Ultrasound imaging systems rely on accurate point spread function (PSF) estimation to support advanced image quality enhancement techniques such as deconvolution and speckle reduction. Phase aberration, caused by sound speed inhomogeneity within biological tissue, is inevitable in ultrasound imaging. It distorts the PSF by increasing sidelobe level and introducing asymmetric amplitude, making PSF estimation under phase aberration highly challenging. In this work, we propose a deep learning framework for estimating phase-aberrated PSFs using U-Net and complex U-Net architectures, operating on RF and complex k-space data, respectively, with the latter demonstrating superior performance. Synthetic phase aberration data, generated using the near-field phase screen model, is employed to train the networks. We evaluate various loss functions and find that log-compressed B-mode perceptual loss achieves the best performance, accurately predicting both the mainlobe and near sidelobe regions of the PSF. Simulation results validate the effectiveness of our approach in estimating PSFs under varying levels of phase aberration.
Using OpenCL-based high-level synthesis, we create a number of spiking neural network (SNN) simulators for the Potjans-Diesmann cortical microcircuit for a high-end Field-Programmable Gate Array (FPGA). Our best simulators simulate the circuit 25\% faster than real-time, require less than 21 nJ per synaptic event, and are bottle-necked by the device's on-chip memory. Speed-wise they compare favorably to the state-of-the-art GPU-based simulators and their energy usage is lower than any other published result. This result is the first for simulating the circuit on a single hardware accelerator. We also extensively analyze the techniques and algorithms we implement our simulators with, many of which can be realized on other types of hardware. Thus, this article is of interest to any researcher or practitioner interested in efficient SNN simulation, whether they target FPGAs or not.
Station keeping is an essential maneuver for Autonomous Surface Vehicles (ASVs), mainly when used in confined spaces, to carry out surveys that require the ASV to keep its position or in collaboration with other vehicles where the relative position has an impact over the mission. However, this maneuver can become challenging for classic feedback controllers due to the need for an accurate model of the ASV dynamics and the environmental disturbances. This work proposes a Model Predictive Controller using Neural Network Simulation Error Minimization (NNSEM-MPC) to accurately predict the dynamics of the ASV under wind disturbances. The performance of the proposed scheme under wind disturbances is tested and compared against other controllers in simulation, using the Robotics Operating System (ROS) and the multipurpose simulation environment Gazebo. A set of six tests were conducted by combining two wind speeds (3 m/s and 6 m/s) and three wind directions (0$^\circ$, 90$^\circ$, and 180$^\circ$). The simulation results clearly show the advantage of the NNSEM-MPC over the following methods: backstepping controller, sliding mode controller, simplified dynamics MPC (SD-MPC), neural ordinary differential equation MPC (NODE-MPC), and knowledge-based NODE MPC (KNODE-MPC). The proposed NNSEM-MPC approach performs better than the rest in 4 out of the 6 test conditions, and it is the second best in the 2 remaining test cases, reducing the mean position and heading error by at least 31\% and 46\% respectively across all the test cases. In terms of execution speed, the proposed NNSEM-MPC is at least 36\% faster than the rest of the MPC controllers. The field experiments on two different ASV platforms showed that ASVs can effectively keep the station utilizing the proposed method, with a position error as low as $1.68$ m and a heading error as low as $6.14^{\circ}$ within time windows of at least $150$s.
This paper describes the design and implementation of parallel neural networks (PNNs) with the novel programming language Golang. We follow in our approach the classical Single-Program Multiple-Data (SPMD) model where a PNN is composed of several sequential neural networks, which are trained with a proportional share of the training dataset. We used for this purpose the MNIST dataset, which contains binary images of handwritten digits. Our analysis focusses on different activation functions and optimizations in the form of stochastic gradients and initialization of weights and biases. We conduct a thorough performance analysis, where network configurations and different performance factors are analyzed and interpreted. Golang and its inherent parallelization support proved very well for parallel neural network simulation by considerable decreased processing times compared to sequential variants.
With the increasing development of neuromorphic platforms and their related software tools as well as the increasing scale of spiking neural network (SNN) models, there is a pressure for interoperable and scalable representations of network state. In response to this, we discuss a parallel extension of a widely used format for efficiently representing sparse matrices, the compressed sparse row (CSR), in the context of supporting the simulation and serialization of large-scale SNNs. Sparse matrices for graph adjacency structure provide a natural fit for describing the connectivity of an SNN, and prior work in the area of parallel graph partitioning has developed the distributed CSR (dCSR) format for storing and ingesting large graphs. We contend that organizing additional network information, such as neuron and synapse state, in alignment with its adjacency as dCSR provides a straightforward partition-based distribution of network state. For large-scale simulations, this means each parallel process is only responsible for its own partition of state, which becomes especially useful when the size of an SNN exceeds the memory resources of a single compute node. For potentially long-running simulations, this also enables network serialization to and from disk (e.g. for checkpoint/restart fault-tolerant computing) to be performed largely independently between parallel processes. We also provide a potential implementation, and put it forward for adoption within the neural computing community.
In this work, we study the numerical solution of inverse eigenvalue problems from a machine learning perspective. Two different problems are considered: the inverse Strum-Liouville eigenvalue problem for symmetric potentials and the inverse transmission eigenvalue problem for spherically symmetric refractive indices. Firstly, we solve the corresponding direct problems to produce the required eigenvalues datasets in order to train the machine learning algorithms. Next, we consider several examples of inverse problems and compare the performance of each model to predict the unknown potentials and refractive indices respectively, from a given small set of the lowest eigenvalues. The supervised regression models we use are k-Nearest Neighbours, Random Forests and Multi-Layer Perceptron. Our experiments show that these machine learning methods, under appropriate tuning on their parameters, can numerically solve the examined inverse eigenvalue problems.
Variational quantum algorithms (VQAs) utilize a hybrid quantum-classical architecture to recast problems of high-dimensional linear algebra as ones of stochastic optimization. Despite the promise of leveraging near- to intermediate-term quantum resources to accelerate this task, the computational advantage of VQAs over wholly classical algorithms has not been firmly established. For instance, while the variational quantum eigensolver (VQE) has been developed to approximate low-lying eigenmodes of high-dimensional sparse linear operators, analogous classical optimization algorithms exist in the variational Monte Carlo (VMC) literature, utilizing neural networks in place of quantum circuits to represent quantum states. In this paper we ask if classical stochastic optimization algorithms can be constructed paralleling other VQAs, focusing on the example of the variational quantum linear solver (VQLS). We find that such a construction can be applied to the VQLS, yielding a paradigm that could theoretically extend to other VQAs of similar form.
Physics-informed neural networks (PINNs) have received significant attention as a unified framework for forward, inverse, and surrogate modeling of problems governed by partial differential equations (PDEs). Training PINNs for forward problems, however, pose significant challenges, mainly because of the complex non-convex and multi-objective loss function. In this work, we present a PINN approach to solving the equations of coupled flow and deformation in porous media for both single-phase and multiphase flow. To this end, we construct the solution space using multi-layer neural networks. Due to the dynamics of the problem, we find that incorporating multiple differential relations into the loss function results in an unstable optimization problem, meaning that sometimes it converges to the trivial null solution, other times it moves very far from the expected solution. We report a dimensionless form of the coupled governing equations that we find most favourable to the optimizer. Additionally, we propose a sequential training approach based on the stress-split algorithms of poromechanics. Notably, we find that sequential training based on stress-split performs well for different problems, while the classical strain-split algorithm shows an unstable behaviour similar to what is reported in the context of finite element solvers. We use the approach to solve benchmark problems of poroelasticity, including Mandel's consolidation problem, Barry-Mercer's injection-production problem, and a reference two-phase drainage problem. The Python-SciANN codes reproducing the results reported in this manuscript will be made publicly available at https://github.com/sciann/sciann-applications.
Modern neuroscience employs in silico experimentation on ever-increasing and more detailed neural networks. The high modelling detail goes hand in hand with the need for high model reproducibility, reusability and transparency. Besides, the size of the models and the long timescales under study mandate the use of a simulation system with high computational performance, so as to provide an acceptable time to result. In this work, we present EDEN (Extensible Dynamics Engine for Networks), a new general-purpose, NeuroML-based neural simulator that achieves both high model flexibility and high computational performance, through an innovative model-analysis and code-generation technique. The simulator runs NeuroML v2 models directly, eliminating the need for users to learn yet another simulator-specific, model-specification language. EDEN's functional correctness and computational performance were assessed through NeuroML models available on the NeuroML-DB and Open Source Brain model repositories. In qualitative experiments, the results produced by EDEN were verified against the established NEURON simulator, for a wide range of models. At the same time, computational-performance benchmarks reveal that EDEN runs up to 2 orders-of-magnitude faster than NEURON on a typical desktop computer, and does so without additional effort from the user. Finally, and without added user effort, EDEN has been built from scratch to scale seamlessly over multiple CPUs and across computer clusters, when available.
This paper reports a comprehensive study on the impacts of temperature-change, process variation, flicker noise and device aging on the inference accuracy of pre-trained all-ferroelectric (FE) FinFET deep neural networks. Multiple-level-cell (MLC) operation with a novel adaptive-program-and-read algorithm with 100ns write pulse has been experimentally demonstrated in 5 nm thick hafnium zirconium oxide (HZO)-based FE-FinFET. With pre-trained neural network (NN) with 97.5% inference accuracy on MNIST dataset as baseline, device to device variation is shown to have negligible impact. Flicker noise characterization at various bias conditions depicts that drain current fluctuation is less than 0.7% with virtually no inference accuracy degradation. The conductance drift of a programmed cell, as an aftermath of temperature change, was captured by a compact model over a wide range of gate biases. Despite significant inference accuracy degradation at 233K for a NN trained at 300K, gate bias optimization for recovering the accuracy is demonstrated. Endurance above 10$^8$ cycles and extrapolated retention above 10 years are shown, which paves the way for edge device artificial intelligence with FE-FinFETs.
Recently, the reconfigurable intelligent surface (RIS), benefited from the breakthrough on the fabrication of programmable meta-material, has been speculated as one of the key enabling technologies for the future six generation (6G) wireless communication systems scaled up beyond massive multiple input multiple output (Massive-MIMO) technology to achieve smart radio environments. Employed as reflecting arrays, RIS is able to assist MIMO transmissions without the need of radio frequency chains resulting in considerable reduction in power consumption. In this paper, we investigate the joint design of transmit beamforming matrix at the base station and the phase shift matrix at the RIS, by leveraging recent advances in deep reinforcement learning (DRL). We first develop a DRL based algorithm, in which the joint design is obtained through trial-and-error interactions with the environment by observing predefined rewards, in the context of continuous state and action. Unlike the most reported works utilizing the alternating optimization techniques to alternatively obtain the transmit beamforming and phase shifts, the proposed DRL based algorithm obtains the joint design simultaneously as the output of the DRL neural network. Simulation results show that the proposed algorithm is not only able to learn from the environment and gradually improve its behavior, but also obtains the comparable performance compared with two state-of-the-art benchmarks. It is also observed that, appropriate neural network parameter settings will improve significantly the performance and convergence rate of the proposed algorithm.
We introduce Arbor, a performance portable library for simulation of large networks of multi-compartment neurons on HPC systems. Arbor is open source software, developed under the auspices of the HBP. The performance portability is by virtue of back-end specific optimizations for x86 multicore, Intel KNL, and NVIDIA GPUs. When coupled with low memory overheads, these optimizations make Arbor an order of magnitude faster than the most widely-used comparable simulation software. The single-node performance can be scaled out to run very large models at extreme scale with efficient weak scaling. HPC, GPU, neuroscience, neuron, software
Monte Carlo Tree Search (MCTS) algorithms perform simulation-based search to improve policies online. During search, the simulation policy is adapted to explore the most promising lines of play. MCTS has been used by state-of-the-art programs for many problems, however a disadvantage to MCTS is that it estimates the values of states with Monte Carlo averages, stored in a search tree; this does not scale to games with very high branching factors. We propose an alternative simulation-based search method, Policy Gradient Search (PGS), which adapts a neural network simulation policy online via policy gradient updates, avoiding the need for a search tree. In Hex, PGS achieves comparable performance to MCTS, and an agent trained using Expert Iteration with PGS was able defeat MoHex 2.0, the strongest open-source Hex agent, in 9x9 Hex.
In this paper, we propose a learning-based low-overhead beam alignment method for vehicle-to-infrastructure communication in vehicular networks. The main idea is to remotely infer the optimal beam directions at a target base station in future time slots, based on the CSI of a source base station in previous time slots. The proposed scheme can reduce channel acquisition and beam training overhead by replacing pilot-aided beam training with online inference from a sequence-to-sequence neural network. Simulation results based on ray-tracing channel data show that our proposed scheme achieves a $8.86\%$ improvement over location-based beamforming schemes with a positioning error of $1$m, and is within a $4.93\%$ performance loss compared with the genie-aided optimal beamformer.
Recently, RRAM-based Binary Neural Network (BNN) hardware has been gaining interests as it requires 1-bit sense-amp only and eliminates the need for high-resolution ADC and DAC. However, RRAM-based BNN hardware still requires high-resolution ADC for partial sum calculation to implement large-scale neural network using multiple memory arrays. We propose a neural network-hardware co-design approach to split input to fit each split network on a RRAM array so that the reconstructed BNNs calculate 1-bit output neuron in each array. As a result, ADC can be completely eliminated from the design even for large-scale neural network. Simulation results show that the proposed network reconstruction and retraining recovers the inference accuracy of the original BNN. The accuracy loss of the proposed scheme in the CIFAR-10 testcase was less than 1.1% compared to the original network. The code for training and running proposed BNN models is available at: https://github.com/YulhwaKim/RRAMScalable_BNN.
