Investors make decisions about buying and selling a financial asset based on available information. The traditional approach in Deep Learning when trying to predict the behavior of an asset is to take a price history, train a model, and forecast one single price in the near future. This is called the frequentist perspective. Uncertainty Quantification is an alternative in which models manage a probability distribution for prediction. It provides investors with more information than the traditional frequentist way, so they can consider the risk of making or not making a certain decision. We systematically reviewed the existing literature on Uncertainty Quantification methods in Deep Learning to predict the behavior of financial assets, such as foreign exchange, stock market, cryptocurrencies and others. The article discusses types of model, categories of financial assets, prediction characteristics and types of uncertainty. We found that, in general terms, references focus on price accuracy as a metric, although other metrics, such as trend accuracy, might be more appropriate. Very few authors analyze both epistemic and aleatoric uncertainty, and none analyze in depth how to decouple them. The time period analyzed includes the years 2001 to 2022.

In Spain, the new Medicine degrees of the European Higher Education Area (EHEA) have incorporated Information Technologies (IT) subjects to develop horizontal competences in their curriculum. Medicine Studies are costly for Higher Education (HE) institutions and, then, every cost-effective educational innovation is welcomed. There are different general approaches for designing and developing serious games in general, but specific target disciplines as Medicine (educational and professional training), lack of tailor-made methodologies, even in eliciting requirements under the Requirements Engineering (RE) framework, as a subfield of Software Engineering. In medical higher education, the main objective of using games is to engage learners in complex problem spaces that replicate real world situations, as it happens in hospital emergencies, surgery operations and so on, without the constraints and risks of the day to day practice (patient injuries, health complications and death). Therefore, the development of serious HE games is a complex, time consuming and costly process, especially for learning medical skills. In this Work in Progress, we address the problem of lacking suitable methodologies in the design and development processes of educational serious games for HE medical environments. Our proposal is a semi-formal and simple method to elicit and formalize the specific teaching needs identified in subjects or practices of Medicine degrees, in order to design and develop serious games based on sound educational principles and theories and on RE rules.

This work presents a new approach based on support vector regression to deal with incomplete input (unseen) data and compares it to other existing techniques. The empirical analysis has been done over 18 real data sets and using five different classifiers, with the aim of foreseeing which technique can be deemed as more suitable for each classifier. Also, this study tries to devise how the relevance of the missing attribute affects the performance of each pair (handling algorithm, classifier). Experimental results demonstrate that no technique is absolutely better than the others for all classifiers. However, combining the proposed strategy with the nearest neighbor classifier appears as the best choice to face the problem of missing attribute values in the input data.

This paper analyzes a generalization of a new metric to evaluate the classification performance in imbalanced domains, combining some estimate of the overall accuracy with a plain index about how dominant the class with the highest individual accuracy is. A theoretical analysis shows the merits of this metric when compared to other well-known measures.

A realistic appearance-based representation of side-view gait sequences is here introduced. It is based on a prior method where a set of appearance-based features of a gait sample is used for gender recognition. These features are computed from parameter values of ellipses that fit body parts enclosed by regions previously defined while ignoring well-known facts of the human body structure. This work presents an improved regionalization method supported by some adaptive heuristic rules to better adjust regions to body parts. As a result, more realistic ellipses and a more meaningful feature space are obtained. Gender recognition experiments conducted on the CASIA Gait Database show better classification results when using the new features.

We study streaming data where the true labels come with a delay. The question is whether the online nearest neighbour classifier (IB2 and IB3 here) should employ the unlabelled data. Three strategies are examined: do-nothing, replace and forget. Experiments with 28 data sets show that IB2 benefits from unlabelled data, while IB3 does not.

In this paper, a band selection technique for hyperspectral image data is proposed. Supervised feature extraction techniques allow a reduction of the dimensionality to extract relevant features through a labeled training set. This implies an analysis of the existing class distributions, which usually means, in the case of hyperspectral imaging, a large number of samples, making the labeling process difficult. A possible alternative could be the use of information measures, which are the basis of the proposed method. The present approach basically behaves as an unsupervised feature selection criterion, to obtain the relevant spectral bands from a set of sample images. The relations of information content between spectral bands are analyzed, leading to the proposed technique based on the minimization of the dependent information between spectral bands, while trying to maximize the conditional entropies of the selected bands.

