Eduardo Morales - Academia.edu (original) (raw)

Papers by Eduardo Morales

Associative classification is a pattern recognition approach that integrates classification and a... more Associative classification is a pattern recognition approach that integrates classification and association rule discovery to build accurate classification models. These models are formed by a collection of contrast patterns that fulfill some restrictions. In this paper, we introduce an experimental comparison of the impact of using different restrictions in the classification accuracy. To the best of our knowledge, this is the first time that such analysis is performed, deriving some interesting findings about how restrictions impact on the classification results. Contrasting these results with previously published papers, we found that their conclusions could be unintentionally biased by the restrictions they used. We found, for example, that the jumping restriction could severely damage the pattern quality in the presence of dataset noise. We also found that the minimal support restriction has a different effect in the accuracy of two associative classifiers, therefore deciding w...

Neural Computing and Applications, 2021

Deep reinforcement learning (DRL) combines the benefits of deep learning and reinforcement learni... more Deep reinforcement learning (DRL) combines the benefits of deep learning and reinforcement learning. However, it still requires long training times and a large number of instances to reach an acceptable performance. Transfer learning (TL) offers an alternative to reduce the training time of DRL agents, using less instances and in some cases improving performance. In this work, we propose a transfer learning formulation for DRL across tasks. A relevant problem of TL that we address herein is how to select a proper pre-trained model that will be useful for the target task. We consider the entropy of feature maps in the hidden layers of the convolutional neural network and their actions spaces as relevant features to select a pre-trained model that is then fine-tuned for the target task. We report experimental results of the proposed source task selection methodology when using Deep Q-Networks for learning to play Atari games. Nevertheless, the proposed method could be used in other DRL algorithms (e.g., DDQN, C51, etc.) and also other domains. Results reveal that most of the time our proposed method is capable of selecting source tasks that improve the performance of a model trained from scratch. Additionally, we introduce a method for selecting the most relevant kernels for the target task, the results show that transferring a subset of the convolutional kernels results in similar performance to training the model from scratch while using less parameters.

IEEE Transactions on Mobile Computing, 2020

Stress assessment is a complex issue and numerous studies have examined factors that influence st... more Stress assessment is a complex issue and numerous studies have examined factors that influence stress in working environments. Research studies have shown that monitoring individuals' behaviour parameters during daily life can also help assess stress levels. In this study, we examine assessment of work-related stress using features derived from sensors in smartphones. In particular, we use information from physical activity levels, location, social-interactions, social-activity and application usage during working days. Our study included 30 employees chosen from two different private companies, monitored over a period of 8 weeks in real work environments. The findings suggest that information from phone sensors shows important correlation with employees perceived stress level. Secondly, we used machine learning methods to classify perceived stress levels based on the analysis of information provided by smartphones. We used decision trees obtaining 67.57% accuracy and 71.73% after applying a semi-supervised method. Our results show that stress levels can be monitored in unobtrusive manner, through analysis of smartphone data.

Adaptive Behavior, 2019

Recent advances in deep learning, in particular in convolutional neural networks (CNNs), have bee... more Recent advances in deep learning, in particular in convolutional neural networks (CNNs), have been widely used in robotics for object classification and action recognition, among others, with very high performance. Nevertheless, this high performance, mostly in classification tasks, is rarely accompanied by reasoning processes that consider the relationships between objects, actions, and effects. In this article, we used three CNNs to classify objects, actions, and effects that were trained with the CERTH-SOR3D dataset that has more than 20,000 RGB-D videos. This dataset involves 14 objects, 13 actions, and in this article was augmented with seven effects. The probabilistic vector output of each trained CNN was combined into a Bayesian network (BN) to capture the relationships between objects, actions, and effects. It is shown that by probabilistically combining information from the three classifiers, it is possible to improve the classification performance of each CNN or to level t...

Lecture Notes in Computer Science, 2002

Abstract. To learn a map of an environment a mobile robot has to ex-plore its workspace. This pap... more Abstract. To learn a map of an environment a mobile robot has to ex-plore its workspace. This paper introduces a new exploration approach that minimizes movements of the robot to reach the nearest unexplored region of the environment. In contrast to other methods, this ...

Lecture Notes in Computer Science

ABSTRACT

Abstract Reinforcement learning deals with learningoptimal or near optimal policies while inter-a... more Abstract Reinforcement learning deals with learningoptimal or near optimal policies while inter-acting with an external environment. Theapplicability of reinforcement learning hasbeen limited by largesearchspacesand by itsinability to re-use previously learned policyon other, although similar, problems. A re-lational representation can be used to allevi-ate both problems. In particular, this papershows how a relational representation can beused to produce powerful abstractions whichcan signiﬁcantly reduced the search space,and that learning over this abstracted spacecan help to re-use previously learned poli-cies on new problems. The proposed stateabstraction requires the deﬁnition of rela-tional properties of the state space and re-lational actions over these states. This papershows how these relatio nal actions can be in-duced using a behavioural cloning approachand how the relations can be learned usingan ILP system based on a state representa-tion. Finally, it is shown when it is neces-sary to change the current state abstractionby learning a new relation. Examples fromgrid based problems, chess and a ﬂight sim-ulator are used to illustrate the main ideasbehind this research.

Res. Comput. Sci., 2020

Reinforcement learning and Causal Inference are indispensable part of machine learning. However, ... more Reinforcement learning and Causal Inference are indispensable part of machine learning. However, they are usually treated separately, although that both are directly relevant to problem solving methods. One of the challenges that emerge in Reinforcement Learning, is the trade-off between exploration and exploitation. In this work we propose to use causal models to attend the learning process of an agent. The causal models helps to restrict the search space by reducing the actions that an agent can take through interventional queries like: Would I have achieved my goal if I had drop the passenger off here?. This simulates common sense that lightens the time it takes the trial and error approach. We attack the classic taxi problem and we show that using causal models in the Q-learning action selection step leads to higher and faster jump-start reward and convergence, respectively.

Lecture Notes in Computer Science, 2021

Deep reinforcement learning has obtained impressive performance in challenging tasks in recent ye... more Deep reinforcement learning has obtained impressive performance in challenging tasks in recent years. Nevertheless, it has important limitations such as long training times and the number instances that are needed to achieve acceptable performance. Transfer learning offers an alternative to alleviate these limitations. In this paper, we propose a novel method for transferring knowledge from more than one source tasks. First, we select the best source tasks using a regressor that predicts the performance of a pre-trained model in the target task. Then, we apply a selection of relevant convolutional kernels for the target task in order to find a target model with similar number of parameters compared to the source ones. According to the results, our approach outperforms the accumulated reward obtained when learning from scratch in 20.62% using lower parameters (about 56% of the total, depending on the specific game).

2017 IEEE International Autumn Meeting on Power, Electronics and Computing (ROPEC), 2017

This paper describes the design and analysis of results of the 2017 RedICA: Text-Image Matching (... more This paper describes the design and analysis of results of the 2017 RedICA: Text-Image Matching (RICATIM) challenge. This academic competition faces the image labeling problem (assigning words to images) as one binary classification. Motivated by recent success of representation learning, we built a data set for binary classification in which each instance is the learned representation of a pair of an image and a word. Instances are labeled as positive, if the word is relevant for describing the content of the image and negative otherwise. Thus, participants of the challenge had to develop binary classification methods to distinguish between relevant and irrelevant text-image matchings. The challenge attracted 43 participants, that provided quite original and competitive solutions. The performance obtained by the top ranked participants was impressive, improving the performance of the baseline considerably. In this paper we describe the approached problem, the challenge design (incl...

Machine Learning and Knowledge Discovery in Databases, 2020

Recent progress in AutoML has lead to state-of-the-art methods (e.g., AutoSKLearn) that can be re... more Recent progress in AutoML has lead to state-of-the-art methods (e.g., AutoSKLearn) that can be readily used by non-experts to approach any supervised learning problem. Whereas these methods are quite effective, they are still limited in the sense that they work for tabular (matrix formatted) data only. This paper describes one step forward in trying to automate the design of supervised learning methods in the context of text mining. We introduce a meta learning methodology for automatically obtaining a representation for text mining tasks starting from raw text. We report experiments considering 60 different textual representations and more than 80 text mining datasets associated to a wide variety of tasks. Experimental results show the proposed methodology is a promising solution to obtain highly effective off the shell text classification pipelines.

Research in Computing Science, 2017

Computación y Sistemas, 2019

This paper describes the design of the 2017 RedICA: Text-Image Matching (RICATIM) challenge, incl... more This paper describes the design of the 2017 RedICA: Text-Image Matching (RICATIM) challenge, including the dataset generation, a complete analysis of results, and the descriptions of the top-ranked developed methods. The academic challenge explores the feasibility of a novel binary image classification scenario, where each instance corresponds to the concatenation of learned representations of an image and a word. Instances are labeled as positive if the word is relevant for describing the visual content of the image, and negative otherwise. This novel approach of the image classification problem poses an alternative scenario where any text-image pair can be represented in such space, so any word could be considered for describing an image. The proposed methods are diverse and competitive, showing considerable improvements over the proposed baselines.

