123 research outputs found
SteelBlastQC Dataset
No full textThe SteelBlastQC dataset consists of 1654 RGB images (512×512 pixels) of steel surfaces that are either shot-blasted or still need shot-blasting to achieve the required texture, forming a binary classification task. The dataset includes 888 “good” (ready for paint) images and 766 “not-good” (needs shot-blasting) images. As declared by the collaborating manufacturer, the ideally treated surface is clean and uniformly coarse with an average roughness level of SA 2.5. The “not-good” class presents several types of defects to the surface, located by industrial shot-blasting experts. These include: fresh welding lines and cuts, abrasion, corrosion, and discoloration. The presented dataset can be used for training computer vision models for automated metal surface quality control, addressing the lack of publicly available datasets containing images of shot-blasted steel. For convenience and reproducibility, the data were split into train and test (80/20 ratio)
Gridbot: towards a neuroinspired navigation system for robot planning
No full textThe ability to orient in an unknown, fast-changing, environment is an unmet challenge for robots but a seamlessly solved problem for the primate brain. This thesis describes the first steps in developing a neuro-inspired “bottom-up” model of the brain’s navigation system to make a mobile robot localize itself, map its surrounding and plan its trajectory. Our model employs neural spikes to encode and process information in real-time. Despite a multitude of Nobel-winning studies that have revealed neurons specializing in self-navigation, such as place, grid, border and head direction cells, their interconnectivity remains elusive. Therefore, any model employing these neurons needs to make quite a lot of extrapolations to fill in the gaps of knowledge. The main challenge was to design a real-time spiking neural network that can compensate for the hardware limitations as well as its own intrinsic imperfections and work in real conditions. To design the first component of our model, the head direction cell layer, we employed mechanisms based on self-organizing and self-sustaining neural activity, or attractor dynamics, resembling those originally proposed in Hebb’s cell assembly theory. The information to be maintained and updated was a continuous variable, or continuous attractor, where a 1D continuum of cell assemblies represented head direction. In theory, our network should give rise to a self-sustained hill of excitation - the attractor. In practice, due to non-ideal speed sensors and the intrinsic spike variability of the spiking network, it was impossible to sustain a correct approximation of the head direction using just this scheme. To correct this, we introduced a spike-based Bayesian inference layer of leaky-integrate-and-fire models of neurons, that combined feedforward (vision) and recursive (kinesthetic) inputs. We show how such a layer can approximate the posterior probability of the preferred state encoded in the spiking probability by adding the logarithms of the simulated dendritic currents, which is a reasonable approximation of the nonlinear dendritic activity. We show that our model accurately estimated head direction and further extend it to include a dynamic network of border cells that can learn to map the observed environment through simulating synaptic plasticity. Solving the localization problem and creating a cognitive map of the surroundings, our thesis paves the way for tackling robot planning through imitating brain structure, its principles and its performance.M.S.Includes bibliographical referencesby Guangzhi Tan
Correction: Corrigendum: Protective effects of ginsenoside Rg1 on intestinal ischemia/reperfusion injury-induced oxidative stress and apoptosis via activation of the Wnt/β-catenin pathway
No full textScientific Reports 6: Article number: 38480; published online: 02 December 2016; updated: 31 May 2017 Guangzhi Wang, Anlong Ji and Xiaofeng Tian were omitted from the author list in the original version of this Article. In addition, an additional affiliation for Guo Zu was omitted. The correct affiliation is listed below:</jats:p
Empirical study on the efficiency of Spiking Neural Networks with axonal delays, and algorithm-hardware benchmarking
Full text linkThe role of axonal synaptic delays in the efficacy and performance of artificial neural networks has been largely unexplored. In step-based analog-valued neural network models (ANNs), the concept is almost absent. In their spiking neuroscience-inspired counterparts, there is hardly a systematic account of their effects on model performance in terms of accuracy and number of synaptic operations. This paper proposes a methodology for accounting for axonal delays in the training loop of deep Spiking Neural Networks (SNNs), intending to efficiently solve machine learning tasks on data with rich temporal dependencies. We then conduct an empirical study of the effects of axonal delays on model performance during inference for the Adding task [1]-[3], a benchmark