Tight Computational Budgets Research Articles

Federated Robustness Propagation: Sharing Adversarial Robustness in Heterogeneous Federated Learning.

Federated learning (FL) emerges as a popular distributed learning schema that learns a model from a set of participating users without sharing raw data. One major challenge of FL comes with heterogeneous users, who may have distributionally different (or non-iid) data and varying computation resources. As federated users would use the model for prediction, they often demand the trained model to be robust against malicious attackers at test time. Whereas adversarial training (AT) provides a sound solution for centralized learning, extending its usage for federated users has imposed significant challenges, as many users may have very limited training data and tight computational budgets, to afford the data-hungry and costly AT. In this paper, we study a novel FL strategy: propagating adversarial robustness from rich-resource users that can afford AT, to those with poor resources that cannot afford it, during federated learning. We show that existing FL techniques cannot be effectively integrated with the strategy to propagate robustness among non-iid users and propose an efficient propagation approach by the proper use of batch-normalization. We demonstrate the rationality and effectiveness of our method through extensive experiments. Especially, the proposed method is shown to grant federated models remarkable robustness even when only a small portion of users afford AT during learning. Source code will be released.

High Dimensional Origin Destination Calibration Using Metamodel Assisted Simultaneous Perturbation Stochastic Approximation

The huge traffic data generated by intelligent transportation system (ITS) leads to the development of many advanced traffic models. These traffic models consist of many adjustable parameters which need to be calibrated before they are used in practice. This paper focuses on the offline calibration of origin-destination (OD) input parameters of a simulation-based traffic model to match its output with sensor data. An improved Simultaneous Perturbation Stochastic Approximation (SPSA) algorithm called Metamodel Assisted Simultaneous Perturbation Stochastic Approximation (MSPSA) is proposed in this paper to calibrate high-dimensional OD parameters within a tight computational budget. The proposed MSPSA combines the gradient of SPSA with the gradient of a differentiable metamodel function to improve the calibration efficiency. An integer program is also used to fine-tune the OD estimates. The proposed MSPSA algorithm is tested on a simple synthetic toy network and complex road network of Kuala Lumpur (KL), Malaysia. The proposed MSPSA algorithm is compared against SPSA and state-of-the-art Weighted Simultaneous Perturbation Stochastic Approximation (WSPSA) in both transportation networks. For KL network, synthetic and real-world sensor measurements are used as ground truth references to evaluate the performance of each approach. Based on the simulation results, the proposed MSPSA algorithm is able to gain at least 50% of improvement as compared to SPSA and WSPSA in both synthetic and real-world scenarios.

A Simulation-Based Model for Continuous Network Design Problem Using Bayesian Optimization

This paper investigates the continuous network design problem (CNDP) and proposes a simulation-based bi-level model and solution framework based on Bayesian optimization. In the bi-level model, the upper level minimizes the total system cost, and the lower level assigns traffic under an approximated dynamic equilibrium condition corresponding to the given network design strategy. A simulation tool integrated micro- and macro- traffic dynamics, namely SUMO, is employed to solve the lower-level problem. The embedded high-fidelity, non-linear understanding of traffic in the simulator oftentimes gains additional complexity due to the lack of tractable mathematical representation. Thus, a Bayesian machine learning technique is utilized to build the link between simulation and optimization. The proposed solution framework takes both the advantages of fine-grained simulations and the efficiency of surrogate-based optimization techniques. Moreover, lane width expansion is innovatively proposed as the decision variable of CNDP to bridge the gap between theory and practice. The relationship between link free-flow speed (FFS) and lane width is also explored based on real data and established to accurately calibrate the simulation input. For demonstrative purposes, numerical experiments on the optimal lane width expansion design were conducted in the inner city of Suzhou, China. The results show that with proper parameter settings, the proposed method is capable to find the global optimal solution within a very tight computational budget, which makes the simulation-based framework an encouraging option for policymakers to enhance transportation network performance.

