We study iterative methods for (two-stage) robust combinatorial optimization problems with discrete uncertainty. We propose a machine-learning-based heuristic to determine starting scenarios that provide strong lower bounds. To this end, we design dimension-independent features and train a Random Forest Classifier on small-dimensional instances. Experiments show that our method improves the solution process for larger instances than contained in the training set and also provides a feature importance-score which gives insights into the role of scenario properties.
Multiobjective combinatorial optimization (MOCO) problems can be found in many real-world applications. However, exactly solving these problems would be very challenging, particularly when they are NP-hard. Many handcrafted heuristic methods have been proposed to tackle different MOCO problems over the past decades. In this work, we generalize the idea of neural combinatorial optimization, and develop a learning-based approach to approximate the whole Pareto set for a given MOCO problem without further search procedure. We propose a single preference-conditioned model to directly generate approximate Pareto solutions for any trade-off preference, and design an efficient multiobjective reinforcement learning algorithm to train this model. Our proposed method can be treated as a learning-based extension for the widely-used decomposition-based multiobjective evolutionary algorithm (MOEA/D). It uses a single model to accommodate all the possible preferences, whereas other methods use a finite number of solutions to approximate the Pareto set. Experimental results show that our proposed method significantly outperforms some other methods on the multiobjective traveling salesman problem, multiobjective vehicle routing problem, and multiobjective knapsack problem in terms of solution quality, speed, and model efficiency.
Graph coloring is a challenging combinatorial optimization problem with a wide range of applications. In this paper, a distribution evolutionary algorithm based on a population of probability model (DEA-PPM) is developed to address it efficiently. Unlike existing estimation of distribution algorithms where a probability model is updated by generated solutions, DEA-PPM employs a distribution population based on a novel probability model, and an orthogonal exploration strategy is introduced to search the distribution space with the assistance of an refinement strategy. By sampling the distribution population, efficient search in the solution space is realized based on a tabu search process. Meanwhile, DEA-PPM introduces an iterative vertex removal strategy to improve the efficiency of $k$-coloring, and an inherited initialization strategy is implemented to address the chromatic problem well. The cooperative evolution of the distribution population and the solution population leads to a good balance between exploration and exploitation. Numerical results demonstrate that the DEA-PPM of small population size is competitive to the state-of-the-art metaheuristics.utes to its competitiveness to the state-of-the-art metaheuristics.
Repair operators are often used for constraint handling in constrained combinatorial optimization. We investigate the (1+1)~EA equipped with a tailored jump-and-repair operation that can be used to probabilistically repair infeasible offspring in graph problems. Instead of evolving candidate solutions to the entire graph, we expand the genotype to allow the (1+1)~EA to develop in parallel a feasible solution together with a growing subset of the instance (an induced subgraph). With this approach, we prove that the EA is able to probabilistically simulate an iterative compression process used in classical fixed-parameter algorithmics to obtain a randomized FPT performance guarantee on three NP-hard graph problems. For $k$-VertexCover, we prove that the (1+1) EA using focused jump-and-repair can find a $k$-vertex cover (if one exists) in $O(2^k n^2\log n)$ iterations in expectation. This leads to an exponential (in $k$) improvement over the best-known parameterized bound for evolutionary algorithms on VertexCover. For the $k$-FeedbackVertexSet problem in tournaments, we prove that the EA finds a feasible feedback set in $O(2^kk!n^2\log n)$ iterations in expectation, and for OddCycleTransversal, we prove the optimization time for the EA is $O(3^k k m n^2\log n)$. For the latter two problems, this constitutes the first parameterized result for any evolutionary algorithm. We discuss how to generalize the framework to other parameterized graph problems closed under induced subgraphs and report experimental results that illustrate the behavior of the algorithm on a concrete instance class.
Molecule design is a fundamental problem in molecular science and has critical applications in a variety of areas, such as drug discovery, material science, etc. However, due to the large searching space, it is impossible for human experts to enumerate and test all molecules in wet-lab experiments. Recently, with the rapid development of machine learning methods, especially generative methods, molecule design has achieved great progress by leveraging machine learning models to generate candidate molecules. In this paper, we systematically review the most relevant work in machine learning models for molecule design. We start with a brief review of the mainstream molecule featurization and representation methods (including 1D string, 2D graph, and 3D geometry) and general generative methods (deep generative and combinatorial optimization methods). Then we summarize all the existing molecule design problems into several venues according to the problem setup, including input, output types and goals. Finally, we conclude with the open challenges and point out future opportunities of machine learning models for molecule design in real-world applications.