Bagging, AdaBoost and Arc-x4 are among the most popular methods for classifier ensembles. All these methods rely on resampling techniques to generate different training subsamples for each of the base classifiers that constitute the ensemble. In the present work, the classical implementations of these algorithms are modified in such a way that resampling is performed separately over the training instances of each class, thus obtaining the same class distribution in each subsample as that of the original training set. Moreover, we also introduce other modifications related to the size of the subsamples and also to the voting strategy. Experimental results for medical and nonmedical databases are here presented and potential benefits of the proposed methods for diagnosis are suggested.

The excessive computational resources required by the Nearest Neighbor rule are a major concern for a number of specialists and practitioners in the Pattern Recognition community. Many proposals for decreasing this computational burden, through reduction of the training sample size, have been published. This paper introduces an algorithm to reduce the training sample size while preserving the original decision boundaries as much as possible. Consequently, the algorithm tends to obtain classification accuracy close to that of the whole training sample. Several experimental results demonstrate the effectiveness of this method when compared to other reduction algorithms based on similar ideas.

For those situations in which the user wants to interact with the system by using, for example, voice commands, it would be convenient to refer to the objects by their names (e.g., "cube") instead of other types of interactions (e.g., "grasp object 1"). Thus, automatic object recognition is the first step in order to acquire a higher level of interaction between the user and the robot. Nevertheless, applying object recognition techniques when the camera images are being transmitted through the web is not an easy task. In this situation, images cannot have a very high resolution, which affects enormously the recognition process due to the inclusion of more errors while digitalizing the real image. Some experiments with the Universitat Jaume I Online Robot evaluate the performance of different neural-network implementations, comparing it to that of some distance-based object recognition algorithms. Results will show which combination of object features, and algorithms (both statistical and neural networks) is more appropriate to our purpose in terms of both effectiveness and computing time.

Within the context of online robots, a considerable amount of research has traditionally focused on the global system functionality, including the way of interaction between the user and the robot. Recent results in different robotics areas have demonstrated the potential of a number of techniques from the Pattern Recognition and Machine Learning domains, although very few work has been specifically addressed to online robots, where the object recognition is directly performed by the user. In this paper, we investigate the feasibility of using a neural network approach to object recognition in the context of online robots, and discuss the main advantages over the application of statistical learning methods. Some experiments with the UJI (Universitat Jaume I) online robot evaluate the performance of different neural network implementations, comparing it to that of some distance-based object recognition algorithms.

This paper describes a web-based telerobotic system with two basic capabilities: object recognition and incremental or continual learning. Within this context, the present paper investigates the feasibility of applying several distance-based classifiers to remote object recognition tasks. On the other hand, incremental learning enables the system to maintain a representative set of past training examples that are used together with new data to appropriately modify the currently held knowledge. This can constitute a very significant advantage because it allows a system to continually increase its knowledge and accordingly, to enhance its performance in recognition as it learns. And also, because this capability allows to detect an incorrect or anomalous employment of this particular remote system. The performance of the system is evaluated by means of an empirical analysis addressed to assess two competing goals: computing time and recognition accuracy.

The system consists of a multirobot architecture that gives access to an educational and an industrial robot, through the Internet, by using advanced multimedia and distributed programming tools like Java, Java3D and CORBA. The main effort is focused on the user interface that allows specification of a task into a 3D model and sending commands to the real robot once the operator approves them. This kind of interaction has the ability to consume very little bandwidth in order to perform a remote manipulation via the Internet. The user interface offers augmented and virtual reality possibilities to the users, as well as object recognition in order to specify commands to the robot in a natural way. This high-level task specification would not be possible without the grasping determination module, which suggests manipulation characteristics for every scene's object.

Intensive fruit and vegetable sorting is a common task in productive regions. In order to meet the market standards, produce is classified according to quality levels that depend on maturity degree, weight, size, density, skin defects, etc. Probably the most important of these tasks involve automatic visual inspection. A distributed and scalable system for sorting automation is presented, that addresses all aspects of the quality classification mentioned above. The main characteristics of the system are: it can control from 1 to 10 conveyor belts; its maximum performance is 15 fruits per second per belt; apart from weight sensors, it combines infrared, color and ultraviolet images; some fruit defect modules are available to take account of the presence of a given defect.

This paper describes an algorithm for oblique decision tree induction. The approach is based on the construction of a Voronoi diagram over the set of points representing patterns in a d-dimensional feature space. The procedure basically consists of a decomposition of the feature space into convex regions with samples from just one class.