Abhinav Jauhri, Brad Stocks, Jian Hui Li, Koichi Yamada, John Paul Shen, 2020. Generating realistic ride-hailing datasets using GANs, ACM Transactions of Spatial Algorithms and Systems. [official]
This article focuses on the synthetic generation of human mobility data in urban areas. We present a novel application of generative adversarial networks (GANs) for modeling and generating human mobility data. We leverage actual ride requests from ride-sharing/hailing services from four major cities to train our GANs model. Our model captures the spatial and temporal variability of the ride request patterns observed for all four cities over a typical week. Previous works have characterized the spatial and temporal properties of human mobility datasets using the fractal dimensionality and the densification power law, respectively, which we utilize to validate our GANs-generated synthetic datasets. We also validate the synthetic datasets using a dynamic vehicle placement application. Such synthetic datasets can avoid privacy concerns and be extremely useful for researchers and policy makers on urban mobility.

Abhinav Jauhri, Carlee Joe-Wong, John Paul Shen, 2017. On the Real-Time Vehicle Placement Problem, NIPS Workshop. [OpenReview] [arXiv] [slides]
Motivated by ride-sharing platforms' efforts to reduce their riders' wait times for a vehicle, this paper introduces a novel problem of placing vehicles to fulfill real-time pickup requests in a spatially and temporally changing environment. The real-time nature of this problem makes it fundamentally different from other placement and scheduling problems, as it requires not only real-time placement decisions but also handling real-time request dynamics, which are influenced by human mobility patterns. We use a dataset of ten million ride requests from four major U.S. cities to show that the requests exhibit significant self-similarity. We then propose distributed online learning algorithms for the real-time vehicle placement problem and bound their expected performance under this observed self-similarity.

Min Hao Chen, Abhinav Jauhri, John Paul Shen, 2017. Data Driven Analysis of the Potentials of Dynamic Ride Pooling, IWCTS. [official] [slides]
This paper focuses on the challenge of dynamically pooling multiple ride requests in real time in order to achieve greater overall efficiency for ride sharing/hailing services. A rigorous formulation of this problem and an efficient pooling method is introduced. This paper adopts a data-driven approach and uses an extensive ride-request data set from the real world to evaluate this method. The experimental results based on the data set for three US cities show that close to 50% of all ride requests can be pooled, and there are significant benefits for both riders and services. Furthermore dynamic ride pooling can potentially yield significant societal benefits in reducing total fuel consumption (by 15%) and alleviating traffic congestion by reducing the total vehicle count (by 30%).

Abhinav Jauhri, Brian Foo, Jérôme Berclaz, Chih Hu, Radek Grzeszczuk, Vasu Parameswaran, John Paul Shen, 2017. Space-Time Graph Modeling of Ride Requests Based on Real-World Data, Workshop on AI and OR for Social Good, AAAI [official] [arXiv] [slides]
This paper focuses on modeling variations in ride requests over geographical space and time from ride-sharing services based on analyzing extensive real-world data. We introduce a graph model that captures the spatial and temporal attributes of ride requests and ride pooling. Using this graph model, we show that the pattern of ride requests and the potential of ride pooling for a city can be characterized by an attribute representing the densification of the graph over time. We show that there is significant variation of this attribute across cities. Furthermore, these graphs belong to a class of graphs with special properties that make it possible to automatically generate synthetic versions that exhibit very similar attributes as the extracted versions from real data. Given these properties, future research can potentially be based on synthetic graphs without requiring access to the real data.

Abhinav Jauhri, Martin Griss and Hakan Erdogmus, 2016. Small Polygon Compression, DCC 2016 & 3rd Conference on Weather Warnings and Communication [arXiv] [talk] [poster] [official] [code]
We decribe several polygon compression techniques to enable efficient transmission of polygons representing geographical targets. The main application is to embed compressed polygons to emergency alert messages that have strict length restrictions, as in the case of Wireless Emergency Alert messages. We are able to compress polygons to between 9.7% and 23.6% of original length, depending on characteristics of the specific polygons, reducing original polygon lengths from 43-331 characters to 8-55 characters. The best techniques apply several heuristics to perform initial compression, and then other algorithmic techniques, including higher base encoding. Further, these methods are respectful of computation and storage constraints typical of cell phones. Two of the best techniques include a “bignum” quadratic combination of integer coordinates and a variable length encoding, which takes advantage of a strongly skewed polygon coordinate distribution. Both techniques applied to one of two “delta” representations of polygons are on average able to reduce the size of polygons by some 80%. A repeated substring dictionary can provide further compression, and a merger of these techniques into a “polyalgorithm” can also provide additional improvements.

Abhinav Jauhri, Bradley McDanel and Chris Connor, 2015. Outlier Detection for Large Scale Manufacturing Processes, IEEE Big Data [official] [pdf]
Integrated circuit manufacturing consists of tests at various stages to ensure functionality and performance using numerous test metrics for each system on chip (SoC) captured as part of assessment. At a later stage, functional units are evaluated in terms of multiple performance characteristics. In this paper, we propose a system that uses test metrics as features for machine learning models to predict the performance characteristics of each SoC. We show that these models are robust against erroneous or noisy signal in test metrics and provide accurate prediction. Given accurate models, we build a system that automatically detects systematic changes in the manufacturing process from week to week and identifies wafers, a grouping of patterned dies in the fabrication process, which have significantly higher than average prediction error and label them as outliers. These outliers are analyzed in order to determine the cause of the discrepancy and to assess potential problems in the manufacturing process. The system has been proven applicable across multiple products and process technologies.

Abhinav Jauhri, Jason D. Lohn and Derek S. Linden, 2014. A Comparison of Antenna Placement Algorithms, PURO Workshop, GECCO [official] [pdf]
Co-location of multiple antenna systems on a single fixed or mobile platform can be challenging due to a variety of factors, such as mutual coupling, individual antenna constraints, multipath, obstructions, and parasitic effects due to the platform. The situation frequently arises where a new communication capability, and hence antenna system, is needed on an existing platform. The problem of placing new antennas requires a long, manual effort in order to complete an antenna placement study. An automated procedure for determining such placements would not only save time, but would be able to optimize the performance of all co-located antenna systems. In this paper we examine a set of stochastic algorithms to determine their effectiveness at finding optimal placements for multiple antennas on a platform. Of the four algorithms studied, simulated annealing and evolutionary strategy were found to be most effective in finding optimal placements.