Inferring and improving street maps with data-driven automation

Automatic map inference, data refinement, and machine-assisted map editing promises more accurate map datasets.

Urban Map Inference by Pervasive Vehicular Sensing Systems with Complementary Mobility

Accurate and up-to-date digital road maps are the foundation of many mobile applications, such as navigation and autonomous driving. A manually-created map suffers from the high cost for creation and maintenance due to constant road network updating. Recently, the ubiquity of GPS devices in vehicular systems has led to an unprecedented amount of vehicle sensing data for map inference. Unfortunately, accurate map inference based on vehicle GPS is challenging for two reasons. First, it is challenging to infer complete road structures due to the sensing deviation, sparse coverage, and low sampling rate of GPS of a fleet of vehicles with similar mobility patterns, e.g., taxis. Second, a road map requires various road properties such as road categories, which is challenging to be inferred by just GPS locations of vehicles. In this paper, we design a map inference system called coMap by considering multiple fleets of vehicles with Complementary Mobility Features. coMap has two key components: a graph-based map sketching component, a learning-based map painting component. We implement coMap with the data from four type-aware vehicular sensing systems in one city, which consists of 18 thousand taxis, 10 thousand private vehicles, 6 thousand trucks, and 14 thousand buses. We conduct a comprehensive evaluation of coMap with two state-of-the-art baselines along with ground truth based on OpenStreetMap and a commercial map provider, i.e., Baidu Maps. The results show that (i) for the map sketching, our work improves the performance by 15.9%; (ii) for the map painting, our work achieves 74.58% of average accuracy on road category classification.

A Map Inference Approach Using Signal Processing from Crowd-sourced GPS Data

The amount of GPS data that can be collected is increasing tremendously, thanks to the increased popularity of Global Position System (GPS) devices (e.g., smartphones). This article aims to develop novel methods of converting crowd-sourced GPS traces into road topology maps. We explore map inference using a three-stage approach, which incorporates a novel Multi-source Variable Rate (MSVR) signal reconstruction mechanism. Unlike conventional map inference methods based on map graph theory, our approach, to the best of our knowledge, is the first use of estimation theory for map inference. In particular, our approach addresses the unique challenges of vehicular GPS data. This data is plentiful but suffers from noise in location and variable coverage of regions. This makes it difficult to differentiate between noise and sparsely covered regions when increasing coverage and reducing noise. Due to the asynchronous, variable sampling rate, and often under-sampled nature of the data, our MSVR approach can better handle inherent GPS errors, reconstruct road shapes more accurately, and better deal with variable GPS data density in empirical environments. We evaluated our method for map inference by comparing to Open Street Map maps as ground truth. We use the F-Measure, Precision, and Recall metrics to evaluate our method on Tsinghua University’s Beijing Taxi Dataset and Shanghai Jiao Tong University’s SUVnet Dataset. On these datasets, we obtained a mean<?brk?> F-Measure, Precision, and Recall of 0.7212, 0.9165, and 0.6021, respectively, outperforming a well-known method based on Kernel Density Estimation in terms of these evaluation metrics.

MAP Inference for Probabilistic Logic Programming

In Probabilistic Logic Programming (PLP) the most commonly studied inference task is to compute the marginal probability of a query given a program. In this paper, we consider two other important tasks in the PLP setting: the Maximum-A-Posteriori (MAP) inference task, which determines the most likely values for a subset of the random variables given evidence on other variables, and the Most Probable Explanation (MPE) task, the instance of MAP where the query variables are the complement of the evidence variables. We present a novel algorithm, included in the PITA reasoner, which tackles these tasks by representing each problem as a Binary Decision Diagram and applying a dynamic programming procedure on it. We compare our algorithm with the version of ProbLog that admits annotated disjunctions and can perform MAP and MPE inference. Experiments on several synthetic datasets show that PITA outperforms ProbLog in many cases.

Exact and Efficient Inference for Collective Flow Diffusion Model via Minimum Convex Cost Flow Algorithm

Collective Flow Diffusion Model (CFDM) is a general framework to find the hidden movements underlying aggregated population data. The key procedure in CFDM analysis is MAP inference of hidden variables. Unfortunately, existing approaches fail to offer exact MAP inferences, only approximate versions, and take a lot of computation time when applied to large scale problems. In this paper, we propose an exact and efficient method for MAP inference in CFDM. Our key idea is formulating the MAP inference problem as a combinatorial optimization problem called Minimum Convex Cost Flow Problem (C-MCFP) with no approximation or continuous relaxation. On the basis of this formulation, we propose an efficient inference method that employs the C-MCFP algorithm as a subroutine. Our experiments on synthetic and real datasets show that the proposed method is effective both in single MAP inference and people flow estimation with EM algorithm.

High-performance sampling of generic determinantal point processes

Determinantal point processes (DPPs) were introduced by Macchi (Macchi 1975 Adv. Appl. Probab. 7 , 83–122) as a model for repulsive (fermionic) particle distributions. But their recent popularization is largely due to their usefulness for encouraging diversity in the final stage of a recommender system (Kulesza & Taskar 2012 Found. Trends Mach. Learn. 5 , 123–286). The standard sampling scheme for finite DPPs is a spectral decomposition followed by an equivalent of a randomly diagonally pivoted Cholesky factorization of an orthogonal projection, which is only applicable to Hermitian kernels and has an expensive set-up cost. Researchers Launay et al. 2018 ( http://arxiv.org/abs/1802.08429 ); Chen & Zhang 2018 NeurIPS ( https://papers.nips.cc/paper/7805-fast-greedy-map-inference-for-determinantal-point-process-to-improve-recommendation-diversity.pdf ) have begun to connect DPP sampling to LDL H factorizations as a means of avoiding the initial spectral decomposition, but existing approaches have only outperformed the spectral decomposition approach in special circumstances, where the number of kept modes is a small percentage of the ground set size. This article proves that trivial modifications of LU and LDL H factorizations yield efficient direct sampling schemes for non-Hermitian and Hermitian DPP kernels, respectively. Furthermore, it is experimentally shown that even dynamically scheduled, shared-memory parallelizations of high-performance dense and sparse-direct factorizations can be trivially modified to yield DPP sampling schemes with essentially identical performance. The software developed as part of this research, Catamari ( hodgestar.com/catamari ) is released under the Mozilla Public License v.2.0. It contains header-only, C++14 plus OpenMP 4.0 implementations of dense and sparse-direct, Hermitian and non-Hermitian DPP samplers. This article is part of a discussion meeting issue ‘Numerical algorithms for high-performance computational science’.