Article
Multidisciplinary Sciences
Benjamin Cramer, Sebastian Billaudelle, Simeon Kanya, Aron Leibfried, Andreas Grubl, Vitali Karasenko, Christian Pehle, Korbinian Schreiber, Yannik Stradmann, Johannes Weis, Johannes Schemmel, Friedemann Zenke
Summary: This study demonstrates the applicability of surrogate gradient learning on analog neuromorphic hardware using an in the-loop approach. The results show that learning can self-correct for device mismatch and achieve competitive spiking network performance on vision and speech benchmarks. This work sets several benchmarks for low-energy spiking network processing on analog neuromorphic hardware and paves the way for future on-chip learning algorithms.
PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED STATES OF AMERICA
(2022)
Article
Multidisciplinary Sciences
Mohamed Kentour, Joan Lu
Summary: The study reveals a good balance between transparency and efficiency can be achieved in deep neural networks by exploring Credit Assignment Paths theory. Experimental results on the Twitter Health News dataset show the model is transparent and traceable, with an overall accuracy of approximately 83% and around 94% correct identification of positive sentiments.
Article
Multidisciplinary Sciences
Weier Wan, Rajkumar Kubendran, Clemens Schaefer, Sukru Burc Eryilmaz, Wenqiang Zhang, Dabin Wu, Stephen Deiss, Priyanka Raina, He Qian, Bin Gao, Siddharth Joshi, Huaqiang Wu, H-S Philip Wong, Gert Cauwenberghs
Summary: This study presents NeuRRAM, a RRAM-based CIM chip that offers versatility, high energy efficiency, and accuracy. By co-optimizing algorithms, architecture, circuits, and devices, the chip can be reconfigured for different model architectures and provides twice the energy efficiency of previous state-of-the-art RRAM-CIM chips across various computational bit-precisions. It achieves inference accuracy comparable to software models quantized to four-bit weights across various AI tasks.
Article
Multidisciplinary Sciences
Ali Siddique, Mang I. Vai, Sio Hang Pun
Summary: This article proposes a novel hardware efficient SNN back-propagation scheme that achieves an accuracy of around 97.5% on the MNIST dataset using only 158,800 synapses. It also presents a high-speed, cost-efficient SNN training engine that can operate at a maximum computational frequency of around 50 MHz on a Virtex 6 FPGA.
SCIENTIFIC REPORTS
(2023)
Article
Multidisciplinary Sciences
Te-Yuan Liu, Ata Mahjoubfar, Daniel Prusinski, Luis Stevens
Summary: Neuromorphic computing, which emulates the neural activity of the brain, offers cost and power efficiency advantages. This study applies Intel's Loihi chip to image retrieval and compares its performance with traditional chips. The results demonstrate significant energy efficiency improvements with the neuromorphic solution.
Article
Computer Science, Artificial Intelligence
Nitin Rathi, Kaushik Roy
Summary: This article proposes a low-latency deep spiking network called DIET-SNN, which optimizes the membrane leak and firing threshold to reduce latency while maintaining competitive accuracy. Through evaluation and comparative experiments, DIET-SNN shows excellent performance in image classification tasks with efficient computational capabilities.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Chengzhi Cao, Xueyang Fu, Yurui Zhu, Zhijing Sun, Zheng-Jun Zha
Summary: With the proposed spiking-convolutional network (SC-Net) architecture, the temporal and spatial information in event data is effectively utilized to restore detailed textures and sharp edges in videos. Experimental results on three video restoration tasks demonstrate the effectiveness of the method.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Multidisciplinary Sciences
Changqing Xu, Yi Liu, Dongdong Chen, Yintang Yang
Summary: A novel training method based on backpropagation for ultra-low-latency spiking neural networks has been proposed in this study, which increases the information capacity of each spike using the multi-threshold Leaky Integrate and Fired (LIF) model. Experimental results demonstrate the method achieves high accuracy on MNIST, FashionMNIST, and CIFAR10 datasets, with significant improvement on CIFAR10 compared to previous approaches.