Efficient calibration of microscopic car-following models for large-scale stochastic network simulators

This paper proposes a simulation-based optimization methodology for the efficient calibration of microscopic traffic flow models (i.e., car-following models) of large-scale stochastic network simulators. The approach is a metamodel simulation-based optimization (SO) method. To improve computational efficiency of the SO algorithm, problem-specific and simulator-specific structural information is embedded into a metamodel. As a closed-form expression is sought, we propose adopting the steady-state solution of the car-following model as an approximation of its simulation-based input-output mapping. This general approach is applied for the calibration of the Gipps car-following model embedded in a microscopic traffic network simulator, on a large network. To this end, a novel formulation for the traffic stream models corresponding to the Gipps car-following law is provided.The proposed approach identifies points with good performance within few simulation runs. Comparing its performances to that of a traditional approach, which does not take advantage of the structural information, the objective function is improved by two orders of magnitude in most experiments. Moreover, this is achieved within tight computational budgets, i.e., few simulation runs. The solutions identified improve the fit to the field measurements by one order of magnitude, on average. The structural information provided to the metamodel is shown to enable the SO algorithm to become robust to both the quality of the initial points and the simulator stochasticity.

A search acceleration method for optimization problems with transport simulation constraints

This work contributes to the rapid approximation of solutions to optimization problems that are constrained by iteratively solved transport simulations. Given an objective function, a set of candidate decision variables and a black-box transport simulation that is solved by iteratively attaining a (deterministic or stochastic) equilibrium, the proposed method approximates the best decision variable out of the candidate set without having to run the transport simulation to convergence for every single candidate decision variable. This method can be inserted into a broad class of optimization algorithms or search heuristics that implement the following logic: (i) Create variations of a given, currently best decision variable, (ii) identify one out of these variations as the new currently best decision variable, and (iii) iterate steps (i) and (ii) until no further improvement can be attained. A probabilistic and an asymptotic performance bound are established and exploited in the formulation of an efficient heuristic that is tailored towards tight computational budgets. The efficiency of the method is substantiated through a comprehensive simulation study with a non-trivial road pricing problem. The method is compatible with a broad range of simulators and requires minimal parametrization.

Time-of-day vehicle mileage fees for congestion mitigation and revenue generation: A simulation-based optimization method and its real-world application

Congestion pricing of a large-scale network is characterized by expensive-to-evaluate objective functions without closed forms. This paper further enhances a computationally efficient simulation-based optimization (SBO) framework to solve the problem within tight computational budget. This paper applies surrogate models to solve the optimization problem with computationally expensive objective functions based on simulation-based dynamic traffic assignment (DTA). DIRECT (a deterministic search algorithm with modification to Lipschitzian optimization) is used for metamodel parameter tuning. A trade-off of different objectives (i.e. the average travel time minimization, expected network throughput maximization, and toll revenue maximization) are converted into a single desirability function. To demonstrate the SBO framework with an application to the vehicle mileage traveled (VMT) based pricing for a real-world freeway network, this paper utilizes a calibrated simulation-based DTA model to evaluate system performance. A stochastic mesoscopic simulator is applied. We investigate the existence of an invariant macroscopic fundamental diagram (MFD) for the network, and compare simulated MFDs with measurements of fixed detectors and probe data. The proposed SBO framework is generic and can be used to solve other congestion pricing problems.

A Computationally Efficient Simulation-Based Optimization Algorithm for Large-Scale Urban Transportation Problems

This paper proposes a computationally efficient simulation-based optimization (SO) algorithm suitable to address large-scale generally constrained urban transportation problems. The algorithm is based on a novel metamodel formulation. We embed the metamodel within a derivative-free trust region algorithm and evaluate the performance of this SO approach considering tight computational budgets. We address a network-wide traffic signal control problem using a calibrated microscopic simulation model of evening peak period traffic of the full city of Lausanne, Switzerland, which consists of more than 600 links and 200 intersections. We control 99 signal phases of 17 intersections distributed throughout the entire network. This SO problem is a high-dimensional nonlinear constrained problem. It is considered large-scale and complex in the fields of derivative-free optimization, traffic signal optimization, and simulation-based optimization. We compare the performance of the proposed metamodel method to that of a traditional metamodel method and that of a widely used commercial signal control software. The proposed method systematically and efficiently identifies signal plans with improved average city-wide travel times.

