Search

Significance: Time-domain functional near-infrared spectroscopy (TD-fNIRS) has been considered as the gold standard of noninvasive optical brain imaging devices. However, due to the high cost, complexity, and large form factor, it has not been as widely adopted as continuous wave NIRS systems. Aim: Kernel Flow is a TD-fNIRS system that has been designed to break through these limitations by maintaining the performance of a research grade TD-fNIRS system while integrating all of the components into a small modular device. Approach: The Kernel Flow modules are built around miniaturized laser drivers, custom integrated circuits, and specialized detectors. The modules can be assembled into a system with dense channel coverage over the entire head. Results: We show performance similar to benchtop systems with our miniaturized device as characterized by standardized tissue and optical phantom protocols for TD-fNIRS and human neuroscience results. Conclusions: The miniaturized design of the Kernel Flow system allows for broader applications of TD-fNIRS.

Across many domains of interaction, both natural and artificial, individuals use past experience to shape future behaviors. The results of such learning processes depend on what individuals wish to maximize. A natural objective is one's own success. However, when two such "selfish" learners interact with each other, the outcome can be detrimental to both, especially when there are conflicts of interest. Here, we explore how a learner can align incentives with a selfish opponent. Moreover, we consider the dynamics that arise when learning rules themselves are subject to evolutionary pressure. By combining extensive simulations and analytical techniques, we demonstrate that selfish learning is unstable in most classical two-player repeated games. If evolution operates on the level of long-run payoffs, selection instead favors learning rules that incorporate social (other-regarding) preferences. To further corroborate these results, we analyze data from a repeated prisoner's dilemma experiment. We find that selfish learning is insufficient to explain human behavior when there is a trade-off between payoff maximization and fairness., 73 pages; final version

MEG based on optically-pumped magnetometry (OP-MEG) operates with miniaturized, wearable insulation, in contrast to massive cryogenic dewars for SQUID-MEG, and allows placement of the sensors close to the scalp. This allows more natural head motion during data recording and localized signal quality comparable to, or surpassing, SQUID-MEG. However no OP-MEG system to date has offered full-head coverage with dense sensor packing, and existing systems - as with SQUID-MEG - require the subject to be sealed in a multilayer, passively-shielded vault in order to suppress ambient magnetic fields. Here we present Kernel Flux, which overcomes these limitations. Kernel Flux uses a collection of alkali vapor sensors in a unique array architecture to directly detect the magnetic fields generated by collective neural activity in the brain, while allowing for comfortable head motion. Each Kernel Flux OP-MEG system was designed from the ground up to work as an integrated system optimized around the user's experience, with relevance to natural home and office contexts.

We have trained a fully convolutional spatio-temporal model for fast and accurate representation learning in the challenging exemplar application area of fusion energy plasma science. The onset of major disruptions is a critically important fusion energy science (FES) issue that must be resolved for advanced tokamak. While a variety of statistical methods have been used to address the problem of tokamak disruption prediction and control, recent approaches based on deep learning have proven particularly compelling. In the present paper, we introduce further improvements to the fusion recurrent neural network (FRNN) software suite. Up to now, FRNN was based on the long short-term memory (LSTM) variant of recurrent neural networks to leverage the temporal information in the data. Here, we implement and apply the temporal convolutional neural network (TCN) architecture to the time-dependent input signals, thus rendering the FRNN architecture fully convolutional. This allows highly optimized convolution operations to carry the majority of the computational load of training, thus enabling a reduction in training time, and the effective use of high performance computing (HPC) resources for hyperparameter tuning. At the same time, the TCN based architecture achieves equal or better predictive performance when compared with the LSTM architecture for a large, representative fusion database. Across data-rich scientific disciplines, these results have implications for the resource-effective training of general spatio-temporal feature extractors based on deep learning. Moreover, this challenging exemplar case study illustrates the advantages of a predictive platform with flexible architecture selection options capable of being readily tuned and adapted for responding to prediction needs that increasingly arise in large modern observational dataset.

In this paper, we evaluate training of deep recurrent neural networks with half-precision floats. We implement a distributed, data-parallel, synchronous training algorithm by integrating TensorFlow and CUDA-aware MPI to enable execution across multiple GPU nodes and making use of high-speed interconnects. We introduce a learning rate schedule facilitating neural network convergence at up to $O(100)$ workers. Strong scaling tests performed on clusters of NVIDIA Pascal P100 GPUs show linear runtime and logarithmic communication time scaling for both single and mixed precision training modes. Performance is evaluated on a scientific dataset taken from the Joint European Torus (JET) tokamak, containing multi-modal time series of sensory measurements leading up to deleterious events called plasma disruptions, and the benchmark Large Movie Review Dataset~\cite{imdb}. Half-precision significantly reduces memory and network bandwidth, allowing training of state-of-the-art models with over 70 million trainable parameters while achieving a comparable test set performance as single precision.

Nuclear fusion power delivered by magnetic-confinement tokamak reactors holds the promise of sustainable and clean energy1. The avoidance of large-scale plasma instabilities called disruptions within these reactors2,3 is one of the most pressing challenges4,5, because disruptions can halt power production and damage key components. Disruptions are particularly harmful for large burning-plasma systems such as the multibillion-dollar International Thermonuclear Experimental Reactor (ITER) project6 currently under construction, which aims to be the first reactor that produces more power from fusion than is injected to heat the plasma. Here we present a method based on deep learning for forecasting disruptions. Our method extends considerably the capabilities of previous strategies such as first-principles-based5 and classical machine-learning7–11 approaches. In particular, it delivers reliable predictions for machines other than the one on which it was trained—a crucial requirement for future large reactors that cannot afford training disruptions. Our approach takes advantage of high-dimensional training data to boost predictive performance while also engaging supercomputing resources at the largest scale to improve accuracy and speed. Trained on experimental data from the largest tokamaks in the United States (DIII-D12) and the world (Joint European Torus, JET13), our method can also be applied to specific tasks such as prediction with long warning times: this opens up the possibility of moving from passive disruption prediction to active reactor control and optimization. These initial results illustrate the potential for deep learning to accelerate progress in fusion-energy science and, more generally, in the understanding and prediction of complex physical systems. Using data from plasma-based tokamak nuclear reactors in the US and Europe, a machine-learning approach based on deep neural networks is taught to forecast disruptions, even those in machines on which the algorithm was not trained.

We extend the simplex-in-cell (SIC) technique recently introduced in the context of collisionless dark matter fluids (Abel et al. 2012; Hahn et al. 2012) to the case of collisionless plasmas. The six-dimensional phase space distribution function $f(\mathbf x,\mathbf v)$ is represented by an ensemble of three-dimensional manifolds, which we refer to as sheets. The electric potential field is obtained by solving the Poisson equation on a uniform mesh, where the charge density is evaluated by a spatial projection of the phase space sheets. The SIC representation of phase space density facilitates robust, high accuracy numerical evolution of the Vlasov-Poisson system using significantly fewer tracer particles than comparable particle-in-cell (PIC) approaches by reducing the numerical shot-noise associated with the latter. We introduce the SIC formulation and describe its implementation in a new code, which we validate using standard test problems including plasma oscillations, Landau damping, and two stream instabilities in one dimension. Merits of the new scheme are shown to include higher accuracy and faster convergence rates in the number of particles. We finally motivate and outline the efficient application of SIC to higher dimensional problems., 22 pages, 9 figures

Inspired by RecA-protein-based homology recognition, we consider the pairing of two long linear arrays of binding sites. We propose a fully reversible, physically realizable biased random walk model for rapid and accurate self-assembly due to the spontaneous pairing of matching binding sites, where the statistics of the searched sample are included. In the model, there are two bound conformations, and the free energy for each conformation is a weakly nonlinear function of the number of contiguous matched bound sites.

It is well known that during homology recognition and strand exchange the double stranded DNA (dsDNA) in DNA/RecA filaments is highly extended, but the functional role of the extension has been unclear. We present an analytical model that calculates the distribution of tension in the extended dsDNA during strand exchange. The model suggests that the binding of additional dsDNA base pairs to the DNA/RecA filament alters the tension in dsDNA that was already bound to the filament, resulting in a non-linear increase in the mechanical energy as a function of the number of bound base pairs. This collective mechanical response may promote homology stringency and underlie unexplained experimental results.

e17520 Background: 25% of cancer drugs currently under development are intended for oral self-medication. This reflects the cancer patients' preferences, 54% - 89% of whom --- depending on the demo...