在尖端的神经形态硬件上实现的反向传播算法
Alpha Renner1,2, Forrest Sheldon3,4, Anatoly Zlotnik5
1Institute of Neuroinformatics, University of Zurich and ETH Zurich, Zurich, 8057, Switzerland.
Nature communications
|November 8, 2024
概括
研究人员为神经形态硬件开发了一种新的尖端反向传播算法. 这种芯片上实现实现了机器学习任务的竞争性准确性,为高效的边缘计算应用铺平了道路.
科学领域:
- 神经科学和人工智能 人工智能
- 神经形态工程的神经形态工程
- 机器学习 机器学习
背景情况:
- 自然的神经系统激发了先进的机器学习和神经形态电路.
- 现代深度学习,特别是反向传播,在神经生理学可信性和硬件实现方面面临挑战.
- 现有的神经形态方法经常难以复制确切的反向传播算法.
研究的目的:
- 在英特尔的Loihi研究处理器上实现了一个神经形态的,尖的反向传播算法.
- 为了证明一条能够进行芯片内学习的三层电路的原理证明.
- 展示完全在芯片上的尖端神经网络 (SNN) 中精确反向传播的可行性.
主要方法:
- 实现一个synfire-gated的动态信息协调和处理算法.
- 在英特尔的Loihi神经形态研究处理器上部署.
- 在MNIST和时尚MNIST数据集上进行培训和测试,用于数字和服装项目分类.
主要成果:
- 成功证明了三层尖端神经网络学习任务的原理证明.
- 获得了与离芯片训练的SNNs相匹敌的分类准确性.
- 展示了一种适用于边缘计算应用的能量延迟产品.
结论:
- 这项工作代表了SNN中第一个完全在芯片上,计算机在循环中的精确反向传播算法的首次完全实现.
- 开发的方法可以在神经形态处理器上实现低功耗,低延迟的深度学习应用程序.
- 突出了将高级机器学习与内存,大规模并行神经形态硬件集成的可行途径.
相关概念视频
01:19The Role of Ion Channels in Neuronal Computation
3.1K
A postsynaptic neuron usually receives numerous impulses from several other presynaptic neurons. The axon hillock of the postsynaptic neuron integrates all these signals and determines the likelihood of firing an action potential.
Sometimes a single EPSP is strong enough to induce an action potential in the postsynaptic neuron. However, multiple presynaptic inputs must often create EPSPs around the same time for the postsynaptic neuron to be sufficiently depolarized to fire an action potential....
Sometimes a single EPSP is strong enough to induce an action potential in the postsynaptic neuron. However, multiple presynaptic inputs must often create EPSPs around the same time for the postsynaptic neuron to be sufficiently depolarized to fire an action potential....
3.1K
01:28Integration of Synaptic Events
1.4K
Synaptic integration mainly includes the summation of graded potentials. Graded potentials, regardless of their type, cause subtle alterations in membrane voltage, resulting in either depolarization or hyperpolarization. These incremental changes, when combined or summed, can propel the neuron toward its threshold. Consider, for example, a membrane experiencing a +15 mV shift, causing it to depolarize from -70 mV to -55 mV. In this scenario, graded potentials govern the membrane's ability...
1.4K
01:25Neural Circuits
1.1K
Neural circuits and neuronal pools are two of the main structures found in the nervous system. Neural circuits are networks of neurons that work together to carry out a specific task or process. They consist of interconnected neurons and glial cells, which provide structural and metabolic support.
Neuronal pools are collections of nerve cells with similar functions and interact through chemical and electrical signals. These pools include both interneurons (the central neural circuit nodes that...
Neuronal pools are collections of nerve cells with similar functions and interact through chemical and electrical signals. These pools include both interneurons (the central neural circuit nodes that...
1.1K