Deep discriminative models provide remarkable insights into hierarchical processing in the brain by predicting neural activity along the visual pathway. However, these models differ from biological systems in their computational and architectural properties. Unlike biological systems, they require teaching signals for supervised learning. Moreover, they rely on feed-forward processing of stimuli, which contrasts with the extensive top-down connections in the ventral pathway. Here, we address both issues by developing a hierarchical deep generative model and show that it predicts an extensive set of experimental results in the primary and secondary visual cortices (V1 and V2). We show that the widely documented sensitivity of V2 neurons to textures is a consequence of learning a hierarchical representation of natural images. Further, we show that top-down influences are inherent to hierarchical inference. Hierarchical inference explains neural signatures of top-down interactions and reveals how higher-level representation shapes low-level representations through modulation of response mean and noise correlations in V1.

Optimal computations under uncertainty require an adequate probabilistic representation about beliefs. Deep generative models, and specifically Variational Autoencoders (VAEs), have the potential to meet this demand by building latent representations that learn to associate uncertainties with inferences while avoiding their characteristic intractable computations. Yet, we show that it is precisely uncertainty representation that suffers from inconsistencies under an array of relevant computer vision conditions: contrast-dependent computations, image corruption, out-of-distribution detection. Drawing inspiration from classical computer vision, we present a principled extension to the standard VAE by introducing a simple yet powerful inductive bias through a global scaling latent variable, which we call the Explaining-Away VAE (EA-VAE). By applying EA-VAEs to a spectrum of computer vision domains and a variety of datasets, spanning standard NIST datasets to rich medical and natural image sets, we show the EA-VAE restores normative requirements for uncertainty. Furthermore, we provide an analytical underpinning of the contribution of the introduced scaling latent to contrast-related and out-of-distribution related modulations of uncertainty, demonstrating that this mild inductive bias has stark benefits in a broad set of problems. Moreover, we find that EA-VAEs recruit divisive normalization, a motif widespread in biological neural networks, to remedy defective inference. Our results demonstrate that an easily implemented, still powerful update to the VAE architecture can remedy defective inference of uncertainty in probabilistic computations.

Interpreting computations in the visual cortex as learning and inference in a generative model of the environment has received wide support both in neuroscience and cognitive science. However, hierarchical computations, a hallmark of visual cortical processing, has remained impervious for generative models because of a lack of adequate tools to address it. Here we capitalize on advances in Variational Autoencoders (VAEs) to investigate the early visual cortex with sparse coding hierarchical VAEs trained on natural images. We design alternative architectures that vary both in terms of the generative and the recognition components of the two latent-layer VAE. We show that representations similar to the one found in the primary and secondary visual cortices naturally emerge under mild inductive biases. Importantly, a nonlinear representation for texture-like patterns is a stable property of the high-level latent space resistant to the specific architecture of the VAE, reminiscent of the secondary visual cortex. We show that a neuroscience-inspired choice of the recognition model, which features a top-down processing component is critical for two signatures of computations with generative models: learning higher order moments of the posterior beyond the mean and image inpainting. Patterns in higher order response statistics provide inspirations for neuroscience to interpret response correlations and for machine learning to evaluate the learned representations through more detailed characterization of the posterior.

Erratum to the article published on December 27th 2015 in Issue 52 of Orvosi Hetilap [Orv. Hetil., 2015, 156(52), 2120–2126, DOI: 10.1556/650.2015.30329]. The name of Dávid Mezey was not correctly typed. The corresponding author asked for the following correction to be published.

Erratum to the article published on December 27th 2015 in Issue 52 of Orvosi Hetilap [Orv. Hetil., 2015, 156(52), 2120-2126, DOI: 10.1556/650.2015.30329]. The name of Dávid Mezey was not correctly typed. The corresponding author asked for the following correction to be published.

Introduction: Two-photon microscopy is the ideal tool to study how signals are processed in the functional brain tissue. However, early raster scanning strategies were inadequate to record fast 3D events like action potentials. Aim: The aim of the authors was to record various neuronal activity patterns with high signal-to-noise ratio in an optical manner. Method: Authors developed new data acquisition methods and microscope hardware. Results: Multiple Line Scanning enables the experimenter to select multiple regions of interests, doing this not just increases repetition speed, but also the signal-to-noise ratio of the fluorescence transients. On the same principle, an acousto-optical deflector based 3D scanning microscope has been developed with a sub-millisecond temporal resolution and a millimeter z-scanning range. Its usability is demonstrated by obtaining 3D optical recordings of action potential backpropagation in several hundred micrometers long neuronal processes of single neurons and by 3D random-access scanning of Ca2+ transients in hundreds of neurons in the mouse visual cortex. Conclusions: Region of interest scanning enables high signal-to-noise ratio and repetition speed, while keeping good depth penetration of the two-photon microscopes. Orv. Hetil., 2015, 156(52), 2120–2126.

Introduction: Two-photon microscopy is the ideal tool to study how signals are processed in the functional brain tissue. However, early raster scanning strategies were inadequate to record fast 3D events like action potentials. Aim: The aim of the authors was to record various neuronal activity patterns with high signal-to-noise ratio in an optical manner. Method: Authors developed new data acquisition methods and microscope hardware. Results: Multiple Line Scanning enables the experimenter to select multiple regions of interests, doing this not just increases repetition speed, but also the signal-to-noise ratio of the fluorescence transients. On the same principle, an acousto-optical deflector based 3D scanning microscope has been developed with a sub-millisecond temporal resolution and a millimeter z-scanning range. Its usability is demonstrated by obtaining 3D optical recordings of action potential backpropagation in several hundred micrometers long neuronal processes of single neurons and by 3D random-access scanning of Ca2+ transients in hundreds of neurons in the mouse visual cortex. Conclusions: Region of interest scanning enables high signal-to-noise ratio and repetition speed, while keeping good depth penetration of the two-photon microscopes.

Racsterelmeleti modszerekkel vizsgaltuk a kvantumszindinamikat veges es zerus hőmersekleten es kemiai potencialon. Terjedelmi okokbol a legfontosabb eredmenyeket tekintjuk csak at az osszefoglaloban. Nulla kemiai potencial mellett meghataroztuk a hadronikus anyag es a kvark-gluon plazma kozti atmenet rendjet, valamint az atmeneti homersekletet a kontinuum limeszben. Ezen eredmenyek a racsterelmelet vegleges eredmenyeinek tekinthetők. Osszehasonlitottuk a Wilson es staggered fermionokat 3 iz hasznalataval. Nem-nulla kemiai potencialnal meghataroztuk a kvark-antikvark potencialt valamint a fazisdiagramot nagy kemiai potencialokra az allapotsűrűseg modszer segitsegevel. A pion szektort megvizsgaltuk kanonikus sokasag segitsegevel. Meghataroztuk a mu-T fazisdiagramot kis kemiai potencialokra 4 kulonboző racsallandonal. Racsszamolasokkal megvizsgaltuk a pentakvarkok letezeset es nem talaltunk erre utalo jelet. Kiszamoltuk a hadron spektrumot Wilson fermionokkal a kontinuum limeszben. Meghataroztuk a pion es kaon leptonikus bomlasi allandoinak aranyat. Implementaltuk a QCD kodjainkat grafikus kartyakra. Overlap fermionokat hasznalo korabbi dinamikus algoritmusunkat tovabbfejlesztettuk. | We have studied quantum chromodynamics at finite and zero temperature and chemical potential using lattice field theory methods. Due to space limitations we present here only the most important results. We have determined at zero chemical potential the order of the transition from hadronic matter to the quark-gluon plasma as well as the transition temperature, both in the continuum limit. These results should be considered as final results of lattice field theory. We have compared Wilson and staggered fermions for 3 flavours. We have determined the quark-antiquark potential as well as the phase diagram for large non-zero chemical potentials using the density of states method. We have considered the pion sector using canonical ensembles. We have determined the mu-T phase diagram for small chemical potentials at 4 different lattice spacings. Using lattice simulations we have considered the problem of existence of pentaquarks and found no evidence for them. We have calculated the hadronic spectrum in the continuum limit using Wilson fermions. We have determined the ratio of the pion and kaon leptonic decay constants. We have implemented our QCD computer codes for the graphical processor cards (GPU). Our dynamical algorithms using overlap fermions have also been extended.

