We introduce an algorithm where the individual bits representing the weights of a neural network are learned. This method allows training weights with integer values on arbitrary bit-depths and naturally uncovers sparse networks, without additional constraints or regularization techniques. We show better results than the standard training technique with fully connected networks and similar performance as compared to standard training for convolutional and residual networks. By training bits in a selective manner we found that the biggest contribution to achieving high accuracy is given by the first three most significant bits, while the rest provide an intrinsic regularization. As a consequence more than 90\% of a network can be used to store arbitrary codes without affecting its accuracy. These codes may be random noise, binary files or even the weights of previously trained networks.

We present some novel, straightforward methods for training the connection graph of a randomly initialized neural network without training the weights. These methods do not use hyperparameters defining cutoff thresholds and therefore remove the need for iteratively searching optimal values of such hyperparameters. We can achieve similar or higher performances than in the case of training all weights, with a similar computational cost as for standard training techniques. Besides switching connections on and off, we introduce a novel way of training a network by flipping the signs of the weights. If we try to minimize the number of changed connections, by changing less than 10 total it is already possible to reach more than 90 standard training. We obtain good results even with weights of constant magnitude or even when weights are drawn from highly asymmetric distributions. These results shed light on the over-parameterization of neural networks and on how they may be reduced to their effective size.

The study shows the methodology for measuring and assessing the radon concentration in indoor and outdoor environments. The results analyzed in the paper show the radioactive measurements of the natural radiation background from the IFIN-HH and ELI-NP, Romania, for radiometrical purposes, for public health and radiological security. The indoor radon concentration varies from 42 to 154 Bq/m(3) with an average value of 109.4 Bq/m(3), and the outdoor concentration varies from 15 to 28 Bq/m(3) with the average value of 23.1 Bq/m(3). Also, we developed a research study in order to investigate the long-term behavior of the diurnal and seasonal fluctuations of radon Rn-222 and the influence of environmental conditions and other climatic factors. In correlation with the radon concentration measured, the effective dose was also calculated.

Recent progress in machine learning techniques have revived interest in building artificial general intelligence using these particular tools. There has been a tremendous success in applying them for narrow intellectual tasks such as pattern recognition, natural language processing and playing Go. The latter application vastly outperforms the strongest human player in recent years. However, these tasks are formalized by people in such ways that it has become easy for automated recipes to find better solutions than humans do. In the sense of John Searle's Chinese Room Argument, the computer playing Go does not actually understand anything from the game. Thinking like a human mind requires to go beyond the curve fitting paradigm of current systems. There is a fundamental limit to what they can achieve currently as only very specific problem formalization can increase their performances in particular tasks. In this paper, we argue than one of the most important aspects of the human mind is its capacity for logical thinking, which gives rise to many intellectual expressions that differentiate us from animal brains. We propose to model the emergence of logical thinking based on Piaget's theory of cognitive development.

Convolutional Neural Networks (CNNs) are build specifically for computer vision tasks for which it is known that the input data is a hierarchical structure based on locally correlated elements. The question that naturally arises is what happens with the performance of CNNs if one of the basic properties of the data is removed, e.g. what happens if the image pixels are randomly permuted? Intuitively one expects that the convolutional network performs poorly in these circumstances in contrast to a multilayer perceptron (MLPs) whose classification accuracy should not be affected by the pixel randomization. This work shows that by randomizing image pixels the hierarchical structure of the data is destroyed and long range correlations are introduced which standard CNNs are not able to capture. We show that their classification accuracy is heavily dependent on the class similarities as well as the pixel randomization process. We also indicate that dilated convolutions are able to recover some of the pixel correlations and improve the performance.

This paper describes the standardization of 67Cu by the 4πβ(PC)-γ coincidence method, and the calibration of ionization chamber. The difficulties were. (i) One of the excited levels of 67Zn has a half life of 9.10 μs. (ii) A recent publication reconsiders the decay scheme parameters. (iii) The solution had a significant content of impurities. The conclusions were that the result of absolute standardization is less influenced by the decay scheme parameters and impurities than the measurements by ionization chamber.

