# Copyright 2020-present, Pietro Buzzega, Matteo Boschini, Angelo Porrello, Davide Abati, Simone Calderara.
# All rights reserved.
# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.
import torch
from models.utils.continual_model import ContinualModel
from utils.args import add_rehearsal_args, ArgumentParser
from utils.buffer import Buffer
[docs]
class Fdr(ContinualModel):
"""Continual learning via Function Distance Regularization."""
NAME = 'fdr'
COMPATIBILITY = ['class-il', 'domain-il', 'task-il', 'general-continual']
[docs]
@staticmethod
def get_parser(parser) -> ArgumentParser:
add_rehearsal_args(parser)
parser.add_argument('--alpha', type=float, required=True,
help='Penalty weight.')
return parser
def __init__(self, backbone, loss, args, transform, dataset=None):
super(Fdr, self).__init__(backbone, loss, args, transform, dataset=dataset)
self.buffer = Buffer(self.args.buffer_size)
self.i = 0
self.soft = torch.nn.Softmax(dim=1)
self.logsoft = torch.nn.LogSoftmax(dim=1)
[docs]
def end_task(self, dataset):
examples_per_task = self.args.buffer_size // self.current_task if self.current_task > 0 else self.args.buffer_size
if self.current_task > 0:
buf_x, buf_log, buf_tl = self.buffer.get_all_data()
self.buffer.empty()
for ttl in buf_tl.unique():
idx = (buf_tl == ttl)
ex, log, tasklab = buf_x[idx], buf_log[idx], buf_tl[idx]
first = min(ex.shape[0], examples_per_task)
self.buffer.add_data(
examples=ex[:first],
logits=log[:first],
task_labels=tasklab[:first]
)
counter = 0
with torch.no_grad():
for i, data in enumerate(dataset.train_loader):
inputs, not_aug_inputs = data[0], data[2]
inputs = inputs.to(self.device)
not_aug_inputs = not_aug_inputs.to(self.device)
outputs = self.net(inputs)
if examples_per_task - counter < 0:
break
self.buffer.add_data(examples=not_aug_inputs[:(examples_per_task - counter)],
logits=outputs.data[:(examples_per_task - counter)],
task_labels=(torch.ones(self.args.batch_size) *
self.current_task)[:(examples_per_task - counter)])
counter += self.args.batch_size
[docs]
def observe(self, inputs, labels, not_aug_inputs, epoch=None):
self.i += 1
self.opt.zero_grad()
outputs = self.net(inputs)
loss = self.loss(outputs, labels)
loss.backward()
self.opt.step()
if not self.buffer.is_empty():
self.opt.zero_grad()
buf_inputs, buf_logits, _ = self.buffer.get_data(self.args.minibatch_size,
transform=self.transform, device=self.device)
buf_outputs = self.net(buf_inputs)
loss = torch.norm(self.soft(buf_outputs) - self.soft(buf_logits), 2, 1).mean()
assert not torch.isnan(loss)
loss.backward()
self.opt.step()
return loss.item()