Search

We design and analyze a new paradigm for building supervised learning networks, driven only by local optimization rules without relying on a global error function. Traditional neural networks with a fixed topology are made up of identical nodes and derive their expressiveness from an appropriate adjustment of connection weights. In contrast, our network stores new knowledge in the nodes accurately and instantaneously, in the form of a lookup table. Only then is some of this information structured and incorporated into the network geometry. The training error is initially zero by construction and remains so throughout the network topology transformation phase. The latter involves a small number of local topological transformations, such as splitting or merging of nodes and adding binary connections between them. The choice of operations to be carried out is only driven by optimization of expressivity at the local scale. What we are primarily looking for in a learning network is its ability to generalize, i.e. its capacity to correctly answer questions for which it has never learned the answers. We show on numerous examples of classification tasks that the networks generated by our algorithm systematically reach such a state of perfect generalization when the number of learned examples becomes sufficiently large. We report on the dynamics of the change of state and show that it is abrupt and has the distinctive characteristics of a first order phase transition, a phenomenon already observed for traditional learning networks and known as grokking. In addition to proposing a non-potential approach for the construction of learning networks, our algorithm makes it possible to rethink the grokking transition in a new light, under which acquisition of training data and topological structuring of data are completely decoupled phenomena., Comment: This paper is submitted to Phys. Rev. X. It's a physicist's study that focus on a new paradigm for deep learning networks. We would have liked to choose other keywords for arXiv to reach a wider community, but don't have the rights to do so

We show that a network of non-identical nodes, with excitable dynamics, pulse-coupled, with coupling delays depending on the Euclidean distance between nodes, is able to adapt the topology of its connections to obtain spike frequency synchronization. The adapted network exhibits remarkable properties: sparse, anti-cluster, necessary presence of a minimum of inhibitory nodes, predominance of connections from inhibitory nodes over those from excitatory nodes and finally spontaneous spatial structuring of the inhibitory projections: the furthest the most intense.

Regimes of operation in tokamaks that are devoid of large ELMs have to be better understood to extrapolate their applicability to reactor-relevant devices. This paper describes non-linear extended MHD simulations that use an experimental equilibrium from an EDA H-mode in ASDEX Upgrade. Linear ideal MHD analysis indicates that the operational point lies slightly inside of the stable region. The non-linear simulations with the visco-resistive extended MHD code, JOREK, sustain non-axisymmetric perturbations that are linearly most unstable with toroidal mode numbers of n = \{6 \dots 9\}, but non-linearly higher and lower n become driven and the low-n become dominant. The poloidal mode velocity during the linear phase is found to correspond to the expected velocity for resistive ballooning modes. The perturbations that exist in the simulations have somewhat smaller poloidal wavenumbers (k_{\theta} \sim 0.1 to 0.5 cm^{-1} ) than the experimental expectations for the quasi-coherent mode in EDA, and cause non-negligible transport in both the heat and particle channels. In the transition from linear to non-linear phase, the mode frequency chirps down from approximately 35 kHz to 13 kHz, which corresponds approximately to the lower end of frequencies that are typically observed in EDA H-modes in ASDEX Upgrade.

In some conditions, I-mode plasmas can feature pedestal relaxation events (PREs) that transiently enhance the energy reaching the divertor target plates. To shed light into their appearance, characteristics and energy reaching the divertor targets, a comparative study between two tokamaks $-$ Alcator C-Mod and ASDEX Upgrade $-$ is carried out. It is found that PREs appear only in a subset of I-mode discharges, mainly when the plasma is close to the H-mode transition. Also, the nature of the triggering instability is discussed by comparing measurements close to the separatrix in both devices. The PRE relative energy loss from the confined region increases with decreasing pedestal top collisionality $\nu_{\mathrm{ped}}^*$. In addition, the relative electron temperature drop at the pedestal top, which is related to the conductive energy loss, rises with decreasing $\nu_{\mathrm{ped}}^*$. Finally, the peak parallel energy fluence due to the PRE measured on the divertor in both devices is compared to the model introduced in [1] for type-I ELMs. The model is shown to provide an upper boundary for PRE energy fluence data, while a lower boundary is found by dividing the model by three. These two boundaries are used to make projections to future devices such as DEMO and ARC.

The I-mode confinement regime can feature small edge temperature drops that can lead to an increase in the energy deposited onto the divertor targets. In this work, we show that these events are associated with a relaxation of both electron temperature and density edge profiles, with the largest drop found at the pedestal top position. Stability analysis of edge profiles reveals that the operational points are far from the ideal peeling-ballooning boundary. Also, we show that these events appear close to the H-mode transition in the typical I-mode operational space in ASDEX Upgrade, and that no further enhancement of energy confinement is found when they occur. Moreover, scrape-off layer transport during these events is found to be very similar to type-I ELMs, with regard to timescales ($\approx$ 800 $\mu$s), filament propagation, toroidally asymmetric energy effluxes at the midplane and asymmetry between inner and outer divertor deposited energy. In particular, the latter reveals that more energy reaches the outer divertor target. Lastly, first measurements of the divertor peak energy fluence are reported, and projections to ARC - a reactor designed to operate in I-mode - are drawn.

Based on a biophysical model of retinal Starburst Amacrine Cell (SAC) \cite{karvouniari-gil-etal:19} we analyse here the dynamics of retinal waves, arising during the visual system development. Waves are induced by spontaneous bursting of SACs and their coupling via acetycholine. We show that, despite the acetylcholine coupling intensity has been experimentally observed to change during development \cite{zheng-lee-etal:04}, SACs retinal waves can nevertheless stay in a regime with power law distributions, reminiscent of a critical regime. Thus, this regime occurs on a range of coupling parameters instead of a single point as in usual phase transitions. We explain this phenomenon thanks to a coherence-resonance mechanism, where noise is responsible for the broadening of the critical coupling strength range.

Signatures of the processes in the early Universe are imprinted in the cosmic web. Some of them may define shell-like structures characterised by typical scales. We search for shell-like structures in the distribution of nearby rich clusters of galaxies drawn from the SDSS DR8. We calculate the distance distributions between rich clusters of galaxies, and groups and clusters of various richness, look for the maxima in the distance distributions, and select candidates of shell-like structures. We analyse the space distribution of groups and clusters forming shell walls. We find six possible candidates of shell-like structures, in which galaxy clusters have maxima in the distance distribution to other galaxy groups and clusters at the distance of about 120 Mpc/h. The rich galaxy cluster A1795, the central cluster of the Bootes supercluster, has the highest maximum in the distance distribution of other groups and clusters around them at the distance of about 120 Mpc/h among our rich cluster sample, and another maximum at the distance of about 240 Mpc/h. The structures of galaxy systems causing the maxima at 120 Mpc/h form an almost complete shell of galaxy groups, clusters and superclusters. The richest systems in the nearby universe, the Sloan Great Wall, the Corona Borealis supercluster and the Ursa Major supercluster are among them. The probability that we obtain maxima like this from random distributions is lower than 0.001. Our results confirm that shell-like structures can be found in the distribution of nearby galaxies and their systems. The radii of the possible shells are larger than expected for a BAO shell (approximately 109 Mpc/h versus approximately 120 Mpc/h), and they are determined by very rich galaxy clusters and superclusters with high density contrast while BAO shells are barely seen in the galaxy distribution. We discuss possible consequences of these differences., Comment: Comments: 9 pages, 10 figures, Astronomy and Astrophysics, in press

No abstract available, since this is a Comment on a published article (cf. title), Comment: This paper is withdrawn, since the Editor suggested that a new version be resubmitted as a Letter. The new version is entitled: Letter to the Editor of Spectrochimica Acta B: External versus self-induced ac Stark shift

The Javalambre-Physics of the Accelerated Universe Astrophysical Survey (J-PAS) is a narrow band, very wide field Cosmological Survey to be carried out from the Javalambre Observatory in Spain with a purpose-built, dedicated 2.5m telescope and a 4.7 sq.deg. camera with 1.2Gpix. Starting in late 2015, J-PAS will observe 8500sq.deg. of Northern Sky and measure $0.003(1+z)$ photo-z for $9\times10^7$ LRG and ELG galaxies plus several million QSOs, sampling an effective volume of $\sim 14$ Gpc$^3$ up to $z=1.3$ and becoming the first radial BAO experiment to reach Stage IV. J-PAS will detect $7\times 10^5$ galaxy clusters and groups, setting constrains on Dark Energy which rival those obtained from its BAO measurements. Thanks to the superb characteristics of the site (seeing ~0.7 arcsec), J-PAS is expected to obtain a deep, sub-arcsec image of the Northern sky, which combined with its unique photo-z precision will produce one of the most powerful cosmological lensing surveys before the arrival of Euclid. J-PAS unprecedented spectral time domain information will enable a self-contained SN survey that, without the need for external spectroscopic follow-up, will detect, classify and measure $\sigma_z\sim 0.5\%$ redshifts for $\sim 4000$ SNeIa and $\sim 900$ core-collapse SNe. The key to the J-PAS potential is its innovative approach: a contiguous system of 54 filters with $145\AA$ width, placed $100\AA$ apart over a multi-degree FoV is a powerful "redshift machine", with the survey speed of a 4000 multiplexing low resolution spectrograph, but many times cheaper and much faster to build. The J-PAS camera is equivalent to a 4.7 sq.deg. "IFU" and it will produce a time-resolved, 3D image of the Northern Sky with a very wide range of Astrophysical applications in Galaxy Evolution, the nearby Universe and the study of resolved stellar populations., Comment: 215 pages, 70 figures. J-PAS "Red Book", containing a description of the technical aspects of the Survey and its main scientific goals