High field quantum electrodynamics experiments will be conducted in the E6 experimental area of the Extreme Light Infrastructure-Nuclear Physics building. Here electrons and protons will be accelerated up to relativistic energies by multi-petawatt laser beam—target interactions. In this respect, the requirements for radiological safety measures are similar to those associated with the operation of conventional high energy accelerators. The paper presents a FLUKA simulation approach to the shielding assessment of the individual experiments. Updated source terms were used in order to compute ambient dose equivalent rates throughout E6 and neighbouring areas and check the compliance of the results with legal dose constraints. We investigated the effectiveness of an ‘all-purpose’ beam dump at E6 and the practicality of local muon shielding.

Charged-particle production was studied in proton–proton collisions collected at the LHC with the ALICE detector at centre-of-mass energies 0.9 TeV and 2.36 TeV in the pseudorapidity range |η| < 1.4. In the central region (|η| < 0.5), at 0.9 TeV, we measure charged-particle pseudorapidity density dNch/dη = 3.02 ± 0.01(stat.) −0.05(syst.) for inelastic interactions, and dNch/dη = 3.58 ± 0.01(stat.) −0.12(syst.) for non-single-diffractive interactions. At 2.36 TeV, we find dNch/dη = 3.77 ± 0.01(stat.) −0.12(syst.) for inelastic, and dNch/dη = 4.43 ± 0.01(stat.) −0.12(syst.) for non-singlediffractive collisions. The relative increase in charged-particle multiplicity from the lower to higher energy is 24.7% ± 0.5%(stat.) −2.8%(syst.) for inelastic and 23.7% ± 0.5%(stat.) −1.1%(syst.) for nonsingle-diffractive interactions. This increase is consistent with that reported by the CMS collaboration for non-single-diffractive events and larger than that found by a number of commonly used models. The multiplicity distribution was measured in different pseudorapidity intervals and studied in terms of KNO variables at both energies. The results are compared to proton–antiproton data and to model

High power lasers have proven being capable to produce high energy gamma rays, charged particles and neutrons to induce all kinds of nuclear reactions. At ELI, the studies with high power lasers will enter for the first time into new domains of power and intensities.

Liquid scintillation samples were filled with (222)Rn and the activity was measured with good precision after reaching the secular equilibrium with the progeny (218)Po, (214)Pb, (214)Bi and (214)Po. After decay of most of (222)Rn activity, the samples contain (210)Pb and progeny. The activities of (210)Pb and progeny can be calculated as a function of time using the initial (222)Rn activity. The samples were measured in a TDCR counter and the experimentally determined counting efficiencies are in accordance with previously published results.

In the present report we propose radiological protection and dosimetry technical systems for ELI-NP facility. The study includes assessments of the facility's radiological installations and experimental areas, as well the location and nature of detectors/instruments that should be appropriate to cover the three main areas of dosimetry for ELI-NP: personnel dosimetry, area dosimetry and environment dosimetry. The devices dedicated to the above first three items must fulfill the legal requirements provided by the regulatory bodies in this field, and will also serve to obtain the operational licenses from the Romanian Nuclear Authority (CNCAN). In order to fulfill the regulatory requirements (Romanian and International) we consider mandatory to consult calibration facilities and methods for the dosimetric systems special characteristics. We identified the structure required in Romania and European countries and a common organizational plan to efficiently design, supervise and monitor all aspects of safety, radiological protection and dosimetry for ELI-NP, as well the required budget and personnel.

This paper refers at the Radionuclide Metrology Laboratory (RML) work accomplished since the last scientific event "2008 Workshop of the Decay Data Evaluation Project" (DDEP-2008), consisting in training sessions for the nuclear decay data evaluators, Bucharest (Romania), during the period 12-14 May, 2008, when the main author presented the activities of the RML from IFIN-HH.

The paper presents the work deployed to improve the measurement uncertainty, for the establishment of the international equivalence and national metrological traceability chain in the activity measurement of 177 Lu. The work is in direct connection with two international projects, involving this radionuclide. Some aspects are analyzed: (i) Absolute standardization of the solution by the method of 4π(PC)β-γ coincidence method; (ii) Measurements by the gamma-ray spectrometry; (iii) Establishment of the equivalence degree through the key comparisons; (iv) Calibration of the secondary standard, ionization chamber.

Journal Article Abstracts of the presentations for the young scientists and professionals award Get access Radiation Protection Dosimetry, Volume 164, Issue 1-2, April 2015, Pages 175–178, https://doi.org/10.1093/rpd/ncu332 Published: 05 February 2015

The paper describes the measurements performed at IFIN-HH regarding the creation of a Romanian (124)I standard, consisting of: absolute standardization of the solution by the application of the 4πβ(PC)-γ coincidence method; Calibration of the CENTRONIC IG12/20A ionization chamber with a standardised solution and comparison with a calculated efficiency; γ-ray spectrometry activity measurement and determination of the impurity levels; Comparison of the results of the three methods.

The paper presents the implication of the Radionuclide Metrology Laboratory (RML) from IFIN-HH in solving the problems connected with the quality assurance of radiopharmaceuticals in production and use, by assuring the metrological traceability in the control of the radiometrological parameters: measurement of activity and radioactive concentration, radionuclide purity as well as measurement of the decay scheme parameters. The following problems are treated: (i) Absolute standardization of PET radionuclides: Cu-64 and Ga-68 and study of their decay scheme parameters; absolute standardization of radionuclides used in targeted - molecular radiotherapy (MRT): Sm-153, Lu-177, Re-186,Re-188 and Sr-89; calibration of the secondary standard ionization chamber and of some commercial radionuclide calibrators. (ii) Radionuclide purity analysis and its influence on the activity calculation. The aspects presented are an overview of published papers.

The work accomplished within the participation at the 2012 key comparison of Tc-99 is presented. The solution was standardized for the first time in IFIN-HH by two methods: LSC-TDCR and 4π(PC)β-γ efficiency tracer. The methods are described and the results are compared. For the LSC-TDCR method, the program TDCR07c, written and provided by P. Cassette, was used for processing the measurement data. The results are 2.1% higher than when applying the TDCR06b program; the higher value, calculated with the software TDCR07c, was used for reporting the final result in the comparison. The tracer used for the 4π(PC)β-γ efficiency tracer method was a standard (60)Co solution. The sources were prepared from the mixture (60)Co+(99)Tc solution and a general extrapolation curve, type: N(βTc-99)/(M)(Tc-99)=f [1-ε(Co-60)], was drawn. This value was not used for the final result of the comparison. The difference between the values of activity concentration obtained by the two methods was within the limit of the combined standard uncertainty of the difference of these two results.

This paper presents the work performed at IFIN-HH and at the users' sites, aimed to establish the (18)FDG Romanian PET traceability chain. It summarizes the operations: (i) Absolute standardization of a (18)F solution, by the 4πβ(PC)-γ coincidence method; (ii) Control of gamma-ray impurity content and measurement of the activity by gamma ray spectrometry; (iii) Measurement of the half life, for detection of some short/long life positron emitter impurities; (iv) Calibration of the CENTRONIC IG12/20A ionization chamber; (v) Calibration of commercial radionuclide calibrators.

The average transverse momentum (p(T)) versus the charged-particle multiplicity N-ch was measured in p-Pb collisions at a collision energy per nucleon-nucleon root S-NN = 5.02 TeV and in pp collisions at collision energies of root s = 0.9, 2.76, and 7 TeV in the kinematic range 0.15 < p(T) < 10.0 GeV/c and vertical bar eta vertical bar < 0.3 with the ALICE apparatus at the LHC. These data are compared to results in Pb-Pb collisions at root S-NN = 2.76 TeV at similar charged-particle multiplicities. In pp and p-Pb collisions, a strong increase of (p(T)) with N-ch is observed, which is much stronger than that measured in Pb-Pb collisions. For pp collisions, this could be attributed, within a model of hadronizing strings, to multiple-parton interactions and to a final-state color reconnection mechanism. The data in p-Pb and Pb-Pb collisions cannot be described by an incoherent superposition of nucleon-nucleon collisions and pose a challenge to most of the event generators. (C) 2013 CERN. Published by Elsevier B.V. All rights reserved.