Search

Aims. The main aim of this work is to study the evolution of the recently introduced relative helicity of the magnetic polarity inversion line (PIL) in a magnetohydrodynamics simulation. Methods. The simulation used is a typical flux emergence simulation in which there is additionally an oblique, pre-existing magnetic field. The interaction of the emerging and ambient fields produces intense coronal activity, with four jets standing out. The 3D magnetic field allows us to compute various energies and helicities, and to study their evolution during the simulation, especially around the identified jets. We examine the evolution of all quantities in three different regions: in the whole volume, in three separate subvolumes of the whole volume, and in a 2D region around the PIL on the photosphere. Results. We find that the helicities are in general more responsive to the jets, followed by the free energy. The eruptivity index, the ratio of the current-carrying helicity to the relative helicity, does not show the typical behaviour it has in other cases, as its variations do not follow the production of the jets. By considering the subvolumes we find that the magnetic field gets more potential and less helical with height. The PIL relative helicity confirms the recent results it showed in observed active regions, exhibiting stronger variations during the jets compared to the standard relative helicity. Moreover, the current-carrying helicity around the PIL has a similar behaviour to the PIL relative helicity, and so this quantity could be equally useful in solar eruptivity studies., Comment: accepted by Astronomy & Astrophysics

Using photospheric vector magnetograms obtained by the Helioseismic and Magnetic Imager on board the Solar Dynamics Observatory and a magnetic connectivity-based method, we compute the magnetic helicity and free magnetic energy budgets of a simple bipolar solar active region (AR) during its magnetic flux emergence phase which lasted $\sim$47 hrs. The AR did not produce any coronal mass ejections (CMEs) or flares with an X-ray class above C1.0 but it was the site of 60 jet events during its flux emergence phase. The helicity and free energy budgets of the AR were below established eruption-related thresholds throughout the interval we studied. However, in addition to their slowly-varying evolution, each of the time profiles of the helicity and free energy budgets showed discrete localized peaks, with eight pairs of them occurring at times of jets emanating from the AR. These jets featured larger base areas and longer durations than the other jets of the AR. We estimated, for the first time, the helicity and free magnetic energy changes associated with these eight jets which were in the ranges of $0.5-7.1 \times 10^{40}$ Mx$^2$ and $1.1-6.9 \times 10^{29}$ erg, respectively. Although these values are one to two orders of magnitude smaller than those usually associated with CMEs, the relevant percentage changes were significant and ranged from 13% to 76% for the normalized helicity and from 9% to 57% for the normalized free magnetic energy. Our study indicates that occasionally jets may have a significant imprint in the evolution of helicity and free magnetic energy budgets of emerging active regions., Comment: A&A (Letters), in press

Biological neural networks do not only include long-term memory and weight multiplication capabilities, as commonly assumed in artificial neural networks, but also more complex functions such as short-term memory, short-term plasticity, and meta-plasticity - all collocated within each synapse. Here, we demonstrate memristive nano-devices based on SrTiO3 that inherently emulate all these synaptic functions. These memristors operate in a non-filamentary, low conductance regime, which enables stable and energy efficient operation. They can act as multi-functional hardware synapses in a class of bio-inspired deep neural networks (DNN) that make use of both long- and short-term synaptic dynamics and are capable of meta-learning or "learning-to-learn". The resulting bio-inspired DNN is then trained to play the video game Atari Pong, a complex reinforcement learning task in a dynamic environment. Our analysis shows that the energy consumption of the DNN with multi-functional memristive synapses decreases by about two orders of magnitude as compared to a pure GPU implementation. Based on this finding, we infer that memristive devices with a better emulation of the synaptic functionalities do not only broaden the applicability of neuromorphic computing, but could also improve the performance and energy costs of certain artificial intelligence applications.

Context. Relative field line helicity (RFLH) is a recently developed quantity which can approximate the density of relative magnetic helicity. Aims. This paper aims to determine whether RFLH can be used as an indicator of solar eruptivity. Methods. Starting from magnetographic observations from the Helioseismic and Magnetic Imager instrument onboard the Solar Dynamic Observatory of a sample of seven solar active regions (ARs), which comprises over 2000 individual snapshots, we reconstruct the AR's coronal magnetic field with a widely-used non-linear force-free method. This enables us to compute RFLH using two independent gauge conditions for the vector potentials. We focus our study around the times of strong flares in the ARs, above the M class, and in regions around the polarity inversion lines (PILs) of the magnetic field, and of RFLH. Results. We find that the temporal profiles of the relative helicity that is contained in the magnetic PIL follow those of the relative helicity that is computed by the accurate volume method for the whole AR. Additionally, the PIL relative helicity can be used to define a parameter similar to the well-known parameter R (Schrijver 2007), whose high values are related with increased flaring probability. This helicity-based R-parameter correlates well with the original one, showing in some cases even higher values, and additionally, it experiences more pronounced decreases during flares. This means that there exists at least one parameter deduced from RFLH, that has important value as a solar eruptivity indicator., Comment: accepted by Astronomy & Astrophysics

