Dr. Alejandro Guerra Hernández

I did my doctorate at the Université de Paris 13, where I worked on the problem of learning in a BDI Multi-Agent System. My master degree, at the Universidad Veracruzana, was about a simulator for behavior networks. I've been working on intentionality and learning in the context of rational agents, agent based simulations, as well as agent based data mining. I also teach artificial intelligence, logic programming, functional programming, and agent oriented programming.

Dr. Alejandro Guerra Hernández ha publicado 35 entradas:

Nuevo artículo en Applied Mathematical Modelling

A. Platas-López, E. Mezura-Montes, N. Cruz-Ramírez, A. Guerra-Hernández. Discriminative learning of bayesian network parameters by differential evolution. Applied Mathematical Modelling 93(2021): 244-256. December 2020. ISSN 0307-904X | ScienceDirect

Abstract: This work proposes Differential Evolution (DE) to train parameters of Bayesian Networks (BN) for optimizing the Conditional Log-Likelihood (Discriminative Learning) instead of the log-likelihood (Generative Learning). Any given BN structure encodes assumptions about conditional independencies among the attributes and will result in error if they do not hold in the data. Such an error includes the classification dimension. The main goal of Discriminative learning is minimize this type of error. In this sense, although BN Discriminative Parameter Learning algorithms have been proposed, to the best of the authors’ knowledge, a meta-heuristic approach has not been devised yet. Thus, this is our main contribution: to come up with this kind of solution and evaluate its behavior so that its feasibility in this domain can be determined. Regarding the proposed method, the bias provided by the best solution in the population improves DE’s performance. DE variants based on JADE, such as L-SHADE and C-JADE, are especially recommended when introducing adaptation mechanisms of mutation and crossover parameters thus reducing the dependence on their calibration. L-SHADE is computationally recommended over any other DE variant. DE approach works well in essentially every standard situation, so DE approach is robust and at least as good, and often better, than the state-of-art method for Discriminative Learning called WANBIA. Our results suggest a potential benefit for Discriminative parameter learning with strong independence assumptions among attributes and that it typically produces more accurate classifiers than generative learning.

 

Categories: Blog personal

Research Lines (LGACs)

I introduced my research line on Agents, Learning and Simulation, for our new students in the Master in AI (MIA): 2020-lgacs-mia. (spanish).

Categories: Blog personal

New article in Research in Computer Science

A. Platas-López, N. Cruz-Ramírez, E. Mezura-Montes, and A. Guerra-Hernández. Discriminative Parameter Learning of Bayesian Networks using differential evolution: A preliminary analysis. Research in Computing Science, 149(3):75–82, 2020.

Abstract. This work proposes Differential Evolution (DE) to train parameters of Bayesian Networks (BN) for optimizing the Conditional Log-Likelihood (Discriminative Learning) instead of the log-likelihood (Generative Learning). Although Discriminative Parameter Learning algorithms have been proposed, to the best of the authors’ knowledge, a metaheuristic approach has not been devised yet. Thus, the objective of this research is to come up with this kind of solution and evaluate its behavior so that its feasibility in this domain can be determined. According to the theory such a solution tends to generate low-bias classifiers that minimize classification error but this is not reflected in results, regarding proposed method, bias in search for best solutions improves DEs performance.

Categories: Blog personal

New article in Research in Computer Science

D. Martínez-Galicia, A. Guerra-Hernández, N. Cruz-Ramírez, X. Limón, and F. Grimaldo. Towards Windowing as a Sub-Sampling Method for Distributed Data Mining. Research in Computing Science, 149(3):57–64, 2020.

Abstract. Windowing is a sub-sampling method that enables the induction of decision trees with large datasets. Using a small sample of the available training examples, the method can achieve levels of accuracy comparable or better than those obtained using the full available dataset. More relevant is the fact that Windowing-based strategies for Distributed Data Mining (DDM) have shown a correlation between the accuracy of the learned decision tree and the number of examples used to learn it, i.e., the higher the accuracy, the fewer examples used to induce the model. This paper corroborates that this behavior is also observed when adopting inductive algorithms of a different nature than C4.5 or ID3, the algorithms usually adopted when windowing, contributing to the use of Windowing as a general sub-sampling method for DDM. The paper also contributes exploring some metrics to the validation of the obtained sub-samples of examples.

