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
cabb5896
Commit
cabb5896
authored
Oct 02, 2018
by
Christoph Lampert
Browse files
Options
Browse Files
Download
Email Patches
Plain Diff
shuffled
parent
a34b7bf1
Changes
2
Hide whitespace changes
Inline
Side-by-side
Showing
2 changed files
with
127 additions
and
49 deletions
+127
-49
finetuning.tex
finetuning.tex
+47
-43
poster.tex
poster.tex
+80
-6
No files found.
finetuning.tex
View file @
cabb5896
\newlength
{
\pacsheight
}
\setlength
{
\pacsheight
}{
2.5cm
}
\vspace
{
0.45cm
}
\begin{minipage}
[c]
{
0.46
\textwidth
}
\textbf
{
Problem Formulation:
}
\begin{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/intensive for large models
\item
might overfit (small
$
X'
$
, many parameters)
\end{itemize}
\end{itemize}
\item
how to fine-tune pretrained model
$
M
$
for new 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/intensive for large models
% \item might overfit (small $X'$, many parameters)
% \end{itemize}
\end{itemize}
\textbf
{
Proposed
}
: flexible Fine-tuning of internal layers
\begin{itemize}
\item
allow any layer to be tuned, not just last
\item
automatic selection criterion
\end{itemize}
\end{minipage}
%
\begin{minipage}
[c]
{
0.49
\textwidth
}
\begin{center}
\newcommand
{
\pacsheight
}{
2.5cm
}
\textbf
{
Example (PACS~
\blue
{
[1]
}
dataset and variants)
}
\vspace
{
0.25cm
}
\begin{minipage}
[c]
{
0.16
\textwidth
}
\begin{minipage}
[c]
{
0.25
\textwidth
}
\includegraphics
[width=\textwidth]
{
fine-tuning/selection
_
criterion
_
pacs.png
}
\end{minipage}
\quad
\begin{minipage}
[c]
{
0.22
\textwidth
}
\begin{minipage}
[c]
{
.4
\textwidth
}
\centering
\includegraphics
[height=\pacsheight]
{
pacs
_
1.png
}
\centering
\scriptsize
\blue
{
Photo (base)
}
\end{minipage}
~
\hspace
{
5pt
}
\vrule\hspace
{
5pt
}
~
\begin{minipage}
[c]
{
0.14
\textwidth
}
\end{minipage}
%~\hspace{5pt}\vrule\hspace{5pt}~
%
\begin{minipage}
[c]
{
.4
\textwidth
}
\centering
\includegraphics
[height=\pacsheight]
{
pacs
_
4.png
}
\centering
\scriptsize
Painting
\end{minipage}
\begin{minipage}
[c]
{
.4
\textwidth
}
\centering
\includegraphics
[height=\pacsheight]
{
pacs
_
2.png
}
\centering
\scriptsize
Noisy
\end{minipage}
~
\begin{minipage}
[c]
{
0.14
\textwidth
}
\centering
\includegraphics
[height=\pacsheight]
{
pacs
_
3.png
}
\end{minipage}
%
\begin{minipage}
[c]
{
.4
\textwidth
}
\centering
\includegraphics
[height=\pacsheight]
{
pacs
_
5.png
}
\centering
\scriptsize
Blurry
\centering
\scriptsize
Cartoon
\end{minipage}
~
\begin{minipage}
[c]
{
0.14
\textwidth
}
\centering
\includegraphics
[height=\pacsheight]
{
pacs
_
4.png
}
\centering
\scriptsize
Painting
\end{minipage}
~
\begin{minipage}
[c]
{
0.14
\textwidth
}
\centering
\includegraphics
[height=\pacsheight]
{
pacs
_
5.png
}
\begin{minipage}
[c]
{
.4
\textwidth
}
\centering
\includegraphics
[height=\pacsheight]
{
pacs
_
3.png
}
\centering
\scriptsize
Cartoon
\centering
\scriptsize
Blurry
\end{minipage}
~
\begin{minipage}
[c]
{
0.14
\textwidth
}
%
\begin{minipage}
[c]
{
.4
\textwidth
}
\centering
\includegraphics
[height=\pacsheight]
{
pacs
_
6.png
}
\centering
\scriptsize
Sketch
\end{minipage}
\end{center}
\end{minipage}
\begin{minipage}
[c]
{
0.45
\textwidth
}
\textbf
{
Proposed
}
: Flexible Fine-tuning of internal layers
\begin{itemize}
\item
Layer Selection Criterion
\item
Experiments in various settings
\item
Projection module (parametrized case)
\end{itemize}
\end{minipage}
\begin{minipage}
[c]
{
0.5
\textwidth
}
\raisebox
{
-0.5
\height
}{
\includegraphics
[width=0.58\textwidth]
{
fine-tuning/seelction
_
criterion.pdf
}}
`~~
\raisebox
{
-0.5
\height
}{
\includegraphics
[width=0.33\textwidth]
{
fine-tuning/selection
_
criterion
_
pacs.png
}}
\end{minipage}
%\begin{center}
% \textbf{Example (PACS~\blue{[1]} dataset and variants)}
%\vspace{0.25cm}
%\end{center}
poster.tex
View file @
cabb5896
...
