Restricted Boltzmann Machines as Models of Interacting Variables

We study the type of distributions that restricted Boltzmann machines (RBMs) with different activation functions can express by investigating the effect of the activation function of the hidden nodes on the marginal distribution they impose on observed bi nary nodes. We report an exact expression for these marginals in the form of a model of interacting binary variables with the explicit form of the interactions depending on the hidden node activation function. We study the properties of these interactions in detail and evaluate how the accuracy with which the RBM approximates distributions over binary variables depends on the hidden node activation function and the number of hidden nodes. When the inferred RBM parameters are weak, an intuitive pattern is found for the expression of the interaction terms, which reduces substantially the differences across activation functions. We show that the weak parameter approximation is a good approximation for different RBMs trained on the MNIST data set. Interestingly, in these cases, the mapping reveals that the inferred models are essentially low order interaction models.

Systematic Evaluation of Kinetics and Distribution of Muscle and Lymph Node Activation Measured by 18F-FDG- and 11C-PBR28-PET/CT Imaging, and Whole Blood and Muscle Transcriptomics After Immunization of Healthy Humans With Adjuvanted and Unadjuvanted Vaccines

Systems vaccinology has been applied to detect signatures of human vaccine induced immunity but its ability, together with high definition in vivo clinical imaging is not established to predict vaccine reactogenicity. Within two European Commission funded high impact programs, BIOVACSAFE and ADITEC, we applied high resolution positron emission tomography/computed tomography (PET/CT) scanning using tissue-specific and non-specific radioligands together with transcriptomic analysis of muscle biopsies in a clinical model systematically and prospectively comparing vaccine-induced immune/inflammatory responses. 109 male participants received a single immunization with licensed preparations of either AS04-adjuvanted hepatitis B virus vaccine (AHBVV); MF59C-adjuvanted (ATIV) or unadjuvanted seasonal trivalent influenza vaccine (STIV); or alum-OMV-meningococcal B protein vaccine (4CMenB), followed by a PET/CT scan (n = 54) or an injection site muscle biopsy (n = 45). Characteristic kinetics was observed with a localized intramuscular focus associated with increased tissue glycolysis at the site of immunization detected by 18F-fluorodeoxyglucose (FDG) PET/CT, peaking after 1–3 days and strongest and most prolonged after 4CMenB, which correlated with clinical experience. Draining lymph node activation peaked between days 3–5 and was most prominent after ATIV. Well defined uptake of the immune cell-binding radioligand 11C-PBR28 was observed in muscle lesions and draining lymph nodes. Kinetics of muscle gene expression module upregulation reflected those seen previously in preclinical models with a very early (~6hrs) upregulation of monocyte-, TLR- and cytokine/chemokine-associated modules after AHBVV, in contrast to a response on day 3 after ATIV, which was bracketed by whole blood responses on day 1 as antigen presenting, inflammatory and innate immune cells trafficked to the site of immunization, and on day 5 associated with activated CD4+ T cells. These observations confirm the use of PET/CT, including potentially tissue-, cell-, or cytokine/chemokine-specific radioligands, is a safe and ethical quantitative technique to compare candidate vaccine formulations and could be safely combined with biopsy to guide efficient collection of samples for integrated whole blood and tissue systems vaccinology in small-scale but intensive human clinical models of immunization and to accelerate clinical development and optimisation of vaccine candidates, adjuvants, and formulations.

Sensor Node Activation Using Bat Algorithm for Connected Target Coverage in WSNs

This paper proposes a sensor node activation method using the nature-inspired algorithm (NIA) for the target coverage problem. The NIAs have been used to solve various optimization problems. This paper formulates the sensor target coverage problem into an object function and solves it with an NIA, specifically, the bat algorithm (BA). Although this is not the first attempt to use the BA for the coverage problem, the proposed method introduces a new concept called bat couple which consists of two bats. One bat finds sensor nodes that need to be activated for sensing, and the other finds nodes for data forwarding from active sensor nodes to a sink. Thanks to the bat couple, the proposed method can ensure connectivity from active sensor nodes to a sink through at least one communication path, focusing on the energy efficiency. In addition, unlike other methods the proposed method considers a practical feature of sensing: The detection probability of sensors decreases as the distance from the target increases. Other methods assume the binary model where the success of target detection entirely depends on whether a target is within the threshold distance from the sensor or not. Our method utilizes the probabilistic sensing model instead of the binary model. Simulation results show that the proposed method outperforms others in terms of the network lifetime.