Categories: Blog personal

New article in Mathematical and Computational Applications

D.Martínez Galicia, A. Guerra-Hernández, N. Cruz-Ramírez, X. Limón, F. Grimaldo. Windowing as a Sub-Sampling Method for Distributed Data Mining. Mathematical and Computational Applications,  25(3)1:39. June 2020. ISSN 2297-8747.

Abstract: Windowing is a sub-sampling method, originally proposed to cope with large datasets when inducing decision trees with the ID3 and C4.5 algorithms. The method exhibits a strong negative correlation between the accuracy of the learned models and the number of examples used to induce them, i.e., the higher the accuracy of the obtained model, the fewer examples used to induce it. This paper contributes to a better understanding of this behavior in order to promote windowing as a sub-sampling method for Distributed Data Mining. For this, the generalization of the behavior of windowing beyond decision trees is established, by corroborating the observed negative correlation when adopting inductive algorithms of different nature. Then, focusing on decision trees, the windows (samples) and the obtained models are analyzed in terms of Minimum Description Length (MDL), Area Under the ROC Curve (AUC), Kulllback–Leibler divergence, and the similitude metric Sim1; and compared to those obtained when using traditional methods: random, balanced, and stratified samplings. It is shown that the aggressive sampling performed by windowing, up to 3% of the original dataset, induces models that are significantly more accurate than those obtained from the traditional sampling methods, among which only the balanced sampling is comparable in terms of AUC. Although the considered informational properties did not correlate with the obtained accuracy, they provide clues about the behavior of windowing and suggest further experiments to enhance such understanding and the performance of the method, i.e., studying the evolution of the windows over time.

Categories: Blog personal

New Article in Studies in Computational Intelligence

A. Platas-López, A. Guerra-Hernández, N. Cruz-Ramírez, M. Quiroz-Castellanos, F. Grimaldo, M. Paolucci, and F. Cecconi. Intuitionistic and Type-2 Fuzzy Logic Ehancements in Neural and Op- timization Algorithms: Theory and Applications, volume 862 of Studies in Computational Intelligence, chapter Towards an Ageng-Based Model for Analysis of Macroeconomic Signals, pages 551–565. Springer, Cham, Switzerland, 2020.

 

Abstract. This work introduces an agent-based model for the analysis of macroeconomic signals. The Bottom-up Adaptive Model (BAM) deploys a closed Walrasian economy where three types of agents (households, firms and banks) interact in three markets (goods, labor and credit) producing some signals of interest, e.g., unemployment rate, GDP, inflation, wealth distribution, etc. Agents are bounded rational, i.e., their behavior is defined in terms of simple rules finitely searching for the best salary, the best price, and the lowest interest rate in the corresponding markets, under incomplete information. The markets define fixed protocols of interaction adopted by the agents. The observed signals are emergent properties of the whole system. All this contrasts with the traditional macroeconomic approach based on the general equilibrium model, where perfect rationality and/or full information availability are assumed. The model is defined following the Overview, Design concepts, and Details Protocol and implemented in NetLogo. BAM is promoted as a toolbox for studying the macroeconomic effects of the agent activities at the service of the elaboration of public policies.

Categories: Blog personal

New paper accepted at CCIA 2019

Our paper “Micro-foundations of macroeconomic dynamics: the agent-based BAM model” has been accepted for presentation at CCIA 2019, to be held in Colonia de Sant Jordi. Congratulations to the co-authors: Alejandro Platas López, Federico Cecconi, Mario Paolucci, and Francisco Grimaldo.

Abstract: This paper presents an open-source agent-based implementation of the BAM model, a micro-founded simulation of macroeconomic basic dynamics defined in the reference book Macroeconomics from the Bottom-up. By exploring the parameter space of our simulation we show that: i) BAM reproduces numerous stylized facts and its parameters influence the outputs plausibly; and ii) the effects of changing the size of markets and introducing shocks of different sizes are as expected. The outcomes are measured in terms of gross domestic product, inflation and unemployment rate, using monthly payments as time scale. These results confirm the fidelity and usability of our implementation, as well as the feasibility of the BAM model.

Categories: Blog personal

New paper at NEO 2019

Our paper “Analysis of Differential Evolution variants for parameter tuning of Decision Trees Inductive Algorithms” was accepted for its presentation at NEO 2019, to be held in Saltillo, México. The full program of the event will be available on August, 20th. Congratulations to the co-authors David Martínez Galicia y Efrén Mezura Montes.