The existence of a well-defined yield stress, where a macroscopic crystal begins to plastically flow, has been a basic observation in materials science. In contrast with macroscopic samples, in microcrystals the strain accumulates in random bursts, which makes controlled plastic formation difficult. Here we study by 2D and 3D simulations the plastic deformation of submicron objects under increasing stress. We show that, while the stress-strain relation of individual samples exhibits jumps, its average and mean deviation still specify a well-defined critical stress. The statistical background of this phenomenon is analyzed through the velocity distribution of dislocations, revealing a universal cubic decay and the appearance of a shoulder due to dislocation avalanches.

We study the relaxation dynamics of systems of straight, parallel crystal dislocations, starting from initially random and uncorrelated positions of the individual dislocations. A scaling model of the relaxation process is constructed by considering the gradual extinction of the initial density fluctuations present in the system. The model is validated by ensemble simulations of the discrete dynamics of dislocations. Convincing agreement is found for systems of edge dislocations in single slip irrespective of the net Burgers vector of the dislocation system. It is also demonstrated that the model does not work in multiple-slip geometries.

Numerical studies of dislocation pair correlations have played a central role in deriving a continuum theory from the equations of motion of 2D dislocation systems in a mathematically rigorous way. As part of an effort to extend this theory into the full 3D dislocation problem, 3D dislocation pair correlations were studied with discrete dislocation dynamics simulation. As a first approximation, dislocations were modeled as uncharged curves in space (their Burgers vectors were disregarded). An inverse square decay with distance was found to describe the numerically obtained pair correlations of the studied curve system.

We present the results of our large scale 4-dimensional (4d) lattice simulations for the MSSM electroweak phase transition (EWPT). We carried out infinite volume and continuum limit extrapolations and found a transition whose strength agrees well with perturbation theory. We determined the properties of the bubble wall that are important for a successful baryogenesis.

Ultra-relativistic collisions, so called “Little Bangs” of gold nuclei are observed at the experiments of the Relativistic Heavy Ion Collider (RHIC) of the Brookhaven National Laboratory, New York. The aim of these experiments is to create and investigate new forms of matter that existed in Nature a few microseconds after the Big Bang, the creation of our Universe. An important (though mathematically never proven) property of the theory of the color degree of freedom of the quarks and gluons (Quantum Chromo Dynamics, QCD) is that they are bound into hadrons in a matter of normal temperature and pressure. In the early Universe, energy density was many orders of magnitude higher than that, thus deconfined phases of colored matter might have existed. Quark Gluon Plasma (QGP) was predicted to be such a possible phase. This type of matter is searched for at the RHIC experiments. A consistent picture emerged after the first three years of running the in RHIC experiments: quarks indeed become deconfined, but also behave collectively, hence this hot matter acts like a liquid [1], not like an ideal gas theorists had anticipated when defining the term QGP. The situation is similar to as if prisoners (quarks and gluons confined in hadrons) have broken out of their cells at nearly the same time, but they find themselves on the crowded jail-yard coupled with all the other escapees. This strong coupling is exactly what happens in a liquid [2]. Based on elliptic flow measurements and the broad range success of analytic hydro models, we can make the definitive statement that in relativistic Au+Au collisions observed at RHIC we see a perfect fluid [3, 4]. Based on our estimates on the temperature [5] and energy density [6] we also conclude that the observed matter is in a deconfined state. We also see a possible signal of partial restoration of the chiral UA(1) symmetry via the mass reduction of η’ bosons [7]. Future plan is to explore all properties of the Quark Matter, by analyzing more data and using higher luminosity. We are after the full map of the QCD phase diagram, and in order to explore it, we also have to go to higher energies and compare them to lower energy data. If the Quark Matter is the New World, then Columbus just realized he is not in India, but on a new continent.

A veges hőmersekletű terelmeletekben elert - Nature-ben is kozolt - eredmenyunk szerint az ősrobbanas utani kvark-hadronikus anyag atmenet folytonos, nem fazisatatalakulas. Racsterelmeleti modszerekkel megadtuk ezen atmenet karakterisztikus hőmersekletet. Az integralhato terelmeletek alkalmazhatok a kondenzalt anyagok fizikajatol a nyilt hurelmeleten keresztul egeszen a reszecskefizikaig. Jelentős lepest tettunk a racionalis konform terelmeletek osztalyozasa fele. A statisztikus fizika teruleten kulon figyelmet szenteltunk az egyensulytol tavoli jelensegek kozul a frontoknak. A kemiai, biologia es magneses frontok mozgasat, alakvaltozasait irtuk le es szabalyozasukat dolgoztuk ki. A fizikai mennyisegek extrem ertekeinek statisztikaja az alkalmazasok szempontjabol is fontos uj terulet, melyen belul transzportjelensegek es 1/f tipusu zaj ingadozasait jellemző mennyisegek maximumanak eloszlasat adtuk meg. A diszlokaciorendszerek statisztikus fizikai vizsgalata segitsegevel tobb attorest ertunk el. A kornyezeti aramlasok temakoreben a Karman Laboratorium aktiv kutatohellye fejlődott. A frontoktol kezdve a ciklonkepződesig szamos kiserletet vegeztunk. Kiemelendő, hogy egy forgokadbeli turbulenciaban a hőmerseklet fluktuacioinak statisztikaja reprodukalja a foldi meteorologiai allomasok adataiet. A kiserletek a klimavaltozas vizsgalataban is igeretesek. A palyazati időszakban 21 doktorandusz temavezeteset lattuk el. | Our result - published also in Nature - on finite temperature field theories states that the quark-hadronic matter transition after the Big Bang happened continuously, not like a phase transition. We determined the characteristic crossover temperature using lattice field theoretical methods. Integrable field theories have applications from condensed matter physics to open string theory to particle physics. We made a major step towards the classification of rational conformal field theories. In statistical physics special attention has been payed to fronts, a phenomenon in far from equilibrium systems. We described and controlled motion and shapes of fronts in chemical, biological, and magnetic systems. Extreme statistics of physical quantities is a new area also important from the viewpoint of applications, where we determined the distribution of maxima in transport phenomena and in quantities characterizing fluctuations of 1/f-type noises. Several breakthroughs were achieved by means of statistical physical studies in dislocation systems. As to the topic of environmental flows, the von Karman Laboratory has become an active research unit where experiments ranging from fronts to cyclone-genesis have been carried out. As a highlight, in the turbulence from a rotating tank the statistics of temperature fluctuations reproduced that of data from meteorological stations. The experiments are promising for studies in climate change. In the grant period we supervised 21 PhD students.

Under stress, many crystalline materials exhibit irreversible plastic deformation caused by the motion of lattice dislocations. In plastically deformed microcrystals, internal dislocation avalanches lead to jumps in the stress-strain curves (strain bursts), whereas in macroscopic samples plasticity appears as a smooth process. By combining three-dimensional simulations of the dynamics of interacting dislocations with statistical analysis of the corresponding deformation behavior, we determined the distribution of strain changes during dislocation avalanches and established its dependence on microcrystal size. Our results suggest that for sample dimensions on the micrometer and submicrometer scale, large strain fluctuations may make it difficult to control the resulting shape in a plastic-forming process.

The relaxation of initially random systems of infinite, straight, parallel edge dislocations is studied in single slip.A relaxation model is developed based on the gradual extinction of initial density fluctuations.Exponents of the predicted power law evolution for the mean value of different powers of dislocation velocity are compared to discrete dislocation dynamics simulations.A satisfying match is found in the theoretical limit of large multipole formation rates and a weak coupling between mobile excess dislocations and background multipoles.