategy. Secondly, considering Gaussian parameterized BP which reconstruct the parameterized message of agents transmitting, this paper proposed the improved DCKF method, moreover, obtaining the locating model of agent node related to posterior distribution of each node in the FG. Finally, based on the localization model, the location estimation of mobile agent node can be obtained utilizing DCKF method to iteratively solve the edge posterior distribution of agent node. Simulation results show that comparing with traditional algorithm, the proposed algorithm is higher on localization accuracy, lower on communication burden.</p> <p> </p>

An improved integrated navigation method with enhanced robustness based on factor graph

In this paper, we present a new approach to solving the distributed facility location problem using the recent modeling and computational methodology of factor graph and message-passing. We first formulate the problem as finding a valid network configuration that minimizes the overall cost. We then represent the problem using a factor graph, and derive simplified, localized, broadcast-based message-passing rules which can elect a near-optimal set of facility nodes in a few iterations. Simulation results for small-world network topologies show that the algorithm is able to achieve good convergence rate and approximation ratio, and scalable to the network size.

A model-based approach to robot kinematics and control using discrete factor graphs with belief propagation

Passive localization based on time of arrival (TOA) measurements is investigated, where the transmitted signal is reflected by a passive target and then received at several distributed receivers. After collecting all measurements at receivers, we can determine the target location. The aim of this paper is to provide a unified factor graph-based framework for passive localization in wireless sensor networks based on TOA measurements. Relying on the linearization of range measurements, we construct a Forney-style factor graph model and conceive the corresponding Gaussian message passing algorithm to obtain the target location. It is shown that the factor graph can be readily modified for handling challenging scenarios such as uncertain receiver positions and link failures. Moreover, a distributed localization method based on consensus-aided operation is proposed for a large-scale resource constrained network operating without a fusion center. Furthermore, we derive the Cramer–Rao bound (CRB) to evaluate the performance of the proposed algorithm. Our simulation results verify the efficiency of the proposed unified approach and of its distributed implementation.

Existing technologies employ different machine learning approachesto predict disasters from historical environmental data. However,for short-term disasters (e.g., earthquakes), historical data alonehas a limited prediction capability. In this work, we consider so-cial media as a supplementary source of knowledge in additionto historical environmental data. Further, we build a joint modelthat learns from disaster-related tweets and environmental data toimprove prediction. We propose the combination of semantically-enriched word embedding to represent entities in tweets with theirsemantics representations computed with the traditionalword2vec.Our experiments show that our proposed approach outperformsthe accuracy of state-of-the-art models in disaster prediction

Event extraction is a sub-task of information extraction and is an important part of natural language processing. Depending on the range of features used, event extraction methods are classified as sentence-level or document-level. However, document-level event extraction is more practical for practical tasks. Document-level event extraction is a difficult task, as it requires features to be extracted from a larger amount of text to determine which span of text is the desired event element. However, most methods do not utilize both sentence-level and document-level features. In order to utilize hybrid feature information and fuse it, this paper proposes a document-level event extraction method that integrating hybrid features. The event extraction method is based on Dynamic Multi-Pooling Convolutional Neural Network (DMCNN) and Bi-directional Long Short-Term Memory (BiLSTM), combined with self-attention mechanisms and Conditional Random Field (CRF). We evaluate the model proposed in this paper on the MUC-4 dataset and the experimental results show that our proposed model outperforms previous work.

Topic detection in group chat has become a promising research due to the widely usage of Instant Messaging (IM) systems. Previous works mainly focus on improving the text similarity between two related messages by utilizing different weighting factors. However, the text similarity of related texts is likely to be zero (or near zero) due to the characteristics of short text messages in group chat. To solve this problem, an innovative topic detection method based on implicit reply which indicates chat messages interact with each other is proposed in this paper. The comparative experiments results on the datasets gathered from QQ groups demonstrate the superiority of the proposed method as compared to the baseline approaches.

Mining users'= topics of interest is one of the most important tasks for social media services. Given known topic associations for some fraction of the users in an online microblogging platform, our goal is to infer the topics of interest for the remaining users in the same site. Specifically, we proposed a novel bi-relational graph model to capture the interactions among users and their shared topics of interests. The proposed graph model contains two sub-graphs: one corresponds to users and the other corresponds to topics. Such a representation allows for effective exploitation of both user homophily relation and topic correlation simultaneously. This is in contrast with previous work where these two factors are considered in isolation. Subsequently, the user interest discovery problem is formulated as a multi-label learning problem on the bi-relational graph, with the goal to estimate the optimized associations between user nodes and topic nodes across the two sub-graphs. Our experiment is carried out with a complete month-long data collected from Twitter and Tumblr via GNIP Decahose1 and Firehose2 respectively. The large-scale studies shed light on the effectiveness of inferring user interests based on the underlying social connections.

Web 2.0 users generate and spread huge amounts of messages in online social media. Such user-generated contents are mixture of temporal topics (e.g., breaking events) and stable topics (e.g., user interests). Due to their different natures, it is important and useful to distinguish temporal topics from stable topics in social media. However, such a discrimination is very challenging because the user-generated texts in social media are very short in length and thus lack useful linguistic features for precise analysis using traditional approaches. In this paper, we propose a novel solution to detect both stable and temporal topics simultaneously from social media data. Specifically, a unified user-temporal mixture model is proposed to distinguish temporal topics from stable topics. To improve this model's performance, we design a regularization framework that exploits prior spatial information in a social network, as well as a burst-weighted smoothing scheme that exploits temporal prior information in the time dimension. We conduct extensive experiments to evaluate our proposal on two real data sets obtained from Del.icio.us and Twitter. The experimental results verify that our mixture model is able to distinguish temporal topics from stable topics in a single detection process. Our mixture model enhanced with the spatial regularization and the burst-weighted smoothing scheme significantly outperforms competitor approaches, in terms of topic detection accuracy and discrimination in stable and temporal topics.