Article
Computer Science, Theory & Methods
Nitin Rathi, Indranil Chakraborty, Adarsh Kosta, Abhronil Sengupta, Aayush Ankit, Priyadarshini Panda, Kaushik Roy
Summary: Neuromorphic Computing is an interdisciplinary field that aims to achieve brain-like efficiency in machine intelligence by optimizing devices, circuits, and algorithms. It introduces Spiking Neural Networks as a new algorithmic paradigm, which represents data as spikes and enables lower power consumption. Challenges lie in developing efficient computing platforms and overcoming the limitations of current technologies.
ACM COMPUTING SURVEYS
(2023)
Article
Multidisciplinary Sciences
Wei Fang, Yanqi Chen, Jianhao Ding, Zhaofei Yu, Timothee Masquelier, Ding Chen, Liwei Huang, Huihui Zhou, Guoqi Li, Yonghong Tian
Summary: Spiking neural networks (SNNs) aim to achieve brain-inspired intelligence on neuromorphic chips with high energy efficiency. However, traditional programming frameworks cannot meet the demands of automatic differentiation, parallel computation acceleration, and deployment. In this study, a framework called SpikingJelly is proposed to address these issues, which can accelerate the training of deep SNNs and provide flexible model acceleration capabilities.
Article
Computer Science, Artificial Intelligence
Xing He, Changgen Peng, Weijie Tan
Summary: In this study, a Wasserstein DLG method named WDLG is proposed to improve the privacy protection of shared gradients in distributed machine learning systems. The method utilizes Wasserstein distance to calculate error loss and enhances model training stability. Experimental results show that WDLG provides more stable virtual data updates, higher attack success rate, faster model convergence, and better image recovery fidelity, as well as support for designing large learning rate strategies.
INTERNATIONAL JOURNAL OF INTELLIGENT SYSTEMS
(2023)
Article
Multidisciplinary Sciences
Logan G. Wright, Tatsuhiro Onodera, Martin M. Stein, Tianyu Wang, Darren T. Schachter, Zoey Hu, Peter L. McMahon
Summary: The study introduced a hybrid in situ-in silico algorithm called physics-aware training, which applies backpropagation to train controllable physical systems for deep physical neural networks. By demonstrating the training of diverse physical neural networks in areas like optics, mechanics, and electronics to perform audio and image classification tasks, the research showcased the universality and effectiveness of the approach.
Article
Neurosciences
Mingcheng Ji, Ziling Wang, Rui Yan, Qingjie Liu, Shu Xu, Huajin Tang
Summary: This paper introduces a novel architecture for event-based object tracking using event cameras and SNN (Spiking Neural Network). By leveraging the unique event-driven computation and energy-efficient computing characteristics of SNN, the architecture better exploits event associations and temporal information, while maintaining sparse representation in segments. A new loss function is proposed to make the architecture more suitable for object tracking. Additionally, a new event-based tracking dataset, named DVSOT21, is presented. Experimental results on DVSOT21 demonstrate competitive performance of the method compared to other competing trackers, with very low energy consumption.
FRONTIERS IN NEUROSCIENCE
(2023)
Article
Computer Science, Artificial Intelligence
Takashi Matsubara, Yuto Miyatake, Takaharu Yaguchi
Summary: The combination of neural networks and numerical integration can provide highly accurate models. The symplectic adjoint method proposed in this study obtains the exact gradient with much less memory cost and is more robust to rounding errors than the naive backpropagation algorithm and checkpointing schemes.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Clarence Tan, Marko Sarlija, Nikola Kasabov
Summary: Emotion recognition is a key challenge in the field of affective computing, and this study proposes a short-term emotion recognition framework based on spiking neural network without handcrafted EEG features. Testing on the DEAP and MAHNOB-HCI databases shows that the SNN-based EEG spiking patterns offer valuable information for short-term emotion recognition.