Following an overview of modeling of (longitudinal) multimode semiconductor laser dynamics, we analyze in detail a model proposed in 2006 to explain deterministic, phase-locked modal alternation, experimentally observed a decade ago. Through a stability analysis, we prove that the numerically obtained electromagnetic field evolution, interpreted as an explanation of the experiments, is nothing more than an extremely long transient, so long as to be hardly identifiable in an entirely numerical approach. Comparison with a model we have recently derived, which predicts a phase instability (Benjamin-Feir-like) compatible with the experimental observations, highlights the crucial ingredient for the dynamics. The wide spectrum of unstable eigenvalues accompanying the phase instability plays the role of an equivalent noise in a fully deterministic description, thus reconciling the heuristic models which could qualitatively reproduce the experimental observation either with deterministic equations in the presence of mode-coupling, or through stochastically driven modal decompositions., Comment: 11 pages. Submitted for the Conference Proceedings of Photonics Europe 2014 -- paper # 9134-38

Thanks to a new derivation of the fundamental equations governing multimode dynamics for a semiconductor laser near its threshold, we identify regimes of existence of a pure phase instability (and of a mixed phase-amplitude turbulence regime) which give an alternative satisfactory interpre- tation of the deterministic multimode dynamics observed in some devices. The existence of intrinsic noise generated by the phase instability reconciles in the same description the deterministic and random features of the semiconductor dynamics.

We study the clustering of galaxies as function of luminosity and redshift in the range $0.35 < z < 1.25$ using data from the Advanced Large Homogeneous Area Medium Band Redshift Astronomical (ALHAMBRA) survey. The ALHAMBRA data used in this work cover $2.38 \mathrm{deg}^2$ in 7 independent fields, after applying a detailed angular selection mask, with accurate photometric redshifts, $\sigma_z \lesssim 0.014 (1+z)$, down to $I_{\rm AB} < 24$. Given the depth of the survey, we select samples in $B$-band luminosity down to $L^{\rm th} \simeq 0.16 L^{*}$ at $z = 0.9$. We measure the real-space clustering using the projected correlation function, accounting for photometric redshifts uncertainties. We infer the galaxy bias, and study its evolution with luminosity. We study the effect of sample variance, and confirm earlier results that the COSMOS and ELAIS-N1 fields are dominated by the presence of large structures. For the intermediate and bright samples, $L^{\rm med} \gtrsim 0.6L^{*}$, we obtain a strong dependence of bias on luminosity, in agreement with previous results at similar redshift. We are able to extend this study to fainter luminosities, where we obtain an almost flat relation, similar to that observed at low redshift. Regarding the evolution of bias with redshift, our results suggest that the different galaxy populations studied reside in haloes covering a range in mass between $\log_{10}[M_{\rm h}/(h^{-1}\mathrm{M}_{\odot})] \gtrsim 11.5$ for samples with $L^{\rm med} \simeq 0.3 L^{*}$ and $\log_{10}[M_{\rm h}/(h^{-1}\mathrm{M}_{\odot})] \gtrsim 13.0$ for samples with $L^{\rm med} \simeq 2 L^{*}$, with typical occupation numbers in the range of $\sim 1 - 3$ galaxies per halo., Comment: 21 pages, 16 figures, 3 tables. Accepted for publication in MNRAS. v2: matches accepted version. Small changes following referee comments, including addition of new fig. 11 and discussion on halo occupation numbers. Main results remain unchanged

Squeezed many-body states of atoms are a valuable resource for high precision frequency metrology and could tremendously boost the performance of atomic lattice clocks. Here, we theoretically demonstrate a viable approach to spin squeezing in lattice clocks via optical dressing of one clock state to a highly excited Rydberg state, generating switchable atomic interactions. For realistic experimental parameters, this is shown to generate over 10 dB of squeezing in a few microseconds interaction time without affecting the subsequent clock interrogation.

We present a study of Goldstone boson scattering based on the flavor SU(3) chiral Lagrangian formulated with vector mesons in the tensor field representation. A coupled-channel channel computation is confronted with the empirical s- and p-wave phase shifts, where good agreement with the data set is obtained up to about 1.2 GeV. There are two relevant free parameters only, the chiral limit value of the pion decay constant and the coupling constant characterizing the decay of the rho meson into a pair of pions. We apply a recently suggested approach that implements constraints from micro- causality and coupled-channel unitarity. Generalized potentials are obtained from the chiral Lagrangian and are expanded in terms of suitably constructed conformal variables. The partial-wave scattering amplitudes are defined as solutions of non-linear integral equations that are solved by means of an N/D ansatz., Comment: 15 pages, 8 figures, typos corrected, accepted for publication in Physics Letters B