Advanced Technology M3 L
A
T EX workshop
Write like a programmer! Martin Klein Schaarsberg
University of T wente
Advanced Technology M 3 L T EX workshop A Write like a - - PowerPoint PPT Presentation
Advanced Technology M 3 L T EX workshop A Write like a programmer! Martin Klein Schaarsberg University of T wente September 11 and 12, 2014 Write like a programmer? Why? Consistency and clarity by separation of content and style
A
University of T wente
◮ You write your document in plain text with commands that
◮ The latex program processes your text and commands to
TeXworks TeXnicCenter Texmaker Notepad
MiKTeX MacTeX TeX Live
Gimp Illustrator Paint
.tex .pdf .eps .png .jpg
JabRef
.bib .sty .cls .dtx
.aux .bbl .log .toc .out .pdf .dvi pdflatex latex bibtex makeindex
AT
\documentclass[10pt]{article} \title{Sample document} \author{My name} \date{\today} \begin{document} \maketitle \end{document}
AT
\documentclass[10pt]{article} \title{Sample document} \author{My name} \date{\today} \begin{document} \maketitle \section{Introduction} \subsection{Food} \end{document}
AT
\documentclass[10pt]{article} \title{Sample document} \author{My name} \date{\today} \begin{document} \maketitle \section{Introduction} This is my \emph{first} document, created on \today. \subsection{Food} I like cookies, Pizza, and chocolate. \end{document}
AT
\documentclass[10pt]{article} \title{Sample document} \author{My name} \date{\today} \begin{document} \maketitle \section{Introduction} This is my \emph{first} document, created on \today. \subsection{Food} I like cookies, Pizza, and chocolate. \end{document}
◮ Commands add extra functions to your text
◮ Format: \command[optional_args]{required_args}
◮ Environments are used to treat sections specially
◮ Format:
\begin{environment}[optional_args] ... \end{environment}
◮ \title{} ◮ \author{} ◮ \date{} ◮ \maketitle{} ◮ \begin{abstract} ... \end{abstract} ◮ \part{} ◮ \chapter{} ◮ \section{} ◮ \subsection{} ◮ \subsubsection{} ◮ \paragraph{} ◮ \footnote{} ◮ \bibliography{}
\begin{itemize} \item First item \item Second item \end{itemize} ◮ First item ◮ Second item \begin{enumerate} \item First item \item Second item \end{enumerate}
\begin{description} \item[the ITco] does not fix your computer \item[the board] Best\'u\'u\'u\'u\'u\'ur! \end{description}
My favourite committee is \begin{enumerate} \item ITco \item BOSS \item Other \begin{itemize} \item NICAT \item INCA \item \dots \end{itemize} \end{enumerate}
◮ NICAT ◮ INCA ◮ . . .
AT
AT
AT
◮ Most errors come from not matching { with a }, a \begin{}
AT
◮ Undefined control sequence means you misspelled a
◮ Putting parts of equations on separate lines in the code will
◮ Exercise 1: see handout ◮ Practice with the following concepts:
◮ Creating PDFs ◮ Document structure ◮ White space, paragraphs and special characters ◮ Lists
◮ Presentation slides are available on
◮ Document classes:
◮ Style packages such as Koma-Script and memoir provide
◮ Options:
◮ Packages add functions to L
AT
◮ In the preamble, before \begin{document}
◮ Format: \usepackage[options]{package}
◮ Use the babel package
◮ (e.g. \usepackage[dutch]{babel}) ◮ Use \texteuro 30 to get €30 (requires the textcomp
package) ◮ Use the icomma package if you want to use a comma as
◮ Create accents like \'e, \`e, \^e (é, è, ê) ◮ Or directly input accented characters:
\usepackage{lmodern} or an other suitable font)
◮ Use $...$ for inline mathematics ◮ Use \begin{equation}...\end{equation} for displayed
◮ Use \begin{equation*} ... \begin{equation*} for
◮ Use the amsmath package for more symbols and
◮ A good reference guide: AMS Short Math Guide
3
1 2
a
i=1
ℓ=1
◮ Normal space disappears when used in equations: a
◮ You can use \quad or \qquad or \hspace{}
◮ Add text in equations with \text{} ◮ Common functions are predefined, such as \sin, \cos, \log,
\begin{multline} v_{n+h|n} = \sigma^2\bigg[1 + \alpha^2(h-1) + \frac{\beta\phi h}{(1-\phi)^2} \left\{2\alpha(1-\phi) + \beta\phi \right\} \\
\left\{ 2\alpha(1-\phi^2) + \beta\phi(1-\phi^h)\right\} \\ + \gamma h_m(2\alpha+\gamma) + \left\{\phi^m(1-\phi^{m h_m})\right\}\bigg]. \end{multline}
λ
◮ Exercise 2: see handout ◮ Practice with the following concepts:
◮ Document class ◮ Packages ◮ Mathematics
◮ Presentation slides are available on
◮ You need the graphicx package ◮ If you use pdfLaTeX, the image should be a jpg, pdf or png
◮ The booktabs package is recommended ◮ There exist tools to convert from Excel
◮ Display code without formatting with
function z = computefactorial(N) z = 1; for k = 1 : N z = z * k; end end ◮ Advanced formatting is possible with the listings package
1
function z = computefactorial(N)
2
z = 1;
3
for k = 1 : N
4
z = z * k;
5
end
6
end
◮ Floats are figures, tables, etc. with a caption ◮ L
AT
◮ You can specify placement options (here if possible, top of
◮ \begin{table}[htb] means put it ‘here’, or ‘top of page’ or
◮ Other possibilities are p for ‘whole page’ and ! meaning
◮ Always put the label after the caption!
◮ Cross-references allow you to refer to figures, tables,
◮ Always a pair of a \label{} and a reference ◮ Put a \label{xx} immediately when a number would have
◮ Add a reference in the text (imagine it is for Figure 3 on page
◮ \ref{xx} becomes ‘3’ ◮ \autoref{xx} becomes ‘Figure 3’ (with the hyperref package) ◮ \eqref{xx} becomes ‘(3)’ (for equations) ◮ \pageref{xx} becomes ‘2’
◮ L
AT
◮ Sections: \tableofcontents
◮ Use \setlength{tocdepth}{2} in the preamble to control
how many levels of sections appear ◮ Figures: \listoffigures ◮ Tables: \listoftables
◮ Exercise 3: see handout ◮ Practice with the following concepts:
◮ Figