004 Datenverarbeitung; Informatik
Refine
Year of publication
Document Type
- Article (19) (remove)
Has Fulltext
- yes (19)
Is part of the Bibliography
- no (19)
Keywords
- synaptic plasticity (4)
- Fisher information (3)
- Hebbian learning (3)
- objective functions (3)
- deep learning (2)
- generating functionals (2)
- homeostatic adaption (2)
- CBM detector (1)
- CLVisc (1)
- Deep learning (1)
Institute
- Physik (19) (remove)
In this proceeding, we review our recent work using deep convolutional neural network (CNN) to identify the nature of the QCD transition in a hybrid modeling of heavy-ion collisions. Within this hybrid model, a viscous hydrodynamic model is coupled with a hadronic cascade “after-burner”. As a binary classification setup, we employ two different types of equations of state (EoS) of the hot medium in the hydrodynamic evolution. The resulting final-state pion spectra in the transverse momentum and azimuthal angle plane are fed to the neural network as the input data in order to distinguish different EoS. To probe the effects of the fluctuations in the event-by-event spectra, we explore different scenarios for the input data and make a comparison in a systematic way. We observe a clear hierarchy in the predictive power when the network is fed with the event-by-event, cascade-coarse-grained and event-fine-averaged spectra. The carefully-trained neural network can extract high-level features from pion spectra to identify the nature of the QCD transition in a realistic simulation scenario.
The Fisher information constitutes a natural measure for the sensitivity of a probability distribution with respect to a set of parameters. An implementation of the stationarity principle for synaptic learning in terms of the Fisher information results in a Hebbian self-limiting learning rule for synaptic plasticity. In the present work, we study the dependence of the solutions to this rule in terms of the moments of the input probability distribution and find a preference for non-Gaussian directions, making it a suitable candidate for independent component analysis (ICA). We confirm in a numerical experiment that a neuron trained under these rules is able to find the independent components in the non-linear bars problem. The specific form of the plasticity rule depends on the transfer function used, becoming a simple cubic polynomial of the membrane potential for the case of the rescaled error function. The cubic learning rule is also an excellent approximation for other transfer functions, as the standard sigmoidal, and can be used to show analytically that the proposed plasticity rules are selective for directions in the space of presynaptic neural activities characterized by a negative excess kurtosis.
We present an effective model for timing-dependent synaptic plasticity (STDP) in terms of two interacting traces, corresponding to the fraction of activated NMDA receptors and the concentration in the dendritic spine of the postsynaptic neuron. This model intends to bridge the worlds of existing simplistic phenomenological rules and highly detailed models, thus constituting a practical tool for the study of the interplay of neural activity and synaptic plasticity in extended spiking neural networks. For isolated pairs of pre- and postsynaptic spikes, the standard pairwise STDP rule is reproduced, with appropriate parameters determining the respective weights and timescales for the causal and the anticausal contributions. The model contains otherwise only three free parameters, which can be adjusted to reproduce triplet nonlinearities in hippocampal culture and cortical slices. We also investigate the transition from time-dependent to rate-dependent plasticity occurring for both correlated and uncorrelated spike patterns.
Generating functionals may guide the evolution of a dynamical system and constitute a possible route for handling the complexity of neural networks as relevant for computational intelligence.We propose and explore a new objective function, which allows to obtain plasticity rules for the afferent synaptic weights. The adaption rules are Hebbian, self-limiting, and result from the minimization of the Fisher information with respect to the synaptic flux. We perform a series of simulations examining the behavior of the new learning rules in various circumstances.The vector of synaptic weights aligns with the principal direction of input activities, whenever one is present. A linear discrimination is performed when there are two or more principal directions; directions having bimodal firing-rate distributions, being characterized by a negative excess kurtosis, are preferred. We find robust performance and full homeostatic adaption of the synaptic weights results as a by-product of the synaptic flux minimization. This self-limiting behavior allows for stable online learning for arbitrary durations.The neuron acquires new information when the statistics of input activities is changed at a certain point of the simulation, showing however, a distinct resilience to unlearn previously acquired knowledge. Learning is fast when starting with randomly drawn synaptic weights and substantially slower when the synaptic weights are already fully adapted.
Poster presentation: The brain is autonomously active and this self-sustained neural activity is in general modulated, but not driven, by the sensory input data stream [1,2]. Traditionally one has regarded this eigendynamics as resulting from inter-modular recurrent neural activity [3]. Understanding the basic modules for cognitive computation is, in this view, the primary focus of research and the overall neural dynamics would be determined by the the topology of the intermodular pathways. Here we examine an alternative point of view, asking whether certain aspects of the neural eigendynamics have a central functional role for overall cognitive computation [4,5]. Transiently stable neural activity is regularly observed on the cognitive time-scale of 80–100 ms, with indications that neural competition [6] plays an important role in the selection of the transiently stable neural ensembles [7], also denoted winning coalitions [8]. We report on a theory approach which implements these two principles, transient-state dynamics and neural competition, in terms of an associative neural network with clique encoding [9]. A cognitive system [10] with a non-trivial internal eigendynamics has two seemingly contrasting tasks to fulfill. The internal processes need to be regular and not chaotic on one side, but sensitive to the afferent sensory stimuli on the other side. We show, that these two contrasting demands can be reconciled within our approach based on competitive transient-state dynamics, when allowing the sensory stimuli to modulate the competition for the next winning coalition. By testing the system with the bars problem, we find an emerging cognitive capability. Only based on the two basic architectural principles, neural competition and transient-state dynamics, with no explicit algorithmic encoding, the system performs on its own a non-linear independent component analysis of input data stream. The system has rudimentary biological features. All learning is local Hebbian-style, unsupervised and online. It exhibits an ever-ongoing eigendynamics and at no time is the state or the value of synaptic strengths reset or the system restarted; there is no separation between training and performance. We believe that this kind of approach – cognitive computation with autonomously active neural networks – to be an emerging field, relevant both for system neuroscience and synthetic cognitive systems.
Understanding the dynamics of recurrent neural networks is crucial for explaining how the brain processes information. In the neocortex, a range of different plasticity mechanisms are shaping recurrent networks into effective information processing circuits that learn appropriate representations for time-varying sensory stimuli. However, it has been difficult to mimic these abilities in artificial neural network models. Here we introduce SORN, a self-organizing recurrent network. It combines three distinct forms of local plasticity to learn spatio-temporal patterns in its input while maintaining its dynamics in a healthy regime suitable for learning. The SORN learns to encode information in the form of trajectories through its high-dimensional state space reminiscent of recent biological findings on cortical coding. All three forms of plasticity are shown to be essential for the network's success. Keywords: synaptic plasticity, intrinsic plasticity, recurrent neural networks, reservoir computing, time series prediction
Recent experimental findings have reported the presence of unconventional charge orders in the enlarged (2 × 2) unit-cell of kagome metals AV3Sb5 (A = K, Rb, Cs) and hinted towards specific topological signatures. Motivated by these discoveries, we investigate the types of topological phases that can be realized in such kagome superlattices. In this context, we employ a recently introduced statistical method capable of constructing topological models for any generic lattice. By analyzing large data sets generated from symmetry-guided distributions of randomized tight-binding parameters, and labeled with the corresponding topological index, we extract physically meaningful information. We illustrate the possible real-space manifestations of charge and bond modulations and associated flux patterns for different topological classes, and discuss their relation to present theoretical predictions and experimental signatures for the AV3Sb5 family. Simultaneously, we predict higher-order topological phases that may be realized by appropriately manipulating the currently known systems.
In this talk we presented a novel technique, based on Deep Learning, to determine the impact parameter of nuclear collisions at the CBM experiment. PointNet based Deep Learning models are trained on UrQMD followed by CBMRoot simulations of Au+Au collisions at 10 AGeV to reconstruct the impact parameter of collisions from raw experimental data such as hits of the particles in the detector planes, tracks reconstructed from the hits or their combinations. The PointNet models can perform fast, accurate, event-by-event impact parameter determination in heavy ion collision experiments. They are shown to outperform a simple model which maps the track multiplicity to the impact parameter. While conventional methods for centrality classification merely provide an expected impact parameter distribution for a given centrality class, the PointNet models predict the impact parameter from 2–14 fm on an event-by-event basis with a mean error of −0.33 to 0.22 fm.
The state-of-the-art pattern recognition method in machine learning (deep convolution neural network) is used to identify the equation of state (EoS) employed in the relativistic hydrodynamic simulations of heavy ion collisions. High-level correlations of particle spectra in transverse momentum and azimuthal angle learned by the network act as an effective EoS-meter in deciphering the nature of the phase transition in QCD. The EoS-meter is model independent and insensitive to other simulation inputs including the initial conditions and shear viscosity for hydrodynamic simulations. Through this study we demonstrate that there is a traceable encoder of the dynamical information from the phase structure that survives the evolution and exists in the final snapshot of heavy ion collisions and one can exclusively and effectively decode these information from the highly complex final output with machine learning when traditional methods fail. Besides the deep neural network, the performance of traditional machine learning classifiers are also provided.