Article
Engineering, Electrical & Electronic
Elena Adomaitiene, Steponas Asmontas, Skaidra Bumeliene, Arunas Tamasevicius
Summary: The I&F electronic neuron model generates short spikes, can be stabilized by feedback, inhibits spikes with high frequency action, and can synchronize when coupled units.
AEU-INTERNATIONAL JOURNAL OF ELECTRONICS AND COMMUNICATIONS
(2021)
Article
Engineering, Electrical & Electronic
Jeeson Kim, Vladmir Kornijcuk, Changmin Ye, Doo Seok Jeong
Summary: This study presents a method to emulate a leaky integrate-and-fire (LIF) model in a hardware-efficient manner using a simplified spike-response model (SRM0) and the template-scaling-based exponential function approximation (TS-EFA). The implementation in FPGA of 512 neurons conforming to SRM0 showcases high precision, low latency, and efficient hardware usage.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS I-REGULAR PAPERS
(2021)
Article
Computer Science, Artificial Intelligence
Tomasz Gorski, Damien Depannemaecker, Alain Destexhe
Summary: The study highlights the limitations of the adaptive exponential integrate-and-fire model and proposes the conductance-based adaptive exponential integrate-and-fire model as a solution to avoid unrealistic behaviors, demonstrating its dynamic characteristics and the variety of firing patterns it can produce.
NEURAL COMPUTATION
(2021)
Article
Engineering, Electrical & Electronic
Ashvinikumar Dongre, Gaurav Trivedi
Summary: This article introduces a self-resetting Integrate and Fire (I&F) neuron circuit based on Resistive Random Access Memory (RRAM). The circuit does not require any external bias voltage, and the integration of control unit into the neuron circuit optimizes its overall power consumption. The circuit consists of two RRAMs for integrate and fire operations, and a pulse propagation and reset circuit with 22 CMOS transistors. It consumes 1.5 fJ per spike, which is 48% and 53% less than recent neurons designed using nanoscale FBFET and PDSOI-MOSFET, respectively. The proposed neuron operates at frequencies ranging from 277 KHz to 03 MHz, at least 7.5% and 10% higher than the operating frequencies of the mentioned recent neurons, respectively. The inclusion of a reset circuit enables the implementation of large scale spiking neural networks (SNN), making it superior in terms of power and energy consumption.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS II-EXPRESS BRIEFS
(2023)
Article
Neurosciences
Sijia Lu, Feng Xu
Summary: This paper establishes a precise mathematical mapping between the biological parameters of the LIF/SNNs and the parameters of ReLU-AN/DNNs, which is validated through simulation and experiments.
FRONTIERS IN NEUROSCIENCE
(2022)
Article
Construction & Building Technology
Xiaobo Shen, Zhaoyang Cao, Haifeng Liu, Beihua Cong, Feng Zhou, Yunsheng Ma, Xiong Zou, Shengke Wei
Summary: In this study, CFD simulations and deep learning were combined to develop a more efficient and intelligent tool for fire investigation. A CFD model for a single room was built and abundant simulations were conducted to collect temperature distribution and smoke layer height data. The dataset was divided into two parts for training and validation of the neural network, and both the inverse and forward models showed high accuracy, with the inverse model reaching over 99% accuracy. The robustness of the inverse model was also examined and validated.
JOURNAL OF BUILDING ENGINEERING
(2023)
Article
Computer Science, Artificial Intelligence
Effrosyni Doutsi, Lionel Fillatre, Marc Antonini, Panagiotis Tsakalides
Summary: This paper introduces a novel coding/decoding mechanism named Dual-SIM quantizer (Dual-SIMQ) that simulates one of the most important properties of the human visual system to enhance the quality of visual perception. By utilizing neuroscience models, the Dual-SIMQ combines time-SIM and rate-SIM mechanisms to achieve high-quality neural coding and simple decoding, ultimately improving the reconstruction quality and performance of the visual stimulus. The proposed mechanism shows promising results in controlling reconstruction accuracy, numerical comparison with state-of-the-art methods, and enhancing perceptual reconstruction quality.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2021)
Article
Engineering, Electrical & Electronic
Wesley H. Brigner, Naimul Hassan, Xuan Hu, Christopher H. Bennett, Felipe Garcia-Sanchez, Can Cui, Alvaro Velasquez, Matthew J. Marinella, Jean Anne C. Incorvia, Joseph S. Friedman
Summary: CMOS devices are not suitable for analog applications, while spintronic devices are well suited. However, a large number of spintronic devices still require the use of CMOS, which decreases system efficiency. This study improves the activation functions of spintronic devices to enable better neural network learning and recognition.