Energy-Efficient Urban Traffic Management: A Microscopic Simulation-Based Approach

Microscopic urban traffic simulators embed the most detailed traveler behavior and network supply models. These simulators represent individual vehicles and can therefore account for vehicle-specific technologies. They can be coupled with instantaneous fuel consumption models to yield detailed network-wide fuel consumption estimates. Nonetheless, there is currently a lack of computationally efficient optimization techniques that enable the use of these complex integrated models to design sustainable transportation strategies. This paper proposes a methodology that combines a stochastic microscopic traffic simulation model with an instantaneous vehicular fuel consumption model. The combined models are embedded within a simulation-based optimization algorithm and used to address a signal control problem that accounts for both travel times and fuel consumption. The proposed technique couples detailed, stochastic, and computationally inefficient models, yet is an efficient optimization technique. Efficiency is achieved by combining simulated observations with analytical approximations of both travel time and fuel consumption. This methodology is applied to a network in the Swiss city of Lausanne. Within a tight computational budget, the proposed method identifies signal plans with improved travel time and fuel consumption metrics. It outperforms traditional methodologies, which use only simulated information or only analytical information. The case study illustrates the added value of combining simulated and analytical information when performance metrics with high variance, such as fuel consumption, are used. This method enables the use of disaggregate instantaneous vehicle-specific information to inform and improve traffic operations at the network-scale.

Urban transportation emissions mitigation: Coupling high-resolution vehicular emissions and traffic models for traffic signal optimization

This paper proposes a methodology that allows high-resolution traffic and emissions models, known as microscopic simulation models, to be efficiently used to address transportation optimization problems that account for complex environmental metrics. The methodology consists of a metamodel simulation-based optimization (SO) approach. The metamodel combines traffic and emissions information from high-resolution microscopic simulators with information from lower-resolution analytical macroscopic models. This paper formulates and uses an analytical and differentiable macroscopic approximation of the non-differentiable simulation-based microscopic emissions model. A differentiable macroscopic traffic model is also used.This paper shows that the analytical structural information provided by macroscopic analytical emissions models can contribute, despite their lower-resolution, to enhance the computational efficiency of algorithms that embed higher-resolution inefficient emissions models. The proposed algorithm is computationally efficient, i.e., it can identify points with improved performance within few simulation runs. More generally, the results of this paper highlight the added value of embedding analytical structural information within SO algorithms to address complex SO problems.A traffic signal control case study is carried out. The proposed metamodel identifies signal plans that improve travel time and various emissions metrics. We present the corresponding monetary savings that can be achieved.This optimization framework enables practitioners to use high-resolution microscopic traffic and emissions models to systematically design transportation schemes that account directly, in the design process, for environmental metrics. Hence, the use of such high-resolution models is no longer limited to the environmental evaluation of a small set of predetermined schemes. The tight computational budgets used in this paper show that such complex problems can be addressed in a computationally efficient manner.

A Simulation-Based Optimization Framework for Urban Transportation Problems

This paper proposes a simulation-based optimization (SO) method that enables the efficient use of complex stochastic urban traffic simulators to address various transportation problems. It presents a metamodel that integrates information from a simulator with an analytical queueing network model. The proposed metamodel combines a general-purpose component (a quadratic polynomial), which provides a detailed local approximation, with a physical component (the analytical queueing network model), which provides tractable analytical and global information. This combination leads to an SO framework that is computationally efficient and suitable for complex problems with very tight computational budgets. We integrate this metamodel within a derivative-free trust region algorithm. We evaluate the performance of this method considering a traffic signal control problem for the Swiss city of Lausanne, different demand scenarios, and tight computational budgets. The method leads to well-performing signal plans. It leads to reduced, as well as more reliable, average travel times.