[TOC]
MorphNet is a method for learning deep network structure during training. The key principle is continuous relaxation of the network-structure learning problem. In short, the MorphNet regularizer pushes the influence of filters down, and once they are small enough, the corresponding output channels are marked for removal from the network.
Specifically, activation sparsity is induced by adding regularizers that target the consumption of specific resources such as FLOPs or model size. When the regularizer loss is added to the training loss and their sum is minimized via stochastic gradient descent or a similar optimizer, the learning problem becomes a constrained optimization of the structure of the network, under the constraint represented by the regularizer. The method was first introduced in our CVPR 2018, paper "MorphNet: Fast & Simple Resource-Constrained Learning of Deep Network Structure". A overview of the approach as well as device-specific latency regularizers were prestend in GTC 2019. [slides, recording: YouTube, GTC on-demand].
NEW: FiGS, is a probabilistic approach to channel regularization that we introduced in Fine-Grained Stochastic Architecture Search. It outperforms our previous regularizers and can be used as either a pruning algorithm or a full fledged Differentiable Architecture Search method.
Suppose you have a working convolutional neural network for image classification but want to shrink the model to satisfy some constraints (e.g., memory, latency). Given an existing model (the “seed network”) and a target criterion, MorphNet will propose a new model by adjusting the number of output channels in each convolution layer.
Note that MorphNet does not change the topology of the network -- the proposed model will have the same number of layers and connectivity pattern as the seed network.
To use MorphNet, you must:
Choose a regularizer from morphnet.network_regularizers
. The choice is
based on
LogisticSigmoid
regularizers.Gamma
regularizer if the seed network
has BatchNorm; use GroupLasso
otherwise.Note: If you use BatchNorm, you must enable the scale parameters (“gamma
variables”), i.e., by setting scale=True
if you are using
tf.keras.layers.BatchNormalization
.
Note: If you are using LogisticSigmoid
don't forget to add the
probabilistic gating op! See below for example.
Initialize the regularizer with a threshold and the output boundary ops and (optionally) the input boundary ops of your model.
MorphNet regularizer crawls your graph starting from the output boundary, and applies regularization to some of the ops it encounters. When it encounters any of the input boundary ops, it does not crawl past them (the ops in the input boundary are not regularized). The threshold determines which output channels can be eliminated.
Add the regularization term to your loss.
As always, regularization loss must be scaled. We recommend to search for
the scaling hyperparameter (regularization strength) along a logarithmic
scale spanning a few orders of magnitude around 1/(initial cost)
. For
example, if the seed network starts with 1e9 FLOPs, explore regularization
strength around 1e-9.
Note: MorphNet does not currently add the regularization loss to the tf.GraphKeys.REGULARIZATION_LOSSES collection; this choice is subject to revision.
Note: Do not confuse get_regularization_term()
(the loss you should add to
your training) with get_cost()
(the estimated cost of the network if the
proposed structure is applied).
Train the model.
Note: We recommend using a fixed learning rate (no decay) for this step, though this is not strictly necessary.
Save the proposed model structure with the StructureExporter
.
The exported files are in JSON format. Note that as the training progresses, the proposed model structure will change. There are no specific guidelines on the stopping time, although you would likely want to wait for the regularization loss (reported via summaries) to stabilize.
(Optional) Create summary ops to monitor the training progress through TensorBoard.
Modify your model using the StructureExporter
output.
Retrain the model from scratch without the MorphNet regularizer.
Note: Use the standard values for all hyperparameters (such as the learning rate schedule).
(Optional) Uniformly expand the network to adjust the accuracy vs. cost trade-off as desired. Alternatively, this step can be performed before the structure learning step.
We refer to the first round of training as structure learning and the second round as retraining.
To summarize, the key hyperparameters for MorphNet are:
Note that the regularizer type is not a hyperparameter because it's uniquely determined by the metric of interest (FLOPs, latency) and the presence of BatchNorm.
Regularizer classes can be found under network_regularizers/
directory. They
are named by the algorithm they use and the target cost they attempt to
minimize. For example, LogisticSigmoidFlopsRegularizer
uses a
Logistic-Sigmoid probabilistic method to to regularize the FLOP cost
and GammaModelSizeRegularizer
uses the batch norm gamma in
order to regularize the model size cost.
gating layers
to your model.The example below demonstrates how to use MorphNet to reduce the number of FLOPs
in your model. In this example, the regularizer will traverse the graph
starting with logits
, and will not go past any op that is earlier in the graph
than the inputs
or labels
; this allows to specify the subgraph for MorphNet to optimize.
#TODO Add Keras example.
from morph_net.network_regularizers import flop_regularizer
from morph_net.tools import structure_exporter
def build_model(inputs, labels, is_training, ...):
gated_relu = activation_gating.gated_relu_activation()
net = tf.layers.conv2d(inputs, kernel=[5, 5], num_outputs=256)
net = gated_relu(net, is_training=is_training)
...
...
net = tf.layers.conv2d(net, kernel=[3, 3], num_outputs=1024)
net = gated_relu(net, is_training=is_training)
logits = tf.reduce_mean(net, [1, 2])
logits = tf.layers.dense(logits, units=1024)
return logits
inputs, labels = preprocessor()
logits = build_model(inputs, labels, is_training=True, ...)
network_regularizer = flop_regularizer.LogisticSigmoidFlopsRegularizer(
output_boundary=[logits.op],
input_boundary=[inputs.op, labels.op],
alive_threshold=0.1 # Value in [0, 1]. This default works well for most cases.
)
regularization_strength = 1e-10
regularizer_loss = (network_regularizer.get_regularization_term() * regularization_strength)
model_loss = tf.nn.sparse_softmax_cross_entropy_with_logits(labels, logits)
optimizer = tf.train.MomentumOptimizer(learning_rate=0.01, momentum=0.9)
train_op = optimizer.minimize(model_loss + regularizer_loss)
You should monitor the progress of structure learning training via Tensorboard. In particular, you should consider adding a summary that computes the current MorphNet regularization loss and the cost if the currently proposed structure is adopted.
tf.summary.scalar('RegularizationLoss', regularizer_loss)
tf.summary.scalar(network_regularizer.cost_name, network_regularizer.get_cost())
Larger values of regularization_strength
will converge to smaller effective
FLOP count. If regularization_strength
is large enough, the FLOP count will
collapse to zero. Conversely, if it is small enough, the FLOP count will remain
at its initial value and the network structure will not vary. The
regularization_strength
parameter is your knob to control where you want to be
on the price-performance curve. The alive_threshold
parameter is used for
determining when an activation is alive.
During training, you should save a JSON file that contains the learned structure of the network, that is the count of activations in a given layer kept alive (as opposed to removed) by MorphNet.
exporter = structure_exporter.StructureExporter(
network_regularizer.op_regularizer_manager)
with tf.Session() as sess:
tf.global_variables_initializer().run()
for step in range(max_steps):
_, structure_exporter_tensors = sess.run([train_op, exporter.tensors])
if (step % 1000 == 0):
exporter.populate_tensor_values(structure_exporter_tensors)
exporter.create_file_and_save_alive_counts(train_dir, step)
Contact: [email protected]