Search

The State of the Art of the young field of Automated Machine Learning (AutoML) is held by the connectionist approach. Several techniques of such an inspiration have recently shown promising results in automatically designing neural network architectures. However, apart from back-propagation, only a few applications of other learning techniques are used for these purposes. The back-propagation process takes advantage of specific optimization techniques that are best suited to specific application domains (e.g., Computer Vision and Natural Language Processing). Hence, the need for a more general learning approach, namely, a basic algorithm able to make inference in different contexts with distinct properties. In this paper, we deal with the problem from a scientific and epistemological point of view. We believe that this is needed to fully understand the mechanisms and dynamics underlying human learning. To this aim, we define some elementary inference operations and show how modern architectures can be built by a combination of those elementary methods. We analyze each method in different settings and find the best-suited application context for each learning algorithm. Furthermore, we discuss experimental findings and compare them with human learning. The discrepancy is particularly evident between supervised and unsupervised learning. Then, we determine which elementary learning rules are best suited for unsupervised systems, and, finally, we propose some improvements in reinforcement learning architectures.

Recommender systems (RSs) represent one of the manifold applications in which Machine Learning can unfold its potential. Nowadays, most of the major online sites selling products and services provide users with RSs that can assist them in their online experience. In recent years, therefore, we have witnessed an impressive series of proposals for novel recommendation techniques that claim to ensure significative improvements compared to classic techniques. In this work, we analyze some of them from a theoretical and experimental point of view and verify whether they can deliver tangible real improvements in terms of performance. Among others, we have experimented with traditional model-based and memory-based collaborative filtering, up to the most recent recommendation techniques based on deep learning. We have chosen the movie domain as an application scenario, and a version of the classic MovieLens as a dataset for training and testing our models.

Positron emission tomography (PET) with 2-[18F]-fluorodeoxyglucose (FDG) has been widely used for the evaluation of cortical glucose metabolism in several neurodegenerative disorders while its potential role in the evaluation of cortical and subcortical activity during a task in the healthy and pathological brain still remains to be a matter of debate. Few studies have been carried out in order to investigate the potential role of this radiotracer for the evaluation of brain glucose consumption during dynamic brain activation. The aim of this review is to provide a general overview of the applications of FDG-PET in the evaluation of cortical activation at rest and during tasks, describing first the physiological basis of FDG distribution in brain and its kinetic in vivo. An overview of the imaging protocols and image interpretation will be provided as well. As a last aspect, the results of the main studies in this field will be summarized and the results of PET findings performed in healthy subjects and patients suffering from various diseases will be reported.

The past years have witnessed an increased use of applied games for developing and evaluating communication skills. These skills benefit from in-terpersonal interactions. Providing feedback to students practicing communica-tion skills is difficult in a traditional class setting with one teacher and many students. This logistic challenge may be partly overcome by providing training using a simulation in which a student practices with communication scenarios. A scenario is a description of a series of interactions, where at each step the player is faced with a choice. We have developed a scenario editor that enables teachers to develop scenarios for practicing communication skills. A teacher can develop a scenario without knowledge of the implementation. This paper presents the implementation architecture for such a scenario-based simulation.

Hypermedia, with its combination of multimedia and non-linear organization of links among informative nodes, provides a highly interactive environment. In structured domains such as Web-based Educational Systems, the complexity of the learning domain often requires a large set of learning nodes and conceptual interrelationships that can cause several issues, e.g.: lack of comprehension, disorientation and inefficacious learning strategies. In this article we propose a new approach to guided navigation in hypermedia-based domains, suitable for helping users in structured and complex learning environments such as cultural heritage domains. Our proposal draws inspiration from the Case-Based Reasoning paradigm associated with a hypermedia structural analysis. In particular, our presentation highlights the use of a hybrid architecture for Adaptive Navigation Support, where the indexing problem of the case-based reasoner is solved by way of a sub-symbolic approach. A case study in the Neo-Realist Italian Cinema domain is discussed along with a formal and controlled evaluation that proves the advantages of the proposed approach.

In this paper, we propose a framework for personalized access to location-aware services. The system relies on the integration of a Geographic Information Retrieval module with a User Modeling component. We describe an application into a specific field, but the platform is easily usable in different contexts. Currently, the framework has been extending to the mobile domain.

There is an increasing reliance on the Internet for information seeking in many everyday activities. While spending on Internet connectivity has increased, besides traditional IR-based search engines, there is still a lack of tools for helping users searching information in this large, self-organizing and dynamic environment. Very few attempts have been undertaken to better understand the user behaviour, preferences and needs during Web information seeking tasks. Cognitive models, as approximations of cognitive processes able to predict and comprehend basic human tasks, have the chance to provide a deep analysis of the user behaviour during Web search activities. In this paper, we summarize the models recently proposed to address this issue, and we present a novel approach based on the SAM theory bent on the recognition of the user needs during Web browsing activities. The simulation of the human memory processes in terms of learning and recall by means of short and long-term semantic memory structures provides relevant information useful to adapt the traditional HCI in several important domains, such as filtering, retrieval and recommending systems.

Several statistical approaches for user profiling have been proposed in order to recognize users’ information needs during their interaction with information sources. Human memory processes in terms of learning and retrieval are with no doubt some of the fundamental elements that take part during this interaction, but actually only a few models for user profiling have been devised considering explicitly these processes. For this reason, a new approach for user modeling is proposed and evaluated. The grounds in the well-studied Search of Associative Memory (SAM) model provides a clear definition of the structure to store information and the processes of learning and retrieval. These assets are missing in other works based for example on simplified versions of semantic memory models and co-occurrence data.

Granulocytic sarcoma (GS) is a rare extramedullary solid tumor defined as an accumulation of myeloblasts or immature myeloid cells. It can cooccur with or precede the acute myeloid leukemia (AML) as well as following treated AML. The incidence of GS in AML patients is 3–8% but it significantly rises in M2 FAB subtype AML. This variety of AML harbors t(8;21) in up to 20–25% of cases (especially in children and black ones of African origin) and, at a molecular level, it is characterized by the generation of a fusion gene known as RUNX1-RUNX1T1. Approximately 10% of M2 AML patients will develop GS, as a consequence, the t(8;21) and the relative transcript represent the most common cytogenetic and molecular abnormalities in GS. FLT3-ITD mutation was rarely described in AML patients presenting with GS. FLT3 ITD is generally strongly associated with poor prognosis in AML, and is rarely reported in patients with t(8;21). GS presentation is extremely variable depending on organs involved; in general, cranial bones and sinus are very rarely affected sites. We report a rare case of GS occurring as a recurrence of a previously treated t(8;21), FLT3-ITD positive AML, involving mastoid bones and paravertebral tissues.

A system prototype for intelligent agents is presented. The system, called Holmes, is specialized in adaptive information filtering and searching of interesting information on the Web. It allows for the execution of searches by exploiting both indications provided by traditional search engines and autonomous exploration. In both cases, we use the model of the user, in the first case to make the proposed results more significant by modifying the proposed order, and in the second one to guide the exploration.