Arthur Carvalho

A cooperative coevolutionary genetic algorithm for learning bayesian network structures

We propose a cooperative coevolutionary genetic algorithm for learning Bayesian network structures from fully observable data sets. Since this problem can be decomposed into two dependent subproblems, that is to find an ordering of the nodes and an optimal connectivity matrix, our algorithm uses two subpopulations, each one representing a subtask. We describe the empirical results obtained with simulations of the Alarm and Insurance networks. We show that our algorithm outperforms the deterministic algorithm K2.

Improving NSGA-II with an adaptive mutation operator

The performance of a Multiobjective Evolutionary Algorithm (MOEA) is crucially dependent on the parameter setting of the operators. The most desired control of such parameters presents the characteristic of adaptiveness, i.e., the capacity of changing the value of the parameter, in distinct stages of the evolutionary process, using feedbacks from the search for determining the direction and/or magnitude of changing. Given the great popularity of the algorithm NSGA-II, the objective of this research is to create adaptive controls for each parameter existing in this MOEA. With these controls, we expect to improve even more the performance of the algorithm. In this work, we propose an adaptive mutation operator that has an adaptive control which uses information about the diversity of candidate solutions for controlling the magnitude of the mutation. A number of experiments considering different problems suggest that this mutation operator improves the ability of the NSGA-II for reaching the Pareto optimal Front and for getting a better diversity among the final solutions.

How many crowdsourced workers should a requester hire

Recent years have seen an increased interest in crowdsourcing as a way of obtaining information from a potentially large group of workers at a reduced cost. The crowdsourcing process, as we consider in this paper, is as follows: a requester hires a number of workers to work on a set of similar tasks. After completing the tasks, each worker reports back outputs. The requester then aggregates the reported outputs to obtain aggregate outputs. A crucial question that arises during this process is: how many crowd workers should a requester hire? In this paper, we investigate from an empirical perspective the optimal number of workers a requester should hire when crowdsourcing tasks, with a particular focus on the crowdsourcing platform Amazon Mechanical Turk. Specifically, we report the results of three studies involving different tasks and payment schemes. We find that both the expected error in the aggregate outputs as well as the risk of a poor combination of workers decrease as the number of workers increases. Surprisingly, we find that the optimal number of workers a requester should hire for each task is around 10 to 11, no matter the underlying task and payment scheme. To derive such a result, we employ a principled analysis based on bootstrapping and segmented linear regression. Besides the above result, we also find that overall top-performing workers are more consistent across multiple tasks than other workers. Our results thus contribute to a better understanding of, and provide new insights into, how to design more effective crowdsourcing processes.

Reinforcement learning for the soccer dribbling task

We propose a reinforcement learning solution to the soccer dribbling task, a scenario in which a soccer agent has to go from the beginning to the end of a region keeping possession of the ball, as an adversary attempts to gain possession. While the adversary uses a stationary policy, the dribbler learns the best action to take at each decision point. After defining meaningful variables to represent the state space, and high-level macro-actions to incorporate domain knowledge, we describe our application of the reinforcement learning algorithm Sarsa with CMAC for function approximation. Our experiments show that, after the training period, the dribbler is able to accomplish its task against a strong adversary around 58% of the time.

A data mining approach to solve the goal scoring problem

In soccer, scoring goals is a fundamental objective which depends on many conditions and constraints. Considering the RoboCup soccer 2D-simulator, this paper presents a data mining-based decision system to identify the best time and direction to kick the ball towards the goal to maximize the overall chances of scoring during a simulated soccer match. Following the CRISP-DM methodology, data for modeling were extracted from matches of major international tournaments (10691 kicks), knowledge about soccer was embedded via transformation of variables and a Multilayer Perceptron was used to estimate the scoring chance. Experimental performance assessment to compare this approach against previous LDA-based approach was conducted from 100 matches. Several statistical metrics were used to analyze the performance of the system and the results showed an increase of 7.7% in the number of kicks, producing an overall increase of 78% in the number of goals scored.

A Study on the Influence of the Number of MTurkers on the Quality of the Aggregate Output

Recent years have seen an increased interest in crowdsourcing as a way of obtaining information from a large group of workers at a reduced cost. In general, there are arguments for and against using multiple workers to perform a task. On the positive side, multiple workers bring different perspectives to the process, which may result in a more accurate aggregate output since biases of individual judgments might offset each other. On the other hand, a larger population of workers is more likely to have a higher concentration of poor workers, which might bring down the quality of the aggregate output.

Electricity Trading Agent for EV-enabled Parking Lots

The reduction of greenhouse gas emissions is seen as an important step towards environmental sustainability. Perhaps not surprising, many governments all around the world are providing incentives for consumers to buy electric vehicles (EVs). A positive response from consumers means that the demand for the charging infrastructure increases as well. We investigate how an existing traditional parking lot, upgraded with chargers, can suit the present demand for charging stations. In particular, a resulting EV-enabled parking lot is an electricity trading agent (i.e., broker) which acts as an energy retailer and as a player on a target electricity market. In this paper, we use agent-based simulation to present the EV-enabled parking lot ecosystem in order to model the underlying dynamics and uncertainties regarding parking lots with electricity trading agent functionalities. We instantiate our agent-based simulations using real-life data in order to perform the what-if analysis. Several key performance indicators (KPIs), including parking utilization, charging utilization and electricity utilization, are proposed. We also illustrate how those KPIs can be used to choose the effective investment strategy with respect to the number and speed of chargers.

Evaluating Policies for Parking Lots Handling Electric Vehicles

The recent advent of electric vehicles (EVs) marks the beginning of a new positive era in the transportation sector. Although the environmental benefits of EVs are well-known today, planning and managing EV charging infrastructure are activities that are still not well-understood. In this paper, we are investigating how the so-called EV-enabled parking lot, a parking lot that is equipped with a certain number of chargers, can define an appropriate parking policy in such a way that satisfies two challenges: EV owners’ needs for recharging as well as the parking lot operator’s goal of profit maximization. Concretely, we present three parking policies that are able to simultaneously deal with both EVs and internal combustion engine vehicles. Detailed sensitivity analysis, based on real-world data and simulations, evaluates the proposed parking policies in a case study concerning parking lots in Melbourne, Australia. This paper produces results that are highly prescriptive in nature because they inform a decision maker under which circumstances a certain parking policy operates optimally. Most notably, we find that the dynamic parking policy, which takes the advantage of advanced information technology (IT) and charging infrastructure by dynamically changing the role of parking spots with chargers, often outperforms the other two parking policies, because it maximizes the profit and minimizes the chance of cars being rejected by the parking lot. We also discuss how making a few parking spots EV-exclusive might be a good policy when the number of available chargers is small and/or the required IT infrastructure is not in place for using the dynamic policy. We conclude this paper proposing a technology roadmap for transforming parking lots into smart EV-enabled parking lots based on the three studied parking policies.

Inducing honest reporting of private information in the presence of social projection.

We discuss payment structures that induce honest reporting of private information by risk-neutral agents in settings involving multiple-choice questions. Such payment structures do not rely on the existence of ground-truth answers, but instead they rely on the assumption that agents exhibit social projection. Social projection is a strong form of the well-known psychological phenomenon called the false-consensus effect, where an agent believes that his private answer to a multiple-choice question is the most popular answer. From a theoretical perspective, we first show that when social projection holds true, honest reporting strictly maximizes an agent’s expected reward from a payment structure that simply compares agents’ reported answers and rewards agreements. Furthermore, we suggest how to induce honest reporting by taking the distance between reported answers into account when social projection is strong, i. e., when an agent believes that his private answer is more likely to be reported by a random peer than all the other answers combined. We also discuss how to derive the above results in terms of proper scoring rules. From an empirical perspective, we investigate the consequences of using a payment structure that rewards agreements in a content-analysis experiment on Amazon Mechanical Turk. We obtain some evidence that, under such a payment structure, agents report more accurate answers than when there are no direct incentives for honest reporting of private answers. Moreover, we find that priming agents by briefly mentioning the theoretical properties of the underlying payment structure results in even more accurate answers.