Extended abstract: This paper presents an empirical comparison of some Differential Evolution variants to solve the parameter tuning of Decision Tree induction algorithms. The aim of this analysis is to identify which one of the variants is more competitive to this problem. In this work, an EVolutionary Agents \& Artifacts approach for the induction of Decision Trees (eva2dt), is adopted to maximize the classification accuracy.

The Agents & Artifacts paradigm naturally accommodates the concept of an evolving population of agents inducing and evaluating decision trees, using different tools implemented as Weka-based artifacts. The use of different inductive algorithms, instead of a single one, is a novelty with respect to related work. Eva2dt agents represent potential solutions for the Differential Evolution algorithm, which are mapped to parameters for inductive algorithms, e.g., j48, SimpleCART, RepTree and RandomTree. Each agent is able to induce decisions trees and to also compute their accuracy through a 10-folds cross-validation process.

The assessed Differential Evolution variants vary in the recombination operator adopted and in some self-adaptive mechanisms. A set of statistical tests were performed to validate the obtained results. All variants were tested on 16 public datasets, obtaining competitive results, particularly for the datasets with the fewer number of training instances. Preliminary results suggest that DE/rand/1/bin variant outperform the others, even when more generations are added. These observations will help to develop other mechanisms to improve the current performance (convergence and accuracy).

Categories: Blog personal

New article in CONISOFT 2018

X. Limón Riaño, A. Guerra-Hernández, A. J. Sánchez-García, J. C. Pérez Arriaga. SagaMAS: a software framework for distributed transactions in the microservice architecture. In: CONISOFT 2018 International Conference in Software Engineering Research and Innovation, San Luis Potosí, México, October 24th-26th, 2018. IEEE Computer Society, Los Alamitos, CA, USA, 2018.

Abstract. This paper introduces SagaMAS: a Multi-Agent based framework on development, dealing with distributed transactions in the microservices architecture. Microservices are an architectural style where the distributed system is decomposed in a series of highly cohesive and independent services. Each microservice can have different implementations and data persistence technologies, resulting in heterogeneous distributed systems. Given its heterogeneity and distributed nature, an open challenge in this architecture is the proper management of distributed transactions that expand through several microservices. MultiAgent Systems are by definition distributed systems suited for complex coordination tasks, such as this. The proposed framework can be seen as a decoupled autonomous layer that coordinates the distributed transactions of the system, relieving the microservice developer from such tasks, and simplifying microservice interactions. Unlike existing approaches, our proposal is conceived and usable at an abstraction level appropriate to express reliability and robustness issues in terms of agent coordination.

Keywords. Microservices, Software Architectures, Distributed Transactions, Multi-Agent Systems

Categories: Blog personal

Nuevo artículo KAIS

X. Limón, A. Guerra-Hernández, N. Cruz-Ramírez, F. Grimaldo. Modeling and implementing distributed data mining strategies in JaCa-DDM. Knowledge and Information Systems  (2018). ISSN: 0239-3116 KAIS

Abstract. This work introduces JaCa-DDM, a novel distributed data mining system founded on the agents and artifacts paradigm, conceived to design, implement, deploy, and evaluate learning strategies. Jason rational agents conform to such strategies to cope with distributed computing environments, where CArtAgO artifacts encapsulate learning algorithms, data sources, evaluation tools, and other services implemented in Weka for data mining tasks. The set of strategies presented in this paper aims at encouraging the use of JaCa-DDM to develop new ones, suited to different needs. For this, our system provides tools to evaluate the resulting models in terms of accuracy, number of instances employed to learn, time of convergence, and volume of communications. Although the emphasis in decision trees, JaCa-DDM can be easily extended by adopting new artifacts, e.g., for meta-learning. The main contributions of the paper are as follows: (i) From the multi-agent systems perspective, our approach illustrates how to exploit the so-called “agentification” of Weka for the sake of code reusability, while preserving the benefits of reasoning at the Belief–Desire–Intention level with Jason; (ii) from the data mining perspective, JaCa-DDM is promoted as an extensible tool to define and test distributed strategies; and (iii) a set of strategies including centralizing, meta-learning and Windowing-based approaches, is carefully analyzed to provide comparisons among them.

Categories: Blog personal

contacto

Dr. Alejandro Guerra Hernández