Article
Neurosciences
Masataka Konishi, Kei M. Igarashi, Keiji Miura
Summary: Representational learning in the middle layer is crucial for efficient learning in deep neural networks. However, the prevailing backpropagation learning rules are not biologically plausible and cannot be implemented in the brain. To address this, it is critical to establish biologically plausible learning rules for memory tasks. Using numerical simulations, biologically plausible learning rules were developed to replicate a laboratory experiment where mice learned to predict reward amounts.
FRONTIERS IN NEUROSCIENCE
(2023)
Article
Mathematical & Computational Biology
Mufeng Tang, Yibo Yang, Yali Amit
Summary: We develop biologically plausible training mechanisms for self-supervised learning in deep networks. By training with contrastive losses and two alternative methods, our framework achieves comparable performance to standard BP learning in linear classifier evaluation of learned embeddings.
FRONTIERS IN COMPUTATIONAL NEUROSCIENCE
(2022)
Article
Mathematical & Computational Biology
Jimmy Gammell, Sonia Buckley, Sae Woo Nam, Adam N. McCaughan
Summary: The paper proposes a method to alleviate the vanishing gradient problem by replacing some connections in a layered network, inspired by small-world networks. This approach is convenient to implement in neuromorphic hardware and is biologically plausible.
FRONTIERS IN COMPUTATIONAL NEUROSCIENCE
(2021)
Article
Computer Science, Artificial Intelligence
Sergio Verduzco-Flores, William Dorrell, Erik De Schutter
Summary: This paper explores a neural control architecture that is both biologically plausible and capable of autonomous learning. It uses feedback controllers that learn to achieve a desired state by selecting the errors that drive them, employing a family of differential Hebbian learning rules. The architecture can control systems with monotonically or non-monotonically changing error responses through reinforcement learning. The use of feedback control simplifies the learning problem and allows for learning of more complex actions.
Article
Computer Science, Artificial Intelligence
Tielin Zhang, Shuncheng Jia, Xiang Cheng, Bo Xu
Summary: Spiking neural networks (SNNs) are considered the third generation of artificial neural networks (ANNs) and have more biologically realistic structures. To address the dynamic characteristics of SNNs, a biologically plausible reward propagation (BRP) algorithm is proposed for training SNNs, which has shown comparable accuracy to state-of-the-art BP-based SNNs while saving computational cost. This approach provides insights toward a better understanding of the intelligent nature of biological systems.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2022)
Article
Materials Science, Multidisciplinary
Sang Hyun Sung, Yujin Jeong, Jung Won Oh, Hyeon-Jin Shin, Jae Hee Lee, Keon Jae Lee
Summary: This article reviews recent progress in bio-plausible mimicry of neural components using memristive devices, focusing on artificial neurons, synapses, and nerve systems. It also explores the relationship between memristive neuromorphic systems and biological neural networks, highlighting the interdisciplinary approaches in neuronal mapping and brain-computer interfaces.
Article
Chemistry, Multidisciplinary
Dong Gue Roe, Seongchan Kim, Yoon Young Choi, Hwije Woo, Moon Sung Kang, Young Jae Song, Jong-Hyun Ahn, Yoonmyung Lee, Jeong Ho Cho
Summary: Inspired by the human brain's memorization process, an artificial synaptic array was created to replicate Ebbinghaus' forgetting curve, demonstrating a biologically plausible memorization process. By selectively applying repetitive learning, the array mimics the selective attention for information prioritization in the human brain. This advancement in bioinspired electronics shows great potential for future development.
ADVANCED MATERIALS
(2021)
Article
Computer Science, Artificial Intelligence
Mohammad Hassan Tayarani Najaran
Summary: In this paper, various transform functions, including Fourier and Wavelet, were applied to extract features from vibration data for fault diagnosis. Different learning algorithms were trained for each feature extraction approach and their results were aggregated in an ensemble machine learning algorithm. An evolutionary algorithm was introduced to optimize the weights of the base learner algorithms and find the best architecture for Convolutional Neural Networks (CNNs) in fault diagnosis. Experimental results on benchmark problems were presented to evaluate the proposed algorithm.
EXPERT SYSTEMS WITH APPLICATIONS
(2023)
Article
Mathematics, Interdisciplinary Applications
Tae-Hyeon Kim, Sungjoon Kim, Kyungho Hong, Jinwoo Park, Yeongjin Hwang, Byung-Gook Park, Hyungjin Kim
Summary: This study explored the feasibility of achieving a larger number of conductance states using the I-cc control method, with experimental results demonstrating 64-level conductance states. Through off-chip learning of the CNN structure, it was verified that the 64-level states showed recognition performance close to that of software-based neural networks. By reducing information loss in the transfer process, the fabricated synaptic device array with the I-cc control programming method is expected to contribute to the development of hardware neural networks.
CHAOS SOLITONS & FRACTALS
(2021)
Article
Engineering, Electrical & Electronic
Sangyeob Kim, Sangjin Kim, Soyeon Um, Soyeon Kim, Kwantae Kim, Hoi-Jun Yoo
Summary: This paper proposes a highly energy-efficient neuromorphic computing-in-memory (Neuro-CIM) processor for ultralow-power deep learning applications. The Neuro-CIM supports spiking neural network (SNN) and eliminates the power and area overhead of previous CIM processors. It reduces power consumption through various techniques such as sign extended bits gating, replacing high-precision ADC with 1-b comparator, and implementing an early stopping scheme. The processor achieves state-of-the-art energy consumption and accuracy for different classification tasks.
