NJU LAMDA Thesis Template
Autor:
Zhuang Zhenhua
Letzte Aktualisierung:
vor 2 Jahren
Lizenz:
Creative Commons CC BY 4.0
Abstrakt:
NJU LAMDA Thesis Template
\begin
Discover why 18 million people worldwide trust Overleaf with their work.
NJU LAMDA Thesis Template
\begin
Discover why 18 million people worldwide trust Overleaf with their work.
\documentclass[aspectratio=169]{beamer}%页面比例16:9
\setbeamercovered{dynamic}%半透明显示
\setbeamertemplate{navigation symbols}{}
\usepackage[backend=bibtex,sorting=none,style=numeric]{biblatex}%设置引用
\addbibresource{ref.bib} %bib数据文件位置
\usepackage{dashrule}
\definecolor{NJUPurple}{rgb}{0.28235, 0.28235, 0.62745}%设置主题颜色
\colorlet{LightNJUPurple}{white!60!NJUPurple}
\colorlet{SuperLightNJUPurple}{white!90!NJUPurple}
\usecolortheme[named=NJUPurple]{structure}
\usepackage[utf8]{inputenc}
\usepackage{graphicx} % Allows including images
\usepackage{booktabs} % Allows the use of \toprule, \midrule and \bottomrule in tables
\usepackage{subfigure}
\usepackage{subfiles}
\usepackage{url}
\usepackage{amssymb}
\usepackage{amsmath}
\usepackage{xcolor,colortbl}
\usepackage[AutoFakeBold, AutoFakeSlant]{xeCJK}
\usefonttheme[onlymath]{serif}
\renewcommand{\today}{\number\year .\number\month .\number\day }
\setbeamertemplate{blocks}[rounded][shadow=true]
\setbeamercolor{block title}{fg=white,bg=NJUPurple}
\setbeamercolor{block body}{fg=black,bg=white}
\setbeamerfont{title}{shape=\bfseries,size=\Large}
\setbeamerfont{author}{shape=\bfseries}
\setbeamercolor{bibliography item}{fg = NJUPurple}
\setbeamercolor{bibliography entry title}{fg = black}
\setbeamercolor{bibliography entry author}{fg = black}
\setbeamercolor{bibliography entry location}{fg = black}
\setbeamercolor{bibliography entry note}{fg = black}
\makeatletter
\def\@makefnmark{\hbox{{{\usebeamercolor[fg]{footnote mark}\usebeamerfont*{footnote mark} [\@thefnmark]}}}}
\def\@makefntext#1{%
\def\insertfootnotetext{ #1}%
\def\insertfootnotemark{{\hbox{{{\usebeamercolor[NJUPurple]{footnote mark}\usebeamerfont*{footnote mark} [\@thefnmark]}}}}}%
\usebeamertemplate***{footnote}}
\makeatother
\renewcommand\footnoterule{\color{NJUPurple}\hdashrule{14cm}{1pt}{1.5mm 2pt} }
\setbeamertemplate{footnote}{
\vspace{0.05cm}
\scriptsize\insertfootnotemark\vspace{-0.38cm}\quad\ \insertfootnotetext\vspace{0.008cm}
}
\setbeamertemplate{title page}{%
\vbox{}
\vfill
\begingroup
\centering
\begin{beamercolorbox}[sep=8pt,center]{title}
\usebeamerfont{title}\inserttitle\par%
\ifx\insertsubtitle\@empty%
\else%
\vskip0.25em%
{\usebeamerfont{subtitle}\usebeamercolor[fg]{subtitle}\insertsubtitle\par}%
\fi%
\end{beamercolorbox}%
\vskip-0.2cm%<- changed
\begin{beamercolorbox}[sep=8pt,center]{institute}
\usebeamerfont{institute}\insertinstitute
\end{beamercolorbox}
\vskip-0.2cm%<- changed
\begin{beamercolorbox}[sep=8pt,center]{author}
\usebeamerfont{author}\insertauthor
\end{beamercolorbox}
\vskip-0.2cm%<- changed
\begin{beamercolorbox}[sep=8pt,center]{date}
\usebeamerfont{date}\insertdate
\end{beamercolorbox}%
\vskip0.5cm%<- changed
\endgroup
% \vfill%<- removed
}
\makeatother
% shape, colour of item, nested item bullets in itemize only
\setbeamertemplate{itemize item}[circle] \setbeamercolor{itemize item}{fg=NJUPurple}
\setbeamertemplate{itemize subitem}[circle] \setbeamercolor{itemize subitem}{fg=NJUPurple}
\setbeamertemplate{itemize subsubitem}[circle] \setbeamercolor{itemize subsubitem}{fg=NJUPurple}
\setbeamertemplate{itemize/enumerate body begin}{\normalsize}
\setbeamertemplate{itemize/enumerate subbody begin}{\normalsize}
\setbeamertemplate{itemize/enumerate subsubbody begin}{\normalsize}
\setbeamertemplate{frametitle}
{\leavevmode%
\begin{beamercolorbox}[wd=\paperwidth, ht=3.8ex, dp=0ex, leftskip = 1.9ex]{titlelike}%
\usebeamerfont{frametitle}\insertframetitle\hspace{0.2cm}\usebeamerfont{framesubtitle}\insertframesubtitle%
\end{beamercolorbox}%
}
\setbeamertemplate{section in toc}[circle] % 目录前设置序号
\setbeamerfont{frametitle}{series=\bfseries}% 标题加粗
\setbeamerfont{framesubtitle}{size=\fontsize{11}{11},series=\bfseries}
\usepackage{latexsym,amsmath,xcolor,multicol,booktabs,calligra}
\usepackage{graphicx,pstricks,listings,stackengine}
\usepackage{wasysym}
%Contents before every section's starting slide
\AtBeginSection[]
{
\begin{frame}
\frametitle{Outline}
\tableofcontents[
currentsection,
currentsubsection,
subsectionstyle=show/show/hide,
sectionstyle=show/shaded
]
\end{frame}
}
% 首页修改
\title{Your Title Your Title Your Title Your Title Your Title Your Title Your Title }
\subtitle{Conference 2024}
\institute{Your Institution}
\author{汇报人：君の名は}
\date{\today}
% 脚注修改
\defbeamertemplate{footline}{NGEGFootlineTemplate}{%
\leavevmode% 离开vmode，也就是离开竖直模式，进入水平模式
\begin{beamercolorbox}[wd=0.975\paperwidth,ht=2.25ex,dp=3ex,right]{title in head/foot}%
\ifnum \the\value{page}>1 \text{\href{http://www.lamda.nju.edu.cn}{http://www.lamda.nju.edu.cn}}\fi
\end{beamercolorbox}%
% \vskip0pt%
}
\setbeamertemplate{footline}[NGEGFootlineTemplate]
%+++++++++++++++++++++++++++++++++++正文开始+++++++++++++++++++++++++++++++++++%
\begin{document}
{
\usebackgroundtemplate{\includegraphics[width=\paperwidth,height=\paperheight]{figs/cover.pdf}}
\frame{\titlepage}
}
{
\usebackgroundtemplate{\includegraphics[width=\paperwidth,height=\paperheight]{figs/content.pdf}}
\begin{frame}
\frametitle{About the Author(分栏显示)}
\begin{columns}
\begin{column}{0.37\linewidth}
\begin{figure}
{\includegraphics[scale=0.16]{figs/profile.jpg}}
\end{figure}
{\scriptsize
{\textbf{LinaBell}}\\Research Fellow in Nanjing University, a member of LAMDA.
}
\end{column}
\begin{column}{0.65\linewidth}
\textbf{Latest works}
\begin{itemize}
\item NeurPIS
\item ICML
\item ICRL
\item TPAMI
\item PRML
\end{itemize}
\end{column}
\end{columns}
\end{frame}
% 储备知识部分
\section{Preliminary}
\begin{frame}
\frametitle{Preliminary(公式展示)}
\begin{itemize}
\item{\textbf{Learning Strategy}}
\\ \hspace*{\fill} \\
Optimization methods:
Pointwise loss (binary cross-entropy, mean square error), pairwise loss (BPR, WARP), and {\color{red}softmax loss}
\begin{gather}
\mathcal{L}_0 = -\sum_{(u,i)\in{O}^{+}}\log\frac{\exp{(\cos(\hat{\theta}_{ui})/\tau)}}{\exp{(\cos(\hat{\theta}_{ui})/\tau)}+\sum_{j\in{N}_{u}}\exp{(\cos(\hat{\theta}_{uj})/\tau)}},
\nonumber
\end{gather}
\end{itemize}
\end{frame}
% 相关工作
\section{Related Work}
\begin{frame}
\frametitle{Related Work(多级列表)}
\textbf{SOTA debiasing strategies}
\begin{itemize}
\item
\textbf{Sample re-weighting methods} (e.g. IPS-CN)\\
exploit the item popularity's inverse to re-weight loss of each instance.
\item
\textbf{Causal inference methods} (e.g. MACR, CausE)\\
\begin{itemize}
\item
specify the role of popularity bias in assumed causal
graphs
\item
mitigate the bias effect on the prediction.