We present the N2Sky system, which provides a framework for the exchange of neural network specific knowledge, as neural network paradigms and objects, by a virtual organization environment. It follows the sky computing paradigm delivering ample resources by the usage of federated Clouds. N2Sky is a novel Cloud-based neural network simulation environment, which follows a pure service oriented approach. The system implements a transparent environment aiming to enable both novice and experienced users to do neural network research easily and comfortably. N2Sky is built using the RAVO reference architecture of virtual organizations which allows itself naturally integrating into the Cloud service stack (SaaS, PaaS, and IaaS) of service oriented architectures.
This paper presents a constructive algorithm that achieves successful one-shot learning of hidden spike-patterns in a competitive detection task. It has previously been shown (Masquelier et al., 2008) that spike-timing-dependent plasticity (STDP) and lateral inhibition can result in neurons competitively tuned to repeating spike-patterns concealed in high rates of overall presynaptic activity. One-shot construction of neurons with synapse weights calculated as estimates of converged STDP outcomes results in immediate selective detection of hidden spike-patterns. The capability of continual learning is demonstrated through the successful one-shot detection of new sets of spike-patterns introduced after long intervals in the simulation time. Simulation expansion (Lightheart et al., 2013) has been proposed as an approach to the development of constructive algorithms that are compatible with simulations of biological neural networks. A simulation of a biological neural network may have orders of magnitude fewer neurons and connections than the related biological neural systems; therefore, simulated neural networks can be assumed to be a subset of a larger neural system. The constructive algorithm is developed using simulation expansion concepts to perform an operation equivalent to the exchange of neurons between the simulation and the larger hypothetical neural system. The dynamic selection of neurons to simulate within a larger neural system (hypothetical or stored in memory) may be a starting point for a wide range of developments and applications in machine learning and the simulation of biology.
Current connectionist simulations require huge computational resources. We describe a neural network simulator for the IBM GF11, an experimental SIMD machine with 566 processors and a peak arithmetic performance of 11 Gigaflops. We present our parallel implementation of the backpropagation learning algorithm, techniques for increasing efficiency, performance measurements on the NetTalk text-to-speech benchmark, and a performance model for the simulator. Our simulator currently runs the back-propagation learning algorithm at 900 million connections per second, where each "connection per second" includes both a forward and backward pass. This figure was obtained on the machine when only 356 processors were working; with all 566 processors operational, our simulation will run at over one billion connections per second. We conclude that the GF11 is well-suited to neural network simulation, and we analyze our use of the machine to determine which features are the most important for high performance.
We measured the impact of long-range exponentially decaying intra-areal lateral connectivity on the scaling and memory occupation of a distributed spiking neural network simulator compared to that of short-range Gaussian decays. While previous studies adopted short-range connectivity, recent experimental neurosciences studies are pointing out the role of longer-range intra-areal connectivity with implications on neural simulation platforms. Two-dimensional grids of cortical columns composed by up to 11 M point-like spiking neurons with spike frequency adaption were connected by up to 30 G synapses using short- and long-range connectivity models. The MPI processes composing the distributed simulator were run on up to 1024 hardware cores, hosted on a 64 nodes server platform. The hardware platform was a cluster of IBM NX360 M5 16-core compute nodes, each one containing two Intel Xeon Haswell 8-core E5-2630 v3 processors, with a clock of 2.40 G Hz, interconnected through an InfiniBand network, equipped with 4x QDR switches.
The development of spiking neural network simulation software is a critical component enabling the modeling of neural systems and the development of biologically inspired algorithms. Existing software frameworks support a wide range of neural functionality, software abstraction levels, and hardware devices, yet are typically not suitable for rapid prototyping or application to problems in the domain of machine learning. In this paper, we describe a new Python package for the simulation of spiking neural networks, specifically geared towards machine learning and reinforcement learning. Our software, called BindsNET, enables rapid building and simulation of spiking networks and features user-friendly, concise syntax. BindsNET is built on top of the PyTorch deep neural networks library, enabling fast CPU and GPU computation for large spiking networks. The BindsNET framework can be adjusted to meet the needs of other existing computing and hardware environments, e.g., TensorFlow. We also provide an interface into the OpenAI gym library, allowing for training and evaluation of spiking networks on reinforcement learning problems. We argue that this package facilitates the use of spiking networks for large-scale machine learning experimentation, and show some simple examples of how we envision BindsNET can be used in practice. BindsNET code is available at https://github.com/Hananel-Hazan/bindsnet