IEEE JOURNAL OF SOLID-STATE CIRCUITS
(2023)
Article
Plant Sciences
Muhammad Hammad Saleem, Kesini Krishnan Velayudhan, Johan Potgieter, Khalid Mahmood Arif
Summary: This research proposes a novel DL-based methodology for the detection and classification of weeds, achieving high accuracy through analysis and optimization of neural networks. The proposed method holds research value for tasks such as real-time detection and reducing computation/training time. The robustness and practicality of the method are validated through the application of the DeepWeeds dataset, making this research an important step towards an efficient and automatic weed control system.
FRONTIERS IN PLANT SCIENCE
(2022)
Article
Acoustics
Thulsiram Gantala, Krishnan Balasubramaniam
Summary: This paper proposes the study of optimizing the hyperparameters of the DPAI model to simulate ultrasonic wave propagation at extended depth with lower error. The effectiveness of the trained DPAI models with varying hyperparameters is demonstrated in reducing the compounding error.
Article
Engineering, Multidisciplinary
Yufeng Shu, Bin Li, Hui Lin
Summary: The surface quality safety inspection of LED chips is essential in production, with traditional methods struggling to keep up with shrinking chip sizes. Deep convolutional neural networks have made significant breakthroughs in this field, surpassing traditional models in performance.
Article
Computer Science, Artificial Intelligence
Takashi Shinozaki
Summary: The study proposes a novel biologically motivated learning method for deep convolutional neural networks, which achieves state-of-the-art performance in image discrimination tasks without requiring a large amount of labeled data. Through unsupervised competitive learning, higher-level learning representations can be achieved solely based on forward propagating signals.
Article
Agronomy
Muhammad Hammad Saleem, Johan Potgieter, Khalid Mahmood Arif
Summary: This research aims to achieve accurate detection of various classes of weeds and a negative class using deep learning, and thoroughly analyze the architectural details of the Faster RCNN model. The results show improved classification and localization performance, validating the effectiveness and robustness of the approach.
Article
Nanoscience & Nanotechnology
L. Canale, A. Laborieux, A. Aroul Mogane, L. Jubin, J. Comtet, A. Leine, L. Bocquet, A. Siria, A. Nigues
Article
Physics, Applied
L. Herrera Diez, Y. T. Liu, D. A. Gilbert, M. Belmeguenai, J. Vogel, S. Pizzini, E. Martinez, A. Lamperti, J. B. Mohammedi, A. Laborieux, Y. Roussigne, A. J. Grutter, E. Arenholtz, P. Quarterman, B. Maranville, S. Ono, M. Salah El Hadri, R. Tolley, E. E. Fullerton, L. Sanchez-Tejerina, A. Stashkevich, S. M. Cherif, A. D. Kent, D. Querlioz, J. Langer, B. Ocker, D. Ravelosona
PHYSICAL REVIEW APPLIED
(2019)
Article
Engineering, Electrical & Electronic
Axel Laborieux, Marc Bocquet, Tifenn Hirtzlin, Jacques-Olivier Klein, Etienne Nowak, Elisa Vianello, Jean-Michel Portal, Damien Querlioz
Summary: The paper focuses on the implementation of ternary neural networks and proposes a two-transistor/two-resistor memory architecture for achieving high energy efficiency at low supply voltage, while studying the bit error rate. Experimental results demonstrate that ternary neural networks can significantly improve neural network performance and exhibit immunity to bit errors.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS I-REGULAR PAPERS
(2021)
Article
Multidisciplinary Sciences
Axel Laborieux, Maxence Ernoult, Tifenn Hirtzlin, Damien Querlioz
Summary: Deep neural networks are susceptible to catastrophic forgetting when learning new tasks. Laborieux et al. propose a training method inspired by neuronal metaplasticity for binarized neural networks to avoid forgetting, which is relevant for neuromorphic applications.
NATURE COMMUNICATIONS
(2021)
Article
Engineering, Electrical & Electronic
Atreya Majumdar, Marc Bocquet, Tifenn Hirtzlin, Axel Laborieux, Jacques-Olivier Klein, Etienne Nowak, Elisa Vianello, Jean-Michel Portal, Damien Querlioz
Summary: This study presents a model of weak RESET process in hafnium oxide RRAM for deep learning and validates its effectiveness through experiments with hybrid CMOS/RRAM technology. The model is used to train binarized neural networks for image recognition tasks, showing that device-to-device variability is the most detrimental imperfection affecting the training process.
IEEE TRANSACTIONS ON ELECTRON DEVICES
(2021)
Proceedings Paper
Computer Science, Artificial Intelligence
A. Laborieux, M. Bocquet, T. Hirtzlin, J-O Klein, L. Herrera Diez, E. Nowak, E. Vianello, J-M Portal, D. Querlioz
2020 2ND IEEE INTERNATIONAL CONFERENCE ON ARTIFICIAL INTELLIGENCE CIRCUITS AND SYSTEMS (AICAS 2020)
(2020)