\end{itemize}
\item
{
\textbf{Regularization-based frameworks} (e.g. Sam-reg) \\
\begin{itemize}
\item Provides a tunable mechanism for controlling the trade-off between recommendation accuracy and coverage.\\
\item
\textbf{Sam-reg} regularizes the biased correlation between user-item relevance and item popularity
\end{itemize}}
\end{itemize}
\end{frame}
% 方法部分
\section{Methodology}
\subsection{BC Loss}
\begin{frame}
\frametitle{Methodology of BC Loss}
\framesubtitle{BC Loss(二级标题)}
\begin{itemize}
\item
\textbf{BC Loss}
\begin{align}\label{equ:bc_loss}
\mathcal{L}_{\text{BC}} =
-\sum_{(u,i)\in{O}^{+}}\log\frac{\exp{(\cos(\hat{\theta}_{ui}{\color{red}+M_{ui}})/\tau)}}{\exp{(\cos(\hat{\theta}_{ui}{\color{red}+M_{ui}})/\tau)}+\sum_{j\in{N}_{u}}\exp{(\cos(\hat{\theta}_{uj})/\tau)}},
\nonumber
\end{align}
$M_{ui}$: the bias-aware angular margin for the interaction $(u,i)$
$$M_{ui} = \min \{\hat{\xi}_{ui}, \pi - \hat{\theta}_{ui}\}$$
\item\textbf{Intuition}\\
If a user-item pair is the hard interaction that can hardly be reconstructed by its popularity statistics, it holds a
high value of $\xi_{ui}$ and leads to a high value of $M_{ui}$. Henceforward, BC loss imposes the large angular
margin $M_{ui}$ between the negative item $j$ and positive item $i$.
\end{itemize}
\end{frame}
% 分析部分
\section{Analyses}
\subsection{Geometric Interpretation}
\begin{frame}
\frametitle{Analyses(图像展示)}
\framesubtitle{Geometric Interpretation}
\begin{itemize}
\item
\textbf{Geometric Interpretation}\\
User $u$ with one observed item $i$ and two unobserved items $j$ and $k$.\\
\begin{figure}
{\includegraphics[scale=0.92]{figs/fig.pdf}}
\end{figure}
\end{itemize}
\end{frame}
\subsection{Theoretical Properties}
\begin{frame}
\frametitle{Analyses(数学环境)}
\framesubtitle{Theoretical Properties}
\begin{itemize}
\item
\textbf{Theoretical Properties}
\begin{proof}
1. There exists an upper bound $m$, s.t. $-1 < \cos(\hat{\theta}_{ui}+M_{ui}) \leq {v}_u^T{v}_i - m < 1 $\\
2. \\
3. \\
4. \\
5. \\
6. \\
\end{proof}
\end{itemize}
\end{frame}
% 实验部分
\section{Experiments}
\begin{frame}
\frametitle{Experiments(表格展示)}
\framesubtitle{Baselines \& Datasets}
\textbf{Baselines}
\begin{itemize}
\item
Backbone: only use softmax loss
\item
IPS-CN: sample re-weighting methods
\item
CausE: bias removal by causal inference
\item
sam + reg: regularization-based framework
\item
MACR: bias removal by causal inference
\end{itemize}
\textbf{Datasets}
\resizebox{\columnwidth}{!}{
\begin{tabular}{lrrrrrrr}
\toprule
& KuaiRec & Douban Movie & Tencent & Amazon-Book & Alibaba-iFashion & Yahoo!R3 & Coat\\ \midrule
\#Users & 7175 & 36,644 & 95,709 & 52,643 & 300,000 & 14382 & 290 \\
\#Items & 10611 & 22,226 & 41,602 & 91,599 & 81,614 & 1000 & 295 \\
\#Interactions & 1062969 & 5,397,926 & 2,937,228 & 2,984,108 & 1,607,813 & 129,748 & 2,776 \\
Sparsity & 0.01396 & 0.00663 & 0.00074 & 0.00062 & 0.00007
& 0.00902 & 0.03245\\ \bottomrule
\end{tabular}}
\end{frame}
% 结论部分
\section{Conclusion}
\begin{frame}
\frametitle{Conclusion(脚注使用)}
\begin{itemize}
\item
\textbf{Contribution}\\
\begin{itemize}
\item
(Originality) Popular bias extractor has an intuitive geometric interpretation.
\item
(Quality) Outperforms existing methods in various evaluation protocols.
\item
(Clarity) Well-written and easy to understand. Theoretical proof is quite solid.
\end{itemize}
\item
\textbf{Limitation}\\
\begin{itemize}
\item
The technical contribution of this paper is limited.It only proposes to employ an extra popularity-based predictor and combine the results with an existing CF model\footfullcite{he2020momentum}.
\item
Overclaims the strength of the proposed BC loss in theoretical analysis. The geometric interpretability and the hard-negative mining ability are actually the same thing\parencite{ pmlr-v119-wang20k, yuan2021one}
\end{itemize}
\end{itemize}
\end{frame}
% 引用部分
\section*{References}
\begin{frame}[allowframebreaks]
\frametitle{References}\color{NJUPurple}{
\printbibliography[heading=none]}
\end{frame}
% 谢辞部分
\section*{Acknowledgement}
\begin{frame}
\frametitle{Acknowledgement}
\textcolor{NJUPurple}{\Huge{\centerline{Thank you!}}}
\end{frame}
}
\end{document}