vat.py

# Copyright 2019 Google LLC
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Virtual adversarial training:a regularization method for supervised and semi-supervised learning.

Application to SSL of https://arxiv.org/abs/1704.03976
"""

import functools
import os

from absl import app
from absl import flags
from easydict import EasyDict
from libml import utils, data, layers, models
import tensorflow as tf
from third_party import vat_utils

FLAGS = flags.FLAGS


class VAT(models.MultiModel):

    def model(self, lr, wd, ema, warmup_pos, vat, vat_eps, entmin_weight, **kwargs):
        hwc = [self.dataset.height, self.dataset.width, self.dataset.colors]
        x_in = tf.placeholder(tf.float32, [None] + hwc, 'x')
        y_in = tf.placeholder(tf.float32, [None] + hwc, 'y')
        l_in = tf.placeholder(tf.int32, [None], 'labels')
        wd *= lr
        warmup = tf.clip_by_value(tf.to_float(self.step) / (warmup_pos * (FLAGS.train_kimg << 10)), 0, 1)

        classifier = functools.partial(self.classifier, **kwargs)
        l = tf.one_hot(l_in, self.nclass)
        logits_x = classifier(x_in, training=True)
        post_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)  # Take only first call to update batch norm.
        logits_y = classifier(y_in, training=True)
        delta_y = vat_utils.generate_perturbation(y_in, logits_y, lambda x: classifier(x, training=True), vat_eps)
        logits_student = classifier(y_in + delta_y, training=True)
        logits_teacher = tf.stop_gradient(logits_y)
        loss_vat = layers.kl_divergence_from_logits(logits_student, logits_teacher)
        loss_vat = tf.reduce_mean(loss_vat)
        loss_entmin = tf.reduce_mean(tf.distributions.Categorical(logits=logits_y).entropy())

        loss = tf.nn.softmax_cross_entropy_with_logits_v2(labels=l, logits=logits_x)
        loss = tf.reduce_mean(loss)
        tf.summary.scalar('losses/xe', loss)
        tf.summary.scalar('losses/vat', loss_vat)
        tf.summary.scalar('losses/entmin', loss_entmin)

        ema = tf.train.ExponentialMovingAverage(decay=ema)
        ema_op = ema.apply(utils.model_vars())
        ema_getter = functools.partial(utils.getter_ema, ema)
        post_ops.append(ema_op)
        post_ops.extend([tf.assign(v, v * (1 - wd)) for v in utils.model_vars('classify') if 'kernel' in v.name])

        train_op = tf.train.AdamOptimizer(lr).minimize(loss + loss_vat * warmup * vat + entmin_weight * loss_entmin,
                                                       colocate_gradients_with_ops=True)
        with tf.control_dependencies([train_op]):
            train_op = tf.group(*post_ops)

        # Tuning op: only retrain batch norm.
        skip_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
        classifier(x_in, training=True)
        train_bn = tf.group(*[v for v in tf.get_collection(tf.GraphKeys.UPDATE_OPS)
                              if v not in skip_ops])

        return EasyDict(
            x=x_in, y=y_in, label=l_in, train_op=train_op, tune_op=train_bn,
            classify_raw=tf.nn.softmax(classifier(x_in, training=False)),  # No EMA, for debugging.
            classify_op=tf.nn.softmax(classifier(x_in, getter=ema_getter, training=False)))


def main(argv):
    del argv  # Unused.
    dataset = data.DATASETS[FLAGS.dataset]()
    log_width = utils.ilog2(dataset.width)
    model = VAT(
        os.path.join(FLAGS.train_dir, dataset.name),
        dataset,
        lr=FLAGS.lr,
        wd=FLAGS.wd,
        arch=FLAGS.arch,
        warmup_pos=FLAGS.warmup_pos,
        batch=FLAGS.batch,
        nclass=dataset.nclass,
        ema=FLAGS.ema,
        smoothing=FLAGS.smoothing,
        vat=FLAGS.vat,
        vat_eps=FLAGS.vat_eps,
        entmin_weight=FLAGS.entmin_weight,

        scales=FLAGS.scales or (log_width - 2),
        filters=FLAGS.filters,
        repeat=FLAGS.repeat)
    model.train(FLAGS.train_kimg << 10, FLAGS.report_kimg << 10)


if __name__ == '__main__':
    utils.setup_tf()
    flags.DEFINE_float('wd', 0.02, 'Weight decay.')
    flags.DEFINE_float('vat', 0.3, 'VAT weight.')
    flags.DEFINE_float('vat_eps', 6, 'VAT perturbation size.')
    flags.DEFINE_float('entmin_weight', 0.06, 'Entropy minimization weight.')
    flags.DEFINE_float('warmup_pos', 0.4, 'Relative position at which constraint loss warmup ends.')
    flags.DEFINE_float('ema', 0.999, 'Exponential moving average of params.')
    flags.DEFINE_float('smoothing', 0.1, 'Label smoothing.')
    flags.DEFINE_integer('scales', 0, 'Number of 2x2 downscalings in the classifier.')
    flags.DEFINE_integer('filters', 32, 'Filter size of convolutions.')
    flags.DEFINE_integer('repeat', 4, 'Number of residual layers per stage.')
    FLAGS.set_default('dataset', 'cifar10.3@250-5000')
    FLAGS.set_default('batch', 64)
    FLAGS.set_default('lr', 0.002)
    FLAGS.set_default('train_kimg', 1 << 16)
    app.run(main)