spikingjelly.activation_based.layer package ============================================================= .. note:: **API稳定性说明** SpikingJelly ``0.0.0.1.0`` 对 ``layer`` 模块的内部实现进行了重构：原来的 ``layer.py`` 文件已被拆分并重组为 ``layer/`` 包，以提升代码的模块化程度和可维护性。 **该改动不会影响对外公开的 API。** 我们强烈建议用户仍然通过 ``layer`` 这一顶层命名空间来访问相关功能，而不是从具体的内部子模块中进行导入。 ``layer`` 层级下的导入路径被视为稳定的公共接口；更深层的子模块仅作为内部实现细节，未来可能发生变化。 .. code:: python from spikingjelly.activation_based.layer import SeqToANNContainer # 推荐 ✅ from spikingjelly.activation_based.layer.container import SeqToANNContainer # 不推荐 ❌ **API Stability Notice** We have refactored the internal implementation of the ``layer`` module. The original ``layer.py`` file has been reorganized into a package ( ``layer/`` ) for better modularity and maintainability. **This change does not affect the public API.** Users are strongly encouraged to continue accessing layers directly from the ``layer`` namespace, rather than importing from specific internal submodules. Import paths under ``layer`` are considered part of the stable public API, while deeper submodule paths are treated as implementation details and may change in future releases. .. code:: python from spikingjelly.activation_based.layer import SeqToANNContainer # recommended ✅ from spikingjelly.activation_based.layer.container import SeqToANNContainer # not recommended ❌ Containers ++++++++++ SpikingJelly 的 **容器** 封装了常见的网络结构，并支持单步（s）和多步（m）步进模式。 ---- SpikingJelly's **containers** wrap common network architectures and support both single-step (s) and multi-step (m) step modes. .. list-table:: * - :class:`MultiStepContainer ` - Container for multi-step forward pass. * - :class:`SeqToANNContainer ` - Container that merges the time and batch dimensions for forward pass. * - :class:`TLastMultiStepContainer ` - Multi-step container whose time dimension is placed at the last. * - :class:`TLastSeqToANNContainer ` - SeqToANNContainer whose time dimension is placed at the last. * - :class:`StepModeContainer ` - Container that sets the step mode for all contained modules. * - :class:`ElementWiseRecurrentContainer ` - Container for element-wise recurrent connections. * - :class:`LinearRecurrentContainer ` - Container for linear recurrent connections. .. toctree:: :hidden: spikingjelly.activation_based.layer.container Wrappers for Stateless Layers +++++++++++++++++++++++++++++ SpikingJelly 的 **无状态层包装器** 封装了 PyTorch 的标准层，并支持脉冲神经网络的时间步进模式。 ---- SpikingJelly's **stateless layer wrappers** wrap PyTorch's standard layers and support the step mode of Spiking Neural Networks. Convolutional Layers -------------------- .. list-table:: * - :class:`Conv1d ` - 1D convolution layer with step mode support. * - :class:`Conv2d ` - 2D convolution layer with step mode support. * - :class:`Conv3d ` - 3D convolution layer with step mode support. Convolutional Transpose Layers ------------------------------- .. list-table:: * - :class:`ConvTranspose1d ` - 1D transposed convolution layer with step mode support. * - :class:`ConvTranspose2d ` - 2D transposed convolution layer with step mode support. * - :class:`ConvTranspose3d ` - 3D transposed convolution layer with step mode support. Upsampling Layers ----------------- .. list-table:: * - :class:`Upsample ` - Upsampling layer with configurable interpolation methods. Pooling Layers -------------- .. list-table:: * - :class:`MaxPool1d ` - 1D max pooling layer with step mode support. * - :class:`MaxPool2d ` - 2D max pooling layer with step mode support. * - :class:`MaxPool3d ` - 3D max pooling layer with step mode support. * - :class:`AvgPool1d ` - 1D average pooling layer with step mode support. * - :class:`AvgPool2d ` - 2D average pooling layer with step mode support. * - :class:`AvgPool3d ` - 3D average pooling layer with step mode support. * - :class:`AdaptiveAvgPool1d ` - 1D adaptive average pooling layer with step mode support. * - :class:`AdaptiveAvgPool2d ` - 2D adaptive average pooling layer with step mode support. * - :class:`AdaptiveAvgPool3d ` - 3D adaptive average pooling layer with step mode support. Linear Layers ------------- .. list-table:: * - :class:`Linear ` - Linear transformation layer with step mode support. * - :class:`Flatten ` - Flatten layer with step mode support. Layers with Weight Standardization --------------------------------------- .. list-table:: * - :class:`WSConv2d ` - Weight Standardization 2D convolution layer with step mode support. * - :class:`WSLinear ` - Weight Standardization linear layer with step mode support. Group Normalization Layers -------------------------- .. list-table:: * - :class:`GroupNorm ` - Group normalization layer with step mode support. .. toctree:: :hidden: spikingjelly.activation_based.layer.stateless_wrapper Attention Layers ++++++++++++++++ SpikingJelly 的 **注意力层** 提供了用于深度脉冲神经网络的注意力机制实现，包括用于卷积 SNN 的注意力层和用于脉冲 Transformers 的注意力层。 ---- SpikingJelly's **attention layers** provide attention mechanisms for deep Spiking Neural Networks, including attention layers for convolutional SNNs and Spiking Transformers. Attention for Convolutional SNNs ---------------------------------------- .. list-table:: * - :class:`TemporalWiseAttention ` - Temporal-wise attention. * - :class:`MultiDimensionalAttention ` - Multi-dimensional attention. Attention for Spiking Transformers ------------------------------------------ .. list-table:: * - :class:`SpikingSelfAttention ` - Spiking self-attention for Spikformer. * - :class:`QKAttention ` - Query-Key attention for QKFormer. * - :class:`TokenQKAttention ` - Token-wise Query-Key attention for QKFormer. * - :class:`ChannelQKAttention ` - Channel-wise Query-Key attention for QKFormer. .. toctree:: :hidden: spikingjelly.activation_based.layer.attention Batch Normalization Variants ++++++++++++++++++++++++++++ SpikingJelly 提供了多种适用于深度 SNN 的 **批归一化层变体** 。 ---- SpikingJelly provides multiple **batch normalization variants** that are optimized for deep SNNs. Standard Batch Normalization ----------------------------- .. list-table:: * - :class:`BatchNorm1d ` - 1D batch normalization layer with step mode support. * - :class:`BatchNorm2d ` - 2D batch normalization layer with step mode support. * - :class:`BatchNorm3d ` - 3D batch normalization layer with step mode support. NeuNorm ------- .. list-table:: * - :class:`NeuNorm ` - Neural Normalization layer. Threshold-Dependent Batch Normalization ---------------------------------------- .. list-table:: * - :class:`ThresholdDependentBatchNorm1d ` - 1D threshold-dependent batch normalization layer. * - :class:`ThresholdDependentBatchNorm2d ` - 2D threshold-dependent batch normalization layer. * - :class:`ThresholdDependentBatchNorm3d ` - 3D threshold-dependent batch normalization layer. Temporal Effective Batch Normalization -------------------------------------- .. list-table:: * - :class:`TemporalEffectiveBatchNorm1d ` - 1D temporal effective batch normalization layer. * - :class:`TemporalEffectiveBatchNorm2d ` - 2D temporal effective batch normalization layer. * - :class:`TemporalEffectiveBatchNorm3d ` - 3D temporal effective batch normalization layer. Batch Normalization Through Time ------------------------------------ .. list-table:: * - :class:`BatchNormThroughTime1d ` - 1D batch normalization through time. * - :class:`BatchNormThroughTime2d ` - 2D batch normalization through time. * - :class:`BatchNormThroughTime3d ` - 3D batch normalization through time. .. toctree:: :hidden: spikingjelly.activation_based.layer.bn Dropout Variants ++++++++++++++++ SpikingJelly 提供了适用于 SNN 的 **Dropout 实现**，支持步进模式。 ---- SpikingJelly provides **dropout implementations** suitable for SNNs with step mode support. .. list-table:: * - :class:`Dropout ` - Dropout layer with step mode support. * - :class:`Dropout2d ` - 2D dropout layer with step mode support. * - :class:`DropConnectLinear ` - DropConnect linear layer with step mode support. .. toctree:: :hidden: spikingjelly.activation_based.layer.dropout Online Learning Modules +++++++++++++++++++++++ SpikingJelly 的 **在线学习模块** 提供了用于在线学习的辅助类和操作。 ---- SpikingJelly's **online learning modules** provide auxiliary classes and operations for online learning. .. list-table:: * - :class:`ReplaceforGrad ` - AutoGrad function for gradient replacement. * - :class:`GradwithTrace ` - Computing gradients using neuron traces. * - :class:`SpikeTraceOp ` - Operation for spike and trace handling in online training. * - :class:`OTTTSequential ` - Sequential container for online training through time. .. toctree:: :hidden: spikingjelly.activation_based.layer.online_learning Miscellaneous +++++++++++++ SpikingJelly 的 **杂项模块** 提供了辅助层和其他实用工具。 ---- SpikingJelly's **miscellaneous module** provides auxiliary layers and other utilities. .. list-table:: * - :class:`SynapseFilter ` - Synapse filter layer with exponential decay. * - :class:`PrintShapeModule ` - Module for printing input shapes during forward pass. * - :class:`VotingLayer ` - Voting layer for multi-class classification. * - :class:`Delay ` - Temporal delay operation. .. toctree:: :hidden: spikingjelly.activation_based.layer.misc