IEEE TRANSACTIONS ON ELECTRON DEVICES
(2022)
Article
Engineering, Electrical & Electronic
Namita Bindal, Ravish Kumar Raj, Brajesh Kumar Kaushik
Summary: Antiferromagnetic (AFM) skyrmion technology is considered a promising direction for the development of the next-generation spintronics-based neuromorphic computing due to its unique features. In this study, a neuron device based on AFM skyrmion is proposed, which demonstrates integrate fire functionality and generates an output signal on a nanotrack. The proposed device offers a 68% reduction in energy dissipation when the number of skyrmions on the nanotrack is increased from 1 to 3.
IEEE TRANSACTIONS ON ELECTRON DEVICES
(2023)
Article
Physics, Multidisciplinary
Viktoras Pyragas, Kestutis Pyragas
Summary: An exact low-dimensional system of mean-field equations for an infinite-size network of pulse-coupled integrate-and-fire neurons with a bimodal distribution of excitability parameter has been derived and studied. Bifurcation analysis reveals a variety of dynamic modes, such as multistable equilibrium states, collective oscillations, and chaos, which do not exist with a unimodal distribution of the excitability parameter. The coexistence of oscillatory modes with stable equilibrium states is also observed.
Article
Optics
Krzysztof Tyszka, Magdalena Furman, Rafal Mirek, Mateusz Krol, Andrzej Opala, Bartlomiej Seredynski, Jan Suffczynski, Wojciech Pacuski, Michal Matuszewski, Jacek Szczytko, Barbara Pietka
Summary: This paper introduces a new approach to neuromorphic photonics using microcavities as building blocks of optical spiking neurons, and demonstrates its feasibility through experimental results. The research shows that exciton-polaritons exhibit characteristics similar to the Leaky Integrate-and-Fire (LIF) mechanism, enabling ultrafast processing on sub-ns timescales.
LASER & PHOTONICS REVIEWS
(2023)
Article
Physics, Multidisciplinary
JunHyuk Woo, Soon Ho Kim, Kyungreem Han, MooYoung Choi
Summary: The integrate-and-fire model is widely used in neuromorphic computing and artificial neural network algorithms. This study characterizes the dynamics and information processing of IF models through computer simulations and information-theoretic approaches, showing the importance of neural coding efficiency and dynamics in neural networks.
JOURNAL OF PHYSICS A-MATHEMATICAL AND THEORETICAL
(2021)
Article
Economics
Vanessa de Souza Gomes, Cassio Augusto Ussi Monti, Carolina Souza Jarochinski e Silva, Lucas Rezende Gomide
Summary: The study focused on evaluating the impact of stochastic delays of forest road maintenance on forest harvesting through a stochastic programming model and simulations. It was found that delays significantly affected the harvesting plan and timber volume, highlighting the importance of effective road maintenance scheduling for better forest management practices.
FOREST POLICY AND ECONOMICS
(2021)
Article
Automation & Control Systems
Hao Yu, Jun Shang, Tongwen Chen
Summary: This paper investigates the discrete-time linear-quadratic-Gaussian problem in the context of partial-information-based sensors and event-based transmissions. By constructing a specific form of stochastic event-based protocols, the optimal control policy is derived, which minimizes an average quadratic cost function under a fixed transmission protocol. Analytical expressions of the corresponding transmission and control performance are obtained. The concept of strict consistency is introduced to evaluate the efficiency of event-based protocols and an iterative parameter design algorithm is proposed to further improve the control performance.
Article
Engineering, Electrical & Electronic
Mudasir A. Khanday, Faisal Bashir, Farooq A. Khanday
Summary: In this work, a single transistor based on germanium (Ge) is used to construct a leaky integrate-and-fire (LIF) neuron, providing significant improvements in energy efficiency, area efficiency, and reduction in cost. Through 2-D calibrated simulation, it is validated that the Ge-mosfet LIF neuron accurately imitates the behavior of a neuron. The Ge-mosfet exhibits low breakdown voltage, high impact ionization coefficient, and sharp breakdown, contributing to low energy per spike and higher spiking current. Compared to a recently reported silicon-based silicon-on-insulator (SOI) mosfet, the proposed Ge-mosfet LIF neuron requires only 8 pJ/spike of energy. The use of gate voltage allows for controllable firing of the Ge-mosfet LIF neuron, improving the energy efficiency of the spiking neural network (SNN) by inducing sparse action.
IEEE TRANSACTIONS ON ELECTRON DEVICES
(2022)