Self-supervised learning (SSL) methods aim to exploit the abundance of unlabelled data for machine learning (ML), however the underlying principles are often method-specific. An SSL framework derived from biological first principles of embodied learning could unify the various SSL methods, help elucidate learning in the brain, and possibly improve ML. SSL commonly transforms each training datapoint into a pair of views, uses the knowledge of this pairing as a positive (i.e. non-contrastive) self-supervisory sign, and potentially opposes it to unrelated, (i.e. contrastive) negative examples. Here, we show that this type of self-supervision is an incomplete implementation of a concept from neuroscience, the Efference Copy (EC). Specifically, the brain also transforms the environment through efference, i.e. motor commands, however it sends to itself an EC of the full commands, i.e. more than a mere SSL sign. In addition, its action representations are likely egocentric. From such a principled foundation we formally recover and extend SSL methods such as SimCLR, BYOL, and ReLIC under a common theoretical framework, i.e. Self-supervision Through Efference Copies (S-TEC). Empirically, S-TEC restructures meaningfully the within- and between-class representations. This manifests as improvement in recent strong SSL baselines in image classification, segmentation, object detection, and in audio. These results hypothesize a testable positive influence from the brain's motor outputs onto its sensory representations., Comment: Accepted at NeurIPS 2022

Recent approximations to backpropagation (BP) have mitigated many of BP's computational inefficiencies and incompatibilities with biology, but important limitations still remain. Moreover, the approximations significantly decrease accuracy in benchmarks, suggesting that an entirely different approach may be more fruitful. Here, grounded on recent theory for Hebbian learning in soft winner-take-all networks, we present multilayer SoftHebb, i.e. an algorithm that trains deep neural networks, without any feedback, target, or error signals. As a result, it achieves efficiency by avoiding weight transport, non-local plasticity, time-locking of layer updates, iterative equilibria, and (self-) supervisory or other feedback signals -- which were necessary in other approaches. Its increased efficiency and biological compatibility do not trade off accuracy compared to state-of-the-art bio-plausible learning, but rather improve it. With up to five hidden layers and an added linear classifier, accuracies on MNIST, CIFAR-10, STL-10, and ImageNet, respectively reach 99.4%, 80.3%, 76.2%, and 27.3%. In conclusion, SoftHebb shows with a radically different approach from BP that Deep Learning over few layers may be plausible in the brain and increases the accuracy of bio-plausible machine learning. Code is available at https://github.com/NeuromorphicComputing/SoftHebb., Comment: Updated to match the published version of the the ICLR 2023 paper (notable-top 25%)

As the modern vehicle becomes more software-defined, it is beginning to take significant effort to avoid serious regression in software design. This is because automotive software architects rely largely upon manual review of code to spot deviations from specified design principles. Such an approach is both inefficient and prone to error. In recent days, neural language models pre-trained on source code are beginning to be used for automating a variety of programming tasks. In this work, we extend the application of such a Programming Language Model (PLM) to automate the assessment of design compliance. Using a PLM, we construct a system that assesses whether a set of query programs comply with Controller-Handler, a design pattern specified to ensure hardware abstraction in automotive control software. The assessment is based upon measuring whether the geometrical arrangement of query program embeddings, extracted from the PLM, aligns with that of a set of known implementations of the pattern. The level of alignment is then transformed into an interpretable measure of compliance. Using a controlled experiment, we demonstrate that our technique determines compliance with a precision of 92%. Also, using expert review to calibrate the automated assessment, we introduce a protocol to determine the nature of the violation, helping eventual refactoring. Results from this work indicate that neural language models can provide valuable assistance to human architects in assessing and fixing violations in automotive software design., Comment: To appear in PROMISE '22

Recently, Strength-based Argumentation Frameworks (StrAFs) have been proposed to model situations where some quantitative strength is associated with arguments. In this setting, the notion of accrual corresponds to sets of arguments that collectively attack an argument. Some semantics have already been defined, which are sensitive to the existence of accruals that collectively defeat their target, while their individual elements cannot. However, until now, only the surface of this framework and semantics have been studied. Indeed, the existing literature focuses on the adaptation of the stable semantics to StrAFs. In this paper, we push forward the study and investigate the adaptation of admissibility-based semantics. Especially, we show that the strong admissibility defined in the literature does not satisfy a desirable property, namely Dung's fundamental lemma. We therefore propose an alternative definition that induces semantics that behave as expected. We then study computational issues for these new semantics, in particular we show that complexity of reasoning is similar to the complexity of the corresponding decision problems for standard argumentation frameworks in almost all cases. We then propose a translation in pseudo-Boolean constraints for computing (strong and weak) extensions. We conclude with an experimental evaluation of our approach which shows in particular that it scales up well for solving the problem of providing one extension as well as enumerating them all., Comment: This is an extended version of a paper accepted at COMMA 2022. 17 pages, 10 figures

Short-term plasticity (STP) is a mechanism that stores decaying memories in synapses of the cerebral cortex. In computing practice, STP has been used, but mostly in the niche of spiking neurons, even though theory predicts that it is the optimal solution to certain dynamic tasks. Here we present a new type of recurrent neural unit, the STP Neuron (STPN), which indeed turns out strikingly powerful. Its key mechanism is that synapses have a state, propagated through time by a self-recurrent connection-within-the-synapse. This formulation enables training the plasticity with backpropagation through time, resulting in a form of learning to learn and forget in the short term. The STPN outperforms all tested alternatives, i.e. RNNs, LSTMs, other models with fast weights, and differentiable plasticity. We confirm this in both supervised and reinforcement learning (RL), and in tasks such as Associative Retrieval, Maze Exploration, Atari video games, and MuJoCo robotics. Moreover, we calculate that, in neuromorphic or biological circuits, the STPN minimizes energy consumption across models, as it depresses individual synapses dynamically. Based on these, biological STP may have been a strong evolutionary attractor that maximizes both efficiency and computational power. The STPN now brings these neuromorphic advantages also to a broad spectrum of machine learning practice. Code is available at https://github.com/NeuromorphicComputing/stpn, Comment: Accepted at ICML 2022

Biological spiking neural networks (SNNs) can temporally encode information in their outputs, e.g. in the rank order in which neurons fire, whereas artificial neural networks (ANNs) conventionally do not. As a result, models of SNNs for neuromorphic computing are regarded as potentially more rapid and efficient than ANNs when dealing with temporal input. On the other hand, ANNs are simpler to train, and usually achieve superior performance. Here we show that temporal coding such as rank coding (RC) inspired by SNNs can also be applied to conventional ANNs such as LSTMs, and leads to computational savings and speedups. In our RC for ANNs, we apply backpropagation through time using the standard real-valued activations, but only from a strategically early time step of each sequential input example, decided by a threshold-crossing event. Learning then incorporates naturally also *when* to produce an output, without other changes to the model or the algorithm. Both the forward and the backward training pass can be significantly shortened by skipping the remaining input sequence after that first event. RC-training also significantly reduces time-to-insight during inference, with a minimal decrease in accuracy. The desired speed-accuracy trade-off is tunable by varying the threshold or a regularization parameter that rewards output entropy. We demonstrate these in two toy problems of sequence classification, and in a temporally-encoded MNIST dataset where our RC model achieves 99.19% accuracy after the first input time-step, outperforming the state of the art in temporal coding with SNNs, as well as in spoken-word classification of Google Speech Commands, outperforming non-RC-trained early inference with LSTMs., Comment: Spotlight paper at ICLR 2022

In this ISSI-supported series of studies on magnetic helicity in the Sun, we systematically implement different magnetic helicity calculation methods on high-quality solar magnetogram observations. We apply finite-volume, discrete flux tube (in particular, connectivity-based) and flux-integration methods to data from Hinode's Solar Optical Telescope. The target is NOAA active region 10930 during a ~1.5 day interval in December 2006 that included a major eruptive flare (SOL2006-12-13T02:14X3.4). Finite-volume and connectivity-based methods yield instantaneous budgets of the coronal magnetic helicity, while the flux-integration methods allow an estimate of the accumulated helicity injected through the photosphere. The objectives of our work are twofold: A cross-validation of methods, as well as an interpretation of the complex events leading to the eruption. To the first objective, we find (i) strong agreement among the finite-volume methods, (ii) a moderate agreement between the connectivity-based and finite-volume methods, (iii) an excellent agreement between the flux-integration methods, and (iv) an overall agreement between finite-volume and flux-integration based estimates regarding the predominant sign and magnitude of the helicity. To the second objective, we are confident that the photospheric helicity flux significantly contributed to the coronal helicity budget, and that a right-handed structure erupted from a predominantly left-handed corona during the X-class flare. Overall, we find that the use of different methods to estimate the (accumulated) coronal helicity may be necessary in order to draw a complete picture of an active-region corona, given the careful handling of identified data (preparation) issues, which otherwise would mislead the event analysis and interpretation., Comment: Accepted for publication in The Astrophysical Journal, 29 pages, 9 figures

Hebbian plasticity in winner-take-all (WTA) networks is highly attractive for neuromorphic on-chip learning, owing to its efficient, local, unsupervised, and on-line nature. Moreover, its biological plausibility may help overcome important limitations of artificial algorithms, such as their susceptibility to adversarial attacks, and their high demands for training-example quantity and repetition. However, Hebbian WTA learning has found little use in machine learning (ML), likely because it has been missing an optimization theory compatible with deep learning (DL). Here we show rigorously that WTA networks constructed by standard DL elements, combined with a Hebbian-like plasticity that we derive, maintain a Bayesian generative model of the data. Importantly, without any supervision, our algorithm, SoftHebb, minimizes cross-entropy, i.e. a common loss function in supervised DL. We show this theoretically and in practice. The key is a "soft" WTA where there is no absolute "hard" winner neuron. Strikingly, in shallow-network comparisons with backpropagation (BP), SoftHebb shows advantages beyond its Hebbian efficiency. Namely, it converges in fewer iterations, and is significantly more robust to noise and adversarial attacks. Notably, attacks that maximally confuse SoftHebb are also confusing to the human eye, potentially linking human perceptual robustness, with Hebbian WTA circuits of cortex. Finally, SoftHebb can generate synthetic objects as interpolations of real object classes. All in all, Hebbian efficiency, theoretical underpinning, cross-entropy-minimization, and surprising empirical advantages, suggest that SoftHebb may inspire highly neuromorphic and radically different, but practical and advantageous learning algorithms and hardware accelerators.

The PolyOculus technology produces large-area-equivalent telescopes by using fiber optics to link modules of multiple semi-autonomous, small, inexpensive, commercial-off-the-shelf telescopes. Crucially, this scalable design has construction costs which are > 10x lower than equivalent traditional large-area telescopes. We have developed a novel photonic lantern approach for the PolyOculus fiber optic linkages which potentially offers substantial advantages over previously considered free-space optical linkages, including much higher coupling efficiencies. We have carried out the first laboratory tests of a photonic lantern prototype developed for PolyOculus, and demonstrate broadband efficiencies of ~91%, confirming the outstanding performance of this technology., Comment: 13 pages, 8 figures

Neural processing on devices and circuits is fast becoming a popular approach to emulate biological neural networks. Elaborate CMOS and memristive technologies have been employed to achieve this, including chalcogenide-based in-memory computing concepts. Here we show that nano-scaled films of chalcogenide semiconductors can serve as building-blocks for novel types of neural computations where their tunable electronic and optical properties are jointly exploited. We demonstrate that ultrathin photoactive cavities of Ge-doped Selenide can emulate the computationally powerful non-linear operations of three-factor neo-Hebbian plasticity and the shunting inhibition. We apply this property to solve a maze game through reinforcement learning, as well as a single-neuron solution to the XOR, which is a linearly inseparable problem with point-neurons. Our results point to a new breed of memristors with broad implications for neuromorphic computing.

In the mammalian nervous system, various synaptic plasticity rules act, either individually or synergistically, and over wide-ranging timescales to dictate the processes that enable learning and memory formation. To mimic biological cognition for artificial intelligence, neuromorphic computing platforms thus call for synthetic synapses, that can faithfully express such complex plasticity and dynamics. Although some plasticity rules have been emulated with elaborate CMOS and memristive circuitry, hardware demonstrations that combine multiple plasticities, such as long-term (LTP) and short-term plasticity (STP) with tunable dynamics and within the same low-power nanoscale devices have been missing. Here, we introduce phase change memtransistive synapse that leverages the non-volatility of memristors and the volatility of transistors for coupling LTP with homo and heterosynaptic STP effects. We show that such biomimetic synapses can enable some powerful cognitive frameworks, such as the short-term spike-timing-dependent plasticity (ST-STDP) and stochastic Hopfield neural networks. We demonstrate how, much like the mammalian brain, such emulations can establish temporal relationships in data streams for the task of sequential learning and combinatorial optimization.

Context. Magnetic helicity is a physical quantity of great importance in the study of astrophysical and natural plasmas. Although a density for helicity cannot be defined, a good proxy for it is field line helicity. The appropriate quantity for use in solar conditions is relative field line helicity (RFLH). Aims. This work aims to study in detail the behaviour of RFLH, for the first time, in a solar active region (AR). Methods. The target active region is the large, eruptive AR 11158. In order to compute RFLH and all other quantities of interest we use a non-linear force-free reconstruction of the AR coronal magnetic field of excelent quality. Results. We find that the photospheric morphology of RFLH is quite different than that of the magnetic field or of the electrical current, and this is not sensitive to the chosen gauge in the computation of RFLH. The value of helicity experiences a large decrease, 25% of its pre-flare value, during an X-class flare of the AR, a change that is also depicted in the photospheric morphology of RFLH. Moreover, the area of this change coincides with the area that encompasses the flux rope, the magnetic structure that later erupted. Conclusions. The use of RFLH can provide important information about the value and location of the magnetic helicity expelled from the solar atmosphere during eruptive events., Comment: accepted by Astronomy & Astrophysics