This paper proposes a method to compute camera 6Dof poses to achieve a user defined coverage. The camera placement problem is modeled as a combinatorial optimization where given the maximum number of cameras, a camera set is selected from a larger pool of possible camera poses. We propose to minimize the squared error between the desired and the achieved coverage, and formulate the non-linear cost function as a mixed integer linear programming problem. A camera lens model is utilized to project the cameras view on a 3D voxel map to compute a coverage score which makes the optimization problem in real environments tractable. Experimental results in two real retail store environments demonstrate the better performance of the proposed formulation in terms of coverage and overlap for triangulation compared to existing methods.
This paper introduces a new fixed effects estimator for linear panel data models with clustered time patterns of unobserved heterogeneity. The method avoids non-convex and combinatorial optimization by combining a preliminary consistent estimator of the slope coefficient, an agglomerative pairwise-differencing clustering of cross-sectional units, and a pooled ordinary least squares regression. Asymptotic guarantees are established in a framework where $T$ can grow at any power of $N$, as both $N$ and $T$ approach infinity. Unlike most existing approaches, the proposed estimator is computationally straightforward and does not require a known upper bound on the number of groups. As existing approaches, this method leads to a consistent estimation of well-separated groups and an estimator of common parameters asymptotically equivalent to the infeasible regression controlling for the true groups. An application revisits the statistical association between income and democracy.
The success of machine learning solutions for reasoning about discrete structures has brought attention to its adoption within combinatorial optimization algorithms. Such approaches generally rely on supervised learning by leveraging datasets of the combinatorial structures of interest drawn from some distribution of problem instances. Reinforcement learning has also been employed to find such structures. In this paper, we propose a radically different approach in that no data is required for training the neural networks that produce the solution. In particular, we reduce the combinatorial optimization problem to a neural network and employ a dataless training scheme to refine the parameters of the network such that those parameters yield the structure of interest. We consider the combinatorial optimization problems of finding maximum independent sets and maximum cliques in a graph. In principle, since these problems belong to the NP-hard complexity class, our proposed approach can be used to solve any other NP-hard problem. Additionally, we propose a universal graph reduction procedure to handle large scale graphs. The reduction exploits community detection for graph partitioning and is applicable to any graph type and/or density. Experimental evaluation on both synthetic graphs and real-world benchmarks demonstrates that our method performs on par with or outperforms state-of-the-art heuristic, reinforcement learning, and machine learning based methods without requiring any data.
Cloud computing is a reliable solution to provide distributed computation power. However, real-time response is still challenging regarding the enormous amount of data generated by the IoT devices in 5G and 6G networks. Thus, multi-access edge computing (MEC), which consists of distributing the edge servers in the proximity of end-users to have low latency besides the higher processing power, is increasingly becoming a vital factor for the success of modern applications. This paper addresses the problem of minimizing both, the network delay, which is the main objective of MEC, and the number of edge servers to provide a MEC design with minimum cost. This MEC design consists of edge servers placement and base stations allocation, which makes it a joint combinatorial optimization problem (COP). Recently, reinforcement learning (RL) has shown promising results for COPs. However, modeling real-world problems using RL when the state and action spaces are large still needs investigation. We propose a novel RL framework with an efficient representation and modeling of the state space, action space and the penalty function in the design of the underlying Markov Decision Process (MDP) for solving our problem.
The Conflict-Free Electric Vehicle Routing Problem (CF-EVRP) is an extension of the Vehicle Routing Problem (VRP), a combinatorial optimization problem of designing routes for vehicles to visit customers such that a cost function, typically the number of vehicles or the total travelled distance, is minimized. The problem finds many logistics applications, particularly for highly automated logistic systems for material handling. The CF-EVRP involves constraints such as time windows on the delivery to the customers, limited operating range of the vehicles, and limited capacity on the number of vehicles that a road segment can accommodate at the same time. In this paper, the compositional algorithm ComSat for solving the CF-EVRP is presented. The algorithm iterates through the sub-problems until a globally feasible solution is found. The proposed algorithm is implemented using an optimizing SMT-solver and is evaluated against an implementation of a previously presented monolithic model. The soundness and completeness of the algorithm are proven, and it is benchmarked on a set of generated problems and found to be able to solve problems of industrial size.
Set-valued prediction is a well-known concept in multi-class classification. When a classifier is uncertain about the class label for a test instance, it can predict a set of classes instead of a single class. In this paper, we focus on hierarchical multi-class classification problems, where valid sets (typically) correspond to internal nodes of the hierarchy. We argue that this is a very strong restriction, and we propose a relaxation by introducing the notion of representation complexity for a predicted set. In combination with probabilistic classifiers, this leads to a challenging inference problem for which specific combinatorial optimization algorithms are needed. We propose three methods and evaluate them on benchmark datasets: a na\"ive approach that is based on matrix-vector multiplication, a reformulation as a knapsack problem with conflict graph, and a recursive tree search method. Experimental results demonstrate that the last method is computationally more efficient than the other two approaches, due to a hierarchical factorization of the conditional class distribution.