...
@@ -8,10 +8,16 @@
\usepackage
[scaled]
{
helvet
}
\usepackage
{
pbox
}
\usepackage
{
framed
}
\newenvironment
{
fshaded
}{
%
\def\FrameCommand
{
\fcolorbox
{
framecolor
}{
shadecolor
}}
%
\MakeFramed
{
\FrameRestore
}}
%
{
\endMakeFramed
}
\definecolor
{
shadecolor
}{
rgb
}{
0.8,0.8,0.8
}
\definecolor
{
framecolor
}{
rgb
}{
.0,.0,.0
}
%
%\usepackage{algorithm}
%\usepackage{algorithmic}
%\usepackage{framed}
%\usepackage{mathrsfs}
%\usepackage{stmaryrd} % for /llbracket, /rrbracket
%\usepackage{rotating}
...
...
@@ -181,16 +187,84 @@
\ \ \
\begin{column}
{
.49
\textwidth
}
\begin{block}
{
\Large
Our Research
}
\definecolor
{
shadecolor
}{
rgb
}{
0.7,0.7,0.7
}
%
%\begin{fshaded}
\textbf
{
Theory (Statistical Machine Learning)
}
\begin{columns}
\begin{column}
{
.25
\textwidth
}
\begin{itemize}
\item
{
Multi-task learning
}
\end{itemize}
\end{column}
%
\begin{column}
{
.27
\textwidth
}
\begin{itemize}
\item
Lifelong learning
\end{itemize}
\end{column}
%
\begin{column}
{
.35
\textwidth
}
\begin{itemize}
\item
Learning with Strong Supervision
\end{itemize}
\end{column}
\end{columns}
%\end{fshaded}
\vspace
{
.5
\baselineskip
}
%\begin{fshaded}
\textbf
{
Models/Algorithms
}
%
\begin{columns}
\begin{column}
{
.25
\textwidth
}
\begin{itemize}
\item
{
Incremental learning
}
\end{itemize}
\end{column}
\begin{column}
{
.27
\textwidth
}
\begin{itemize}
\item
Multi-stage architectures
\end{itemize}
\end{column}
\begin{column}
{
.35
\textwidth
}
\begin{itemize}
\item
\LARGE
Lots
\item
of
\item
Cool
\item
Stuff
\item
Non-standard forms of supervision
\end{itemize}
\end{column}
\end{columns}
%\end{fshaded}
%\begin{fshaded}
\vspace
{
.5
\baselineskip
}
\textbf
{
Applications (in Computer Vision)
}
\begin{columns}
\begin{column}
{
.25
\textwidth
}
\begin{itemize}
\item
{
Object recognition
}
\end{itemize}
\end{column}
\begin{column}
{
.27
\textwidth
}
\begin{itemize}
\item
{
Object detection
}
\end{itemize}
\end{column}
\begin{column}
{
.35
\textwidth
}
\begin{itemize}
\item
{
Semantic image representations
}
\end{itemize}
\end{column}
\end{columns}
%\end{fshaded}
\end{block}
\begin{block}
{
\Large
Example: Flex-tuning
}
\vspace
{
1
\baselineskip
}
\begin{block}
{
\Large
Flexible Fine-tuning (Flex-Tuning)
}
\input
{
finetuning.tex
}
\end{block}
...
...
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