for sequential regression, and for the Spiking Heidelberg Digits dataset (SHD) [4], commonly used for evaluating event-driven models. Quantitative results on the SHD show that SNNs incorporating axonal delays instead of explicit recurrent synapses achieve state-of-the-art, over 90% test accuracy while needing less than half trainable synapses. Additionally, we estimate the required memory in terms of total parameters and energy consumption of accomodating such delay-trained models on a modern neuromorphic accelerator [5], [6]. These estimations are based on the number of synaptic operations and the reference GF-22nm FDX CMOS technology. As a result, we demonstrate that a reduced parameterization, which incorporates axonal delays, leads to approximately 90% energy and memory reduction in digital hardware implementations for a similar performance in the aforementioned task.</p
ONLINE ANOMALY ANALYSIS AND SELF PROTECTION AGAINST NETWORK ATTACKS by
No full textRights Copyright © is held by the author. Digital access to this material is made possible by the University Libraries, University of Arizona. Further transmission, reproduction or presentation (such as public display or performance) of protected items is prohibited except with permission of the author
Biologically inspired spiking neural networks for energy-efficient robot learning and control
No full textEnergy-efficient learning and control are becoming increasingly crucial for robots that solve complex real-world tasks with limited onboard resources. Although deep neural networks (DNN) have been successfully applied to robotics, their high energy consumption limits their use in low-power edge applications. Biologically inspired spiking neural networks (SNN), facilitated by the advances in neuromorphic processors, have started to deliver energy-efficient, massively parallel, and low-latency solutions to robotics. This dissertation presents our energy-efficient neuromorphic solutions to robot navigation, control, and learning, using SNNs on the neuromorphic processor. First, we propose a biologically constrained SNN, mimicking the brain's spatial system, solving the unidimensional SLAM problem while only consuming 1% of energy compared with the conventional filter-based approach. In addition, when extending the model to 2D environments by adding biologically realistic hippocampal neurons, the SNN formed cognitive maps in real-time and helped study the neuronal interconnectivity and cognitive functions. Next, the dissertation shows how the neuromorphic approach can be extended to high-level cognitive functions such as learning control policies. Specifically, we propose a reinforcement co-learning framework that jointly trains a spiking actor network (SAN) with a deep critic network using backpropagation to learn optimal policies for both mapless navigation and high-dimensional continuous control. Compared with state-of-the-art DNN approaches, our method results in up to 140 times less energy consumption during inference, while generating a superior successful rate on mapless navigation, and achieves the same level of performance on high-dimensional continuous control when using the population-coded spiking actor network (PopSAN). Lastly, we explore how these energy gains can further be extended to training through the development of a biologically plausible gradient-based learning framework on the neuromorphic processor. The learning method is functionally equivalent to the spatiotemporal backpropagation but solely relies on spike-based communication, local information processing, and rapid online computation, which are the main neuromorphic principles that mimic the brain. Overall, work in this dissertation pushes the frontiers of SNN applications to energy-efficient robotic control and learning, and hence paves the way toward the introduction of a biologically inspired alternative solution for autonomous robots running on energy-efficient neuromorphic processors.Ph.D.Includes bibliographical reference
Open the box of digital neuromorphic processor: Towards effective algorithm-hardware co-design
No full textThis work is partially funded by research and innovation projects ANDANTE (ECSEL JU under grant agreement No 876925), DAIS (KDT JU under grant agreement No 101007273) and MemScale (Horizon EU under grant agreement 871371). The JU receives support from the European Union's Horizon 2020 research and innovation programme and Sweden, Spain, Portugal, Belgium, Germany, Slovenia, Czech Republic, Netherlands, Denmark, Norway and Turkey
SENECA: building a fully digital neuromorphic processor, design trade-offs and challenges
Full text link& nbsp;This work was partially funded by research and innovation projects ANDANTE (ECSEL JU under grant agreement No. 876925), DAIS (KDT JU under grant agreement No. 101007273), and MemScale (Horizon EU under grant agreement 871371). The JU receives support from the European Union's Horizon 2020 research and innovation programme and Sweden, Spain, Portugal, Belgium, Germany, Slovenia, Czech Republic, Netherlands, Denmark, Norway, and Turkey
- …
