为卷积神经网络的光子神经形加速器基于一个集成的可重新配置的网格
Aris Tsirigotis1, George Sarantoglou1,2, Stavros Deligiannidis3
1Department of Information and Communication Systems Engineering, University of the Aegean, Karlovasi, Samos, Greece.
Communications engineering
|April 28, 2025
概括
本研究介绍了一种使用光学光谱切片的光子集成的神经形态加速器. 它在MNIST数据集上实现了高精度,为机器学习应用程序提供了显著的节能.
科学领域:
- 光子学是指光子学的使用方法.
- 神经形态计算是一种神经形态计算.
- 机器学习 机器学习
背景情况:
- 光子加速器为机器学习提供能源效率和低延迟.
- 升级集成光子电路用于复杂的任务,如卷积层仍然是一个挑战.
研究的目的:
- 为了实验验证一个光子集成的神经形态加速器.
- 为了证明光谱切片对机器学习任务的有效性.
- 展示一个硬件友好的方法,以实现高效的计算.
主要方法:
- 使用可重新配置的光子网用于光学光谱切片.
- 实现了用于光学领域预处理的模拟卷积引擎.
- 在MNIST数据集上测试了加速器的性能.
主要成果:
- 在MNIST数据集上实现了98.6%的数值准确度和97.7%的实验准确度.
- 证明了一些光子节点可以取代数字卷积神经网络模块.
- 与数字同行相比,估计减少了高达30%的电力消耗.
结论:
- 光子集成的神经形态加速器是机器学习的可行,节能的解决方案.
- 光谱切片为硬件加速的人工智能提供了一种强大的技术.
- 这种方法促进了光学领域的高效信息预处理和特征提取.
更多相关视频
10:45Anatomically Inspired Three-dimensional Micro-tissue Engineered Neural Networks for Nervous System Reconstruction, Modulation, and Modeling
Published on: May 31, 2017
12.7K
14:09High-Throughput Total Internal Reflection Fluorescence and Direct Stochastic Optical Reconstruction Microscopy Using a Photonic Chip
Published on: November 16, 2019
6.8K
相关概念视频
Ampere-Maxwell's Law: Problem-Solving
459
A parallel-plate capacitor with capacitance C, whose plates have area A and separation distance d, is connected to a resistor R and a battery of voltage V. The current starts to flow at t = 0. What is the displacement current between the capacitor plates at time t? From the properties of the capacitor, what is the corresponding real current?
To solve the problem, we can use the equations from the analysis of an RC circuit and Maxwell's version of Ampère's law.
For the first part of...
To solve the problem, we can use the equations from the analysis of an RC circuit and Maxwell's version of Ampère's law.
For the first part of...
459
Convolution: Math, Graphics, and Discrete Signals
201
In any LTI (Linear Time-Invariant) system, the convolution of two signals is denoted using a convolution operator, assuming all initial conditions are zero. The convolution integral can be divided into two parts: the zero-input or natural response and the zero-state or forced response, with t0 indicating the initial time.
To simplify the convolution integral, it is assumed that both the input signal and impulse response are zero for negative time values. The graphical convolution process...
To simplify the convolution integral, it is assumed that both the input signal and impulse response are zero for negative time values. The graphical convolution process...
201
Mesh Analysis with Current Sources
1.2K
Mesh analysis becomes simpler when analyzing circuits with current sources, whether independent or dependent. The presence of current sources reduces the number of equations required for analysis. Two cases illustrate this:
Current Source in One Mesh: The analysis process is straightforward when a current source is found in only one mesh within the circuit. Mesh currents are assigned as usual, with the mesh containing the current source excluded from the analysis. Kirchhoff's voltage law...
Current Source in One Mesh: The analysis process is straightforward when a current source is found in only one mesh within the circuit. Mesh currents are assigned as usual, with the mesh containing the current source excluded from the analysis. Kirchhoff's voltage law...
1.2K
Semiconductors
472
There is variation in the electrical conductivity of materials - metals, semiconductors, and insulators that are showcased with the help of the energy band diagrams.
Metals such as copper (Cu), zinc (Zn), or lead (Pb) have low resistivity and feature conduction bands that are either not fully occupied or overlap with the valence band, making a bandgap non-existent. This allows electrons in the highest energy levels of the valence band to easily transition to the conduction band upon gaining...
Metals such as copper (Cu), zinc (Zn), or lead (Pb) have low resistivity and feature conduction bands that are either not fully occupied or overlap with the valence band, making a bandgap non-existent. This allows electrons in the highest energy levels of the valence band to easily transition to the conduction band upon gaining...
472