Random convolution kernel transform (Rocket) is a fast, efficient, and novel approach for time series feature extraction using a large number of independent randomly initialized 1-D convolution kernels of different configurations. The output of the convolution operation on each time series is represented by a partial positive value (PPV). A concatenation of PPVs from all kernels is the input feature vector to a Ridge regression classifier. Unlike typical deep learning models, the kernels are not trained and there is no weighted/trainable connection between kernels or concatenated features and the classifier. Since these kernels are generated randomly, a portion of these kernels may not positively contribute in performance of the model. Hence, selection of the most important kernels and pruning the redundant and less important ones is necessary to reduce computational complexity and accelerate inference of Rocket for applications on the edge devices. Selection of these kernels is a combinatorial optimization problem. In this paper, we propose a scheme for selecting these kernels while maintaining the classification performance. First, the original model is pre-trained at full capacity. Then, a population of binary candidate state vectors is initialized where each element of a vector represents the active/inactive status of a kernel. A population-based optimization algorithm evolves the population in order to find a best state vector which minimizes the number of active kernels while maximizing the accuracy of the classifier. This activation function is a linear combination of the total number of active kernels and the classification accuracy of the pre-trained classifier with the active kernels. Finally, the selected kernels in the best state vector are utilized to train the Ridge regression classifier with the selected kernels.
This paper surveys the trend of leveraging machine learning to solve mixed integer programming (MIP) problems. Theoretically, MIP is an NP-hard problem, and most of the combinatorial optimization (CO) problems can be formulated as the MIP. Like other CO problems, the human-designed heuristic algorithms for MIP rely on good initial solutions and cost a lot of computational resources. Therefore, we consider applying machine learning methods to solve MIP, since ML-enhanced approaches can provide the solution based on the typical patterns from the historical data. In this paper, we first introduce the formulation and preliminaries of MIP and several traditional algorithms to solve MIP. Then, we advocate further promoting the different integration of machine learning and MIP and introducing related learning-based methods, which can be classified into exact algorithms and heuristic algorithms. Finally, we propose the outlook for learning-based MIP solvers, direction towards more combinatorial optimization problems beyond MIP, and also the mutual embrace of traditional solvers and machine learning components.
Combinatorial optimization is a well-established area in operations research and computer science. Until recently, its methods have focused on solving problem instances in isolation, ignoring that they often stem from related data distributions in practice. However, recent years have seen a surge of interest in using machine learning as a new approach for solving combinatorial problems, either directly as solvers or by enhancing exact solvers. Based on this context, the ML4CO aims at improving state-of-the-art combinatorial optimization solvers by replacing key heuristic components. The competition featured three challenging tasks: finding the best feasible solution, producing the tightest optimality certificate, and giving an appropriate solver configuration. Three realistic datasets were considered: balanced item placement, workload apportionment, and maritime inventory routing. This last dataset was kept anonymous for the contestants.
The traveling salesman problem is a fundamental combinatorial optimization problem with strong exact algorithms. However, as problems scale up, these exact algorithms fail to provide a solution in a reasonable time. To resolve this, current works look at utilizing deep learning to construct reasonable solutions. Such efforts have been very successful, but tend to be slow and compute intensive. This paper exemplifies the integration of entropic regularized optimal transport techniques as a layer in a deep reinforcement learning network. We show that we can construct a model capable of learning without supervision and inferences significantly faster than current autoregressive approaches. We also empirically evaluate the benefits of including optimal transport algorithms within deep learning models to enforce assignment constraints during end-to-end training.
The simultaneous stochastic optimization of mining complexes (SSOMC) is a large-scale stochastic combinatorial optimization problem that simultaneously manages the extraction of materials from multiple mines and their processing using interconnected facilities to generate a set of final products, while taking into account material supply (geological) uncertainty to manage the associated risk. Although simulated annealing has been shown to outperform comparing methods for solving the SSOMC, early performance might dominate recent performance in that a combination of the heuristics' performance is used to determine which perturbations to apply. This work proposes a data-driven framework for heuristic scheduling in a fully self-managed hyper-heuristic to solve the SSOMC. The proposed learn-to-perturb (L2P) hyper-heuristic is a multi-neighborhood simulated annealing algorithm. The L2P selects the heuristic (perturbation) to be applied in a self-adaptive manner using reinforcement learning to efficiently explore which local search is best suited for a particular search point. Several state-of-the-art agents have been incorporated into L2P to better adapt the search and guide it towards better solutions. By learning from data describing the performance of the heuristics, a problem-specific ordering of heuristics that collectively finds better solutions faster is obtained. L2P is tested on several real-world mining complexes, with an emphasis on efficiency, robustness, and generalization capacity. Results show a reduction in the number of iterations by 30-50% and in the computational time by 30-45%.
