Bridging logic and kernel machines

We propose a general framework to incorporate first-order logic (FOL) clauses, that are thought of as an abstract and partial representation of the environment, into kernel machines that learn within a semi-supervised scheme. We rely on a multi-task learning scheme where each task is associated with a unary predicate defined on the feature space, while higher level abstract representations consist of FOL clauses made of those predicates. We re-use the kernel machine mathematical apparatus to solve the problem as primal optimization of a function composed of the loss on the supervised examples, the regularization term, and a penalty term deriving from forcing real-valued constraints deriving from the predicates. Unlike for classic kernel machines, however, depending on the logic clauses, the overall function to be optimized is not convex anymore. An important contribution is to show that while tackling the optimization by classic numerical schemes is likely to be hopeless, a stage-based learning scheme, in which we start learning the supervised examples until convergence is reached, and then continue by forcing the logic clauses is a viable direction to attack the problem. Some promising experimental results are given on artificial learning tasks and on the automatic tagging of bibtex entries to emphasize the comparison with plain kernel machines.     

Inductive inference from noisy examples using the hybrid finitestate filter

Recurrent neural networks processing symbolic strings can be regarded as adaptive neural parsers. Given a set of positive and negative examples, picked up from a given language, adaptive neural parsers can effectively be trained to infer the language grammar. In this paper we use adaptive neural parsers to face the problem of inferring grammars from examples that are corrupted by a kind of noise that simply changes their membership. We propose a training algorithm, referred to as hybrid finite state filter, which is based on a parsimony principle that penalizes the development of complex rules. We report very promising experimental results showing that the proposed inductive inference scheme is indeed capable of capturing rules, while removing noise.     

Finite-tree analysis for constraint logic-based languages

Logic languages based on the theory of rational, possibly infinite, trees have much appeal in that rational trees allow for faster unification (due to the safe omission of the occurs-check) and increased expressivity (cyclic terms can provide very efficient representations of grammars and other useful objects). Unfortunately, the use of infinite rational trees has problems. For instance, many of the built-in and library predicates are ill-defined for such trees and need to be supplemented by run-time checks whose cost may be significant. Moreover, some widely used program analysis and manipulation techniques are correct only for those parts of programs working over finite trees. It is thus important to obtain, automatically, a knowledge of the program variables (the finite variables) that, at the program points of interest, will always be bound to finite terms. For these reasons, we propose here a new data-flow analysis, based on abstract interpretation, that captures such information. We present a parametric domain where a simple component for recording finite variables is coupled, in the style of the open product construction of Cortesi et al., with a generic domain (the parameter of the construction) providing sharing information. The sharing domain is abstractly specified so as to guarantee the correctness of the combined domain and the generality of the approach. This finite-tree analysis domain is further enhanced by coupling it with a domain of Boolean functions, called finite-tree dependencies, that precisely captures how the finiteness of some variables influences the finiteness of other variables. We also summarize our experimental results showing how finite-tree analysis, enhanced with finite-tree dependencies, is a practical means of obtaining precise finiteness information.     

The Graph Neural Network Model

Many underlying relationships among data in several areas of science and engineering, e.g., computer vision, molecular chemistry, molecular biology, pattern recognition, and data mining, can be represented in terms of graphs. In this paper, we propose a new neural network model, called graph neural network (GNN) model, that extends existing neural network methods for processing the data represented in graph domains. This GNN model, which can directly process most of the practically useful types of graphs, e.g., acyclic, cyclic, directed, and undirected, implements a function tau(G,n) isin IR^m that maps a graph G and one of its nodes n into an m-dimensional Euclidean space. A supervised learning algorithm is derived to estimate the parameters of the proposed GNN model. The computational cost of the proposed algorithm is also considered. Some experimental results are shown to validate the proposed learning algorithm, and to demonstrate its generalization capabilities.     

miRNA Expression Profiling in Subcutaneous Adipose Tissue of Monozygotic Twins Discordant for HIV Infection: Validation of Differentially Expressed miRNA and Bioinformatic Analysis

