A5026640054- Advanced Memory and Neural Computing
- Neural dynamics and brain function
- Neuroscience and Neural Engineering
- Ferroelectric and Negative Capacitance Devices
- Neural Networks and Reservoir Computing
- CCD and CMOS Imaging Sensors
- Neurobiology and Insect Physiology Research
- Plant and animal studies
- Insect Pheromone Research and Control
- Modular Robots and Swarm Intelligence
- Neural Networks and Applications
- Insect and Arachnid Ecology and Behavior
- Robotics and Sensor-Based Localization
- Philosophy, Science, and History
- Parallel Computing and Optimization Techniques
- Advanced Optical Imaging Technologies
- Insect Utilization and Effects
Forschungszentrum Jülich
2022-2025
University of Zurich
2018-2024
ETH Zurich
2018-2024
SIB Swiss Institute of Bioinformatics
2018-2024
United States University
2022
Los Alamos National Laboratory
2019-2022
University of Konstanz
2018
Event-based vision sensors achieve up to three orders of magnitude better speed vs. power consumption trade off in high-speed control UAVs compared conventional image sensors. cameras produce a sparse stream events that can be processed more efficiently and with lower latency than images, enabling ultra-fast vision-driven control. Here, we explore how an event-based algorithm implemented as spiking neuronal network on neuromorphic chip used drone controller. We show seamless integration...
In recent years, it has become evident that olfaction is a fast sense, and millisecond short differences in stimulus onsets are used by animals to analyze their olfactory environment. contrast, receptor neurons thought be relatively slow temporally imprecise. These observations have led conundrum: how, then, can an animal resolve dynamics smell with high temporal acuity? Using parallel recordings from Drosophila, we found hitherto unknown precise odorant-evoked spike responses, first...
In this work, we present a spiking neural network (SNN) based PID controller on neuromorphic chip.On-chip SNNs are currently being explored in low-power AI applications.Due to potentially ultra-low power consumption, low latency, and high processing speed, on-chip promising tool for control of power-constrained platforms, such as Unmanned Aerial Vehicles (UAV).To obtain highly efficient fast end-toend controllers, the SNN-based architectures must be seamlessly integrated with motor...
The field of neuromorphic computing holds great promise in terms advancing efficiency and capabilities by following brain-inspired principles. However, the rich diversity techniques employed research has resulted a lack clear standards for benchmarking, hindering effective evaluation advantages strengths methods compared to traditional deep-learning-based methods. This paper presents collaborative effort, bringing together members from academia industry, define benchmarks computing:...
Abstract Programming recurrent spiking neural networks (RSNNs) to robustly perform multi-timescale computation remains a difficult challenge. To address this, we describe single-shot weight learning scheme embed robust dynamics into attractor-based RSNNs, by exploiting the properties of high-dimensional distributed representations. We finite state machines RSNN superimposing symmetric autoassociative matrix and asymmetric transition terms, which are each formed vector binding an input...
In this paper, we investigate the use of ultra low-power, mixed signal analog/digital neuromorphic hardware for implementation biologically inspired neuronal path integration and map formation a mobile robot. We perform spiking network simulations developed architecture, interfaced to simulated robotic vehicle. then port architecture on two connected devices, one which features on-board plasticity, demonstrate feasibility realization simultaneous localization mapping (SLAM).
Spiking Neuronal Networks (SNNs) realized in neuromorphic hardware lead to low-power and low-latency neuronal computing architectures. Neuromorphic systems are most efficient when all of perception, decision making, motor control seamlessly integrated into a single architecture that can be on the hardware. Many network architectures address perception tasks, while work controllers is scarce. Here, we present an improved implementation PID controller. The controller was Intel's research chip...
We present a fully event-driven vision and processing system for selective attention tracking, realized on neuromorphic processor Loihi interfaced to an event-based Dynamic Vision Sensor DAVIS. The mechanism is as recurrent spiking neural network that implements attractor-dynamics of dynamic fields. demonstrate capability the create sustained activation supports object tracking when distractors are or slows down stops, reducing number generated events.
Abstract The capabilities of natural neural systems have inspired new generations machine learning algorithms as well neuromorphic very large-scale integrated (VLSI) circuits capable fast, low-power information processing. However, it has been argued that most modern are not neurophysiologically plausible. In particular, the workhorse deep learning, backpropagation algorithm, proven difficult to translate hardware. this study, we present a neuromorphic, spiking algorithm based on...
In this work, we present a neuromorphic architecture for head pose estimation and scene representation the humanoid iCub robot. The spiking neuronal network is fully realized in Intel's research chip, Loihi, precisely integrates issued motor commands to estimate iCub's path-integration process. vision system of used correct drift estimation. Positions objects front robot are memorized using on-chip synaptic plasticity. We real-time robotic experiments 2 degrees freedom (DoF) robot's show...
The capabilities of natural neural systems have inspired both new generations machine learning algorithms as well neuromorphic, very large-scale integrated circuits capable fast, low-power information processing. However, it has been argued that most modern are not neurophysiologically plausible. In particular, the workhorse deep learning, backpropagation algorithm, proven difficult to translate neuromorphic hardware. This study presents a spiking algorithm based on synfire-gated dynamical...
Neuromorphic hardware offers computing platforms for the efficient implementation of spiking neural networks (SNNs) that can be used robot control. Here, we present such an SNN on a neuromorphic chip solves number tasks related to simultaneous localization and mapping (SLAM): forming map unknown environment and, at same time, estimating robot's pose. In particular, mechanism detect estimate errors when revisits known landmark updates both path integration speed reduce error. The whole system...
Vector Symbolic Architectures (VSA) were first proposed as connectionist models for symbolic reasoning, leveraging parallel and in-memory computing in brains neuromorphic hardware that enable low-power, low-latency applications. Symbols are defined VSAs points/vectors a high-dimensional neural state-space. For spiking (and brains), particularly sparse representations of interest, they minimize the number costly spikes. Furthermore, can be efficiently stored simple Hebbian auto-associative...
We present a spiking neural network (SNN) for visual pattern recognition with on-chip learning on neuromorphichardware. show how this can learn simple patterns composed of horizontal and vertical bars sensed by Dynamic Vision Sensor, using local spike-based plasticity rule. During recognition, the classifies pattern's identity while at same time estimating its location scale. build previous work that used neuromorphic hardware in loop demonstrate proposed properly operate learning,...
The Locally Competitive Algorithm (LCA) [17, 18] was put forward as a model of primary visual cortex [14, 17] and has been used extensively sparse coding algorithm for multivariate data. LCA seen implementations on neuromorphic processors, including IBM's TrueNorth processor [10], Intel's research processor, Loihi, which show that it can be very efficient with respect to the power resources consumes [8]. When combined dictionary learning [13], encounters synaptic instability [24], where,...
We present a fully event-driven vision and processing system for selective attention tracking implemented on Intel's neuromorphic research chip, Loihi, directly interfaced with an event-based Dynamic Vision Sensor, DAVIS. The mechanism is realized as recurrent spiking neural network (SNN) that forms sustained activation-bump attractors. dynamics support object when distractors are the slows down or stops.
Understanding a visual scene by inferring identities and poses of its individual objects is still open problem. Here we propose neuromorphic solution that utilizes an efficient factorization network based on three key concepts: (1) computational framework Vector Symbolic Architectures (VSA) with complex-valued vectors; (2) the design Hierarchical Resonator Networks (HRN) to deal non-commutative nature translation rotation in scenes, when both are used combination; (3) multi-compartment...
Programming recurrent spiking neural networks (RSNNs) to robustly perform multi-timescale computation remains a difficult challenge. To address this, we show how the distributed approach offered by vector symbolic architectures (VSAs), which uses high-dimensional random vectors as smallest units of representation, can be leveraged embed robust dynamics into attractor-based RSNNs. We finite state machines RSNN superimposing symmetric autoassociative weight matrix and asymmetric transition...
Thanks to their parallel and sparse activity features, recurrent neural networks (RNNs) are well-suited for hardware implementation in low-power neuromorphic hardware. However, mapping rate-based RNNs hardware-compatible spiking (SNNs) remains challenging. Here, we present a ${\Sigma}{\Delta}$-low-pass RNN (lpRNN): an architecture employing adaptive neuron model that encodes signals using ${\Sigma}{\Delta}$-modulation enables precise mapping. The ${\Sigma}{\Delta}$-neuron communicates analog...
Instabilities in neuromorphic machine learning can occur when synaptic updates meant to encode matrix transforms are not normalized. This phenomenon is encountered Hebbian [5], where, as a synapse's strength grows, post-synaptic activity increases, further enhancing strength, leading runaway condition, where becomes saturated [3, 7]. A number of mechanisms have been suggested for regulating and stabilizing this [1, 2, 4, 6]. Here, we present new algorithm directly normalize connectivity. The...