The bin packing problem exists widely in real logistic scenarios (e.g., packing pipeline, express delivery), with its goal to improve the packing efficiency and reduce the transportation cost. In this NP-hard combinatorial optimization problem, the position and quantity of each item in the box are strictly restricted by complex constraints and special customer requirements. Existing approaches are hard to obtain the optimal solution since rigorous constraints cannot be handled within a reasonable computation load. In this paper, for handling this difficulty, the packing knowledge is extracted from historical data collected from the packing pipeline of Huawei. First, by fully exploiting the relationship between historical packing records and input orders(orders to be packed) , the problem is reformulated as a set cover problem. Then, two novel strategies, the constraint handling and process acceleration strategies are applied to the classic column generation approach to solve this set cover problem. The cost of solving pricing problem for generating new columns is high due to the complex constraints and customer requirements. The proposed constraints handling strategy exploits the historical packing records with the most negative value of the reduced cost. Those constraints have been implicitly satisfied in these historical packing records so that there is no need to conduct further evaluation on constraints, thus the computational load is saved. To further eliminate the iteration process of column generation algorithm and accelerate the optimization process, a Learning to Price approach called Modified Pointer Network is proposed, by which we can determine which historical packing records should be selected directly. Through experiments on realworld datasets, we show our proposed method can improve the packing success rate and decrease the computation time simultaneously.
We propose a novel physical layer security scheme that exploits an optimization method as a one-way function. The proposed scheme builds on nonsquare differential multiple-input multiple-output (MIMO), which is capable of noncoherent detection even in massive MIMO scenarios and thus resilient against risky pilot insertion and pilot contamination attacks. In contrast to conventional nonsquare differential MIMO schemes, which require space-time projection matrices designed via highly complex, discrete, and combinatorial optimization, the proposed scheme utilizes projection matrices constructed via low-complexity continuous optimization designed to maximize the coding gain of the system. Furthermore, using a secret key generated from the true randomness nature of the wireless channel as an initial value, the proposed continuous optimization-based projection matrix construction method becomes a one-way function, making the proposed scheme a physical layer secure differential MIMO system. An attack algorithm to challenge the proposed scheme is also devised, which demonstrates that the security level achieved improves as the number of transmit antennas increases, even in an environment where the eavesdropper can perfectly estimate channel coefficients and experience asymptotically large signal-to-noise ratios.
This article is a gentle discussion about the field of reinforcement learning in practice, about opportunities and challenges, touching a broad range of topics, with perspectives and without technical details. The article is based on both historical and recent research papers, surveys, tutorials, talks, blogs, books, (panel) discussions, and workshops/conferences. Various groups of readers, like researchers, engineers, students, managers, investors, officers, and people wanting to know more about the field, may find the article interesting. In this article, we first give a brief introduction to reinforcement learning (RL), and its relationship with deep learning, machine learning and AI. Then we discuss opportunities of RL, in particular, products and services, games, bandits, recommender systems, robotics, transportation, finance and economics, healthcare, education, combinatorial optimization, computer systems, and science and engineering. Then we discuss challenges, in particular, 1) foundation, 2) representation, 3) reward, 4) exploration, 5) model, simulation, planning, and benchmarks, 6) off-policy/offline learning, 7) learning to learn a.k.a. meta-learning, 8) explainability and interpretability, 9) constraints, 10) software development and deployment, 11) business perspectives, and 12) more challenges. We conclude with a discussion, attempting to answer: "Why has RL not been widely adopted in practice yet?" and "When is RL helpful?".
As the share of renewable sources is increasing the need for multiple storage units appropriately sized and located is essential to achieve better inertial response. This work focuses on the question of "how to distribute constant number of storage units in the gird under transient events such that the inertial response of the maximum frequency deviation is minimized?". To answer this question, we provide a comprehensive modeling framework for energy storage units placement and size for frequency stability under spatial effects. The distributed storage units are modeled as grid supporting inverters and the total storage capacity in the grid is bounded based on the allowed steady-state frequency deviation after disturbances. The problem of finding the optimal distributions can be considered as combinatorial problem which consists of high dimensional solutions. In this light, we develop two numeric approaches based on Brute-force search and adaptation of the Cross-entropy method for finding the best distribution and examined it on a case study of the future Israeli grid. The results on the case study provide a new insight---the storage units should be placed around the area of the disturbances, including in sites with high inertia in accordance to the network topology.