Combined AntiRetroviral Treatments (cARTs) used for HIV infection may result in varied metabolic complications, which in some cases, may be related to patient genetic factors, particularly microRNAs. The use of monozygotic twins, differing only for HIV infection, presents a unique and powerful model for the controlled analysis of potential alterations of miRNAs regulation consequent to cART treatment. Profiling of 2578 mature miRNA in the subcutaneous (SC) adipose tissue and plasma of monozygotic twins was investigated by the GeneChip^® miRNA 4.1 array. Real-time PCR and ddPCR experiments were performed in order to validate differentially expressed miRNAs. Target genes of deregulated miRNAs were predicted by the miRDB database (prediction score > 70) and enrichment analysis was carried out with g:Profiler. Processes in SC adipose tissue most greatly affected by miRNA up-regulation included (i) macromolecular metabolic processes, (ii) regulation of neurogenesis, and (iii) protein phosphorylation. Furthermore, KEGG analysis revealed miRNA up-regulation involvement in (i) insulin signaling pathways, (ii) neurotrophin signaling pathways, and (iii) pancreatic cancer. By contrast, miRNA up-regulation in plasma was involved in (i) melanoma, (ii) p53 signaling pathways, and (iii) focal adhesion. Our findings suggest a mechanism that may increase the predisposition of HIV⁺ patients to insulin resistance and cancer.     

Membrane bio-reactor for advanced textile wastewater treatment and reuse.

Textile wastewater contains slowly- or non-biodegradable organic substances whose removal or transformation calls for advanced tertiary treatments downstream Activated Sludge Treatment Plants (ASTP). This work is focused on the treatment of textile industry wastewater using Membrane Bio-reactor (MBR) technology. An experimental activity was carried out at the Baciacavallo Wastewater Treatment Plant (WWTP) (Prato, Italy) to verify the efficiency of a pilot-scale MBR for the treatment of municipal wastewater, in which textile industry wastewater predominates. In the Baciacavallo WWTP the biological section is followed by a coagulation-flocculation treatment and ozonation. During the 5 months experimental period, the pilot-scale MBR proved to be very effective for wastewater reclamation. On average, removal efficiency of the pilot plant (93% for COD, 96% for ammonium and 99% for total suspended solids) was higher than the WWTP ones. Color was removed as in the WWTP. Anionic surf actants removal of pilot plant and WWTP were very similar (92.5 and 93.3% respectively), while the non-ionic surfactants removal was higher in the pilot plant (99.2 vs. 97.1). In conclusion the MBR technology demonstrated to be effective for textile wastewater reclamation, leading both to an improvement of pollutants removal and to a draw-plate simplification.     

Recursive processing of cyclic graphs

Recursive neural networks are a powerful tool for processing structured data. According to the recursive learning paradigm, the input information consists of directed positional acyclic graphs (DPAGs). In fact, recursive networks are fed following the partial order defined by the links of the graph. Unfortunately, the hypothesis of processing DPAGs is sometimes too restrictive, being the nature of some real-world problems intrinsically cyclic. In this paper, a methodology is proposed, which allows us to process any cyclic directed graph. Therefore, the computational power of recursive networks is definitely established, also clarifying the underlying limitations of the model.     

Optimization of wide-band linear arrays

An optimization method is proposed for linear arrays to be used in ultrasound systems under wide-band operation. A fast algorithm, the threshold accepting, has been utilized to determine the element positions and weight coefficients of a linear array that generates a desired beam pattern. To reduce the computational burden in the optimization procedure, an efficient numerical routine for the beam pattern evaluation has been implemented. We address the optimization problem of both dense and sparse wideband arrays. In the first case, the goal is to minimize the side-lobe energy by varying the element weights; we compare the optimized beam pattern with that obtained with classical shading functions, showing that better results can be achieved with a wide-band optimization. We also consider the optimization of the layout (positions and weights) of a sparse linear array to achieve a desired beam pattern with a fixed or minimum number of array elements. The comparison of the proposed method with a narrow-band optimization algorithm is presented, showing that better performances (about -7 dB further reduction of the side-lobe level) can be achieved with a wide-band sparse array optimization. Further numerical simulations are given, showing that the proposed method yields better results than wideband sparse random arrays and periodic arrays with the same aperture width     

On the implementation of frontier-to-root tree automata inrecursive neural networks

We explore the node complexity of recursive neural network implementations of frontier-to-root tree automata (FRA). Specifically, we show that an FRAO (Mealy version) with m states, l input-output labels, and maximum rank N can be implemented by a recursive neural network with O( radical(log l+log m)lm(N)/log l+N log m) units and four computational layers, i.e., without counting the input layer. A lower bound is derived which is tight when no restrictions are placed on the number of layers. Moreover, we present a construction with three computational layers having node complexity of O((log l+log m) radicallm (N)) and O((log l+log m)lm(N)) connections. A construction with two computational layers is given that implements any given FRAO with a node complexity of O(lm(N)) and O((log l+log m)lm(N)) connections. As a corollary we also get a new upper bound for the implementation of finite-state automata into recurrent neural networks with three computational layers.