Skip to content
GitLab
Projects
Groups
Snippets
Help
Loading...
Help
What's new
10
Help
Support
Community forum
Keyboard shortcuts
?
Submit feedback
Contribute to GitLab
Sign in
Toggle navigation
Open sidebar
Bui Thi Mai PHUONG
ard_poster
Commits
a7e83b82
Commit
a7e83b82
authored
Oct 02, 2018
by
Amelie Royer
Browse files
Options
Browse Files
Download
Email Patches
Plain Diff
problem formulation
parent
29defe1f
Changes
1
Hide whitespace changes
Inline
Side-by-side
Showing
1 changed file
with
16 additions
and
10 deletions
+16
-10
finetuning.tex
finetuning.tex
+16
-10
No files found.
finetuning.tex
View file @
a7e83b82
\begin{minipage}
[c]
{
0.55
\textwidth
}
\textbf
{
Setting:
}
\vspace
{
0.45cm
}
\begin{minipage}
[c]
{
0.46
\textwidth
}
\textbf
{
Problem Formulation:
}
\begin{itemize}
\item
classes appear sequentially (or in batches)
% $c_1,c_2,\dots,c_T$
\item
each class
$
y
$
has training images
$
X
^
y
=
\{
x
^
y
_
1
,
\dots
,x
^
y
_{
n
_
y
}
\}
$
\end{itemize}
\item
Pretrained model
$
M
$
for task
$
T: X
\rightarrow
Y
$
\item
How to most efficiently fine-tune
$
M
$
for new task
$
T': X'
\rightarrow
Y
$
where
$
\mathcal
{
P
}
(
X
)
\neq
\mathcal
{
P
}
(
X'
)
$
\item
\textbf
{
Baseline:
}
Fine-tuning
\begin{itemize}
\item
slow for large models
\item
overfitting (small
$
X'
$
, many parameters)
\end{itemize}
\end{itemize}
\end{minipage}
~
\begin{minipage}
[c]
{
0.4
\textwidth
}
%
\begin{minipage}
[c]
{
0.4
9
\textwidth
}
\begin{center}
\newcommand
{
\pacsheight
}{
2.5cm
}
\textbf
{
Example (PACS~
\blue
{
[1]
}
dataset and variants)
}
\begin{minipage}
[c]
{
0.1
4
\textwidth
}
\begin{minipage}
[c]
{
0.1
6
\textwidth
}
\centering
\includegraphics
[height=\pacsheight]
{
pacs
_
1.png
}
\centering
\scriptsize
Photo
\end{minipage}
~
\centering
\scriptsize
\blue
{
Photo (base)
}
\end{minipage}
~
~~
\begin{minipage}
[c]
{
0.14
\textwidth
}
\centering
\includegraphics
[height=\pacsheight]
{
pacs
_
2.png
}
...
...
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
.
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment