tcn

Temporal Convolutional Network (TCN)

Overview

Temporal convolutional network (TCN) is a type of convolutional neural network (CNN) that is commonly used for sequence modeling tasks such as speech recognition, text generation, and video classification. TCN is a fully convolutional network that consists of a series of dilated causal convolutional layers. The dilated convolutional layers allow TCN to have a large receptive field while maintaining a small number of parameters. TCN is also fully parallelizable, which allows for faster training and inference times.

For more info, refer to the original paper Temporal Convolutional Networks: A Unified Approach to Action Segmentation.

Parameters are in helia_edge.models.tcn_params.

Functions

Name	Description
build	TCN model from TcnParams
normalization	Normalization layer
tcn_block_lg	TCN large block
tcn_block_mb	TCN mbconv block
tcn_block_sm	TCN small block
tcn_core	TCN core
tcn_layer	TCN functional layer

Additions

The TCN architecture has been modified to allow the following:

• Convolutional pairs can be factorized into depthwise separable convolutions.
• Squeeze and excitation (SE) blocks can be added between convolutional pairs.
• Normalization can be set between batch normalization and layer normalization.
• ReLU is replaced with the approximated ReLU6.

Usage

The following example builds a TCN from TcnParams and TcnBlockParams through a ModelSpec.

import helia_edge as helia

params = helia.models.TcnParams(
    input_kernel=(1, 3),
    input_norm="batch",
    blocks=[
        helia.models.TcnBlockParams(filters=8, kernel=(1, 3), dilation=(1, 1), dropout=0.1, ex_ratio=1, se_ratio=0, norm="batch"),
        helia.models.TcnBlockParams(filters=16, kernel=(1, 3), dilation=(1, 2), dropout=0.1, ex_ratio=1, se_ratio=0, norm="batch"),
        helia.models.TcnBlockParams(filters=24, kernel=(1, 3), dilation=(1, 4), dropout=0.1, ex_ratio=1, se_ratio=4, norm="batch"),
        helia.models.TcnBlockParams(filters=32, kernel=(1, 3), dilation=(1, 8), dropout=0.1, ex_ratio=1, se_ratio=4, norm="batch"),
    ],
    output_kernel=(1, 3),
    include_top=True,
    num_classes=5,
    use_logits=True,
)
model = helia.models.build(helia.models.ModelSpec(params=params, input_shape=(1, 800, 1)))

Machine-readable model

• normalizationfunctionNormalization layer
• tcn_block_lgfunctionTCN large block
• tcn_block_mbfunctionTCN mbconv block
• tcn_block_smfunctionTCN small block
• tcn_corefunctionTCN core
• tcn_layerfunctionTCN functional layer
• buildfunctionBuild a TCN model.

function

normalization

Python

Normalization layer

helia_edge/models/tcn.py:63

normalization(norm: str, name: str) -> keras.Layer

Normalization layer

Parameters

Parameters of normalization

Name	Type	Default	Description
norm	str	Required	Normalization type
name	str	Required	Name

Returns

Returns of normalization

Type	Description
keras.Layer	keras.Layer: Layer

function

tcn_block_lg

Python

TCN large block

helia_edge/models/tcn.py:92

tcn_block_lg(params: TcnBlockParams, name: str) -> keras.Layer

TCN large block

Parameters

Parameters of tcn_block_lg

Name	Type	Default	Description
params	TcnBlockParams	Required	Parameters
name	str	Required	Name

Returns

Returns of tcn_block_lg

Type	Description
keras.Layer	keras.Layer: Layer

function

tcn_block_mb

Python

TCN mbconv block

helia_edge/models/tcn.py:157

tcn_block_mb(params: TcnBlockParams, name: str) -> keras.Layer

TCN mbconv block

Parameters

Parameters of tcn_block_mb

Name	Type	Default	Description
params	TcnBlockParams	Required	Parameters
name	str	Required	Name

Returns

Returns of tcn_block_mb

Type	Description
keras.Layer	keras.Layer: Layer

function

tcn_block_sm

Python

TCN small block

helia_edge/models/tcn.py:258

tcn_block_sm(params: TcnBlockParams, name: str) -> keras.Layer

TCN small block

Parameters

Parameters of tcn_block_sm

Name	Type	Default	Description
params	TcnBlockParams	Required	Parameters
name	str	Required	Name

Returns

Returns of tcn_block_sm

Type	Description
keras.Layer	keras.Layer: Layer

function

tcn_core

Python

TCN core

helia_edge/models/tcn.py:344

tcn_core(params: TcnParams) -> keras.Layer

TCN core

Parameters

Parameters of tcn_core

Name	Type	Default	Description
params	TcnParams	Required	Parameters

Returns

Returns of tcn_core

Type	Description
keras.Layer	keras.Layer: Layer

function

tcn_layer

Python

TCN functional layer

helia_edge/models/tcn.py:373

tcn_layer(x: keras.KerasTensor, params: TcnParams) -> keras.KerasTensor

TCN functional layer

Parameters

Parameters of tcn_layer

Name	Type	Default	Description
x	keras.KerasTensor	Required	Input tensor
params	TcnParams	Required	Parameters

Returns

Returns of tcn_layer

Type	Description
keras.KerasTensor	keras.KerasTensor: Output tensor

function

build

Python

Build a TCN model.

helia_edge/models/tcn.py:428

build(
    params: TcnParams,
    input_shape: tuple[int | None, ...],
    *,
    batch_size: int | None = None,
    name: str | None = None,
) -> keras.Model

Build a TCN model.

Parameters

Parameters of build

Name	Type	Default	Description
params	TcnParams	Required	Model parameters.
input_shape	tuple[int | None, ...]	Required	Input shape without the batch axis; None for a variable axis.
batch_size	int | None	None	Static batch size; None for a dynamic batch.
name	str | None	None	Model name; the family when None.

Returns

Returns of build

Type	Description
keras.Model	keras.Model: The model, named ``tcn`` unless ``name`` is given.