2007 Sixth Mexican International Conference on Artificial Intelligence, Special Session (MICAI), 2007

MDQL is an algorithm, based on reinforcement learning, for solving multiple objective optimizatio... more MDQL is an algorithm, based on reinforcement learning, for solving multiple objective optimization problems, that has been tested on several applications with promising results [6, 7]. MDQL discretizes the decision variables into a set of states, each associated with actions to move agents to contiguous states. A group of agents explore this state space and are able to find Pareto sets applying a distributed reinforcement learning algorithm. The precision of the Pareto solutions depends on the chosen granularity of the states. A finer granularity on the states creates more precise solutions but at the expense of a larger search space, and consequently the need for more computational resources. In this paper, a very important improvement is introduced into the original MDQL algorithm to incrementally refined the Pareto solutions. The new algorithm, called IMDQL, starts with a coarse granularity to find an initial Pareto set. A vicinity for each of the Pareto solutions in refined and a new Pareto set is founded in this refined state space. This process continues until there is no more improvement within a small threshold value. It is shown that IMDQL not only improves the solutions found by MDQL, but also converges faster. MDQL has also been tested on the solutions of dynamic optimization problems. In this paper, it is also shown that the adaptation capabilities observed in MDQL can be improved with IMDQL. IMDQL was tested on the benchmark problems proposed by Jin [4]. Performance evaluation was made using the Collective Mean Fitness metric proposed by Morrison [10]. IMDQL was compared against an standard evolution strategy with the covariance matrix adaptation (CMA-ES) with very promising results.

In this paper we present Sabina, a service robot developed by the Markovito team at INAOE. Sabina... more In this paper we present Sabina, a service robot developed by the Markovito team at INAOE. Sabina is based on a PatrolBot robot platform and incorporates a set of general purpose modules for service robots that achieve basic robot skills, such as map building; localiza- tion and navigation; object and people recognition and tracking; and human interaction using facial animation, speech and gestures and ma- nipulation. All these modules are integrated in a layered behavior-based architecture implemented on the Robot Operating System (ROS). In ad- dition to these Sabina's capabilities, we also describe a novel approach to autonomous 3D object reconstruction. The Markovito team has par- ticipated in the Robocup@Home category in previous Robocup compe- titions; in Turkey 2011 our team qualied for the second stage of the competition. Last year, in 2015, the Markovito team won a 1st place in the mexican RoboCup@home competition.

Object recognition is a relevant task for many areas and, in particular, for service robots. Rece... more Object recognition is a relevant task for many areas and, in particular, for service robots. Recently object recognition has been dominated by the use of Deep Neural Networks (DNN), however, they required a large number of images and long training times. If a user asks a service robot to search for an unknown object, it has to deal with selecting relevant images to learn a model, deal with polysemy, and learn a model relatively quickly to be of any use to the user. In this paper we describe an object recognition system that deals with the above challenges by: (i) a user interface to reduce different object interpretations, (ii) downloading on-the-fly images from Internet to train a model, and (iii) using the outputs of a trimmed pre-trained DNN as attributes for a SVM. The whole process (selecting and downloading images and training a model) of learning a model for an unknown object takes around two minutes. The proposed method was tested on 72 common objects found in a house enviro...

Lecture Notes in Computer Science, 2018

The analysis of natural images has been the topic of research in uncountable articles in computer... more The analysis of natural images has been the topic of research in uncountable articles in computer vision and pattern recognition (e.g., natural images has been used as benchmarks for object recognition and image retrieval). However, despite the research progress in such field, there is a gap in the analysis of certain type of natural images, for instance, those in the context of animal behavior. In fact, biologists perform the analysis of natural images manually without the aid of techniques that were supposedly developed for this purpose. In this context, this paper presents a study on automated methods for the analysis of natural images of hummingbirds with the goal to assist biologists in the study of animal behavior. The automated analysis of hummingbird behavior is challenging mainly because of (1) the speed at which these birds move and interact; (2) the unpredictability of their trajectories; and (3) its camouflage skills. We report a comparative study of two deep learning approaches for the detection of hummingbirds in their nest. Two variants of transfer learning from convolutional neural networks (CNNs) are evaluated in real imagery for hummingbird behavior analysis. Transfer learning is adopted because not enough images are available for training a CNN from scratch, besides, transfer learning is less time consuming. Experimental results are encouraging, as acceptable classification performance is achieved with CNN-based features. Interestingly, a pretrained CNN without fine tunning and a standard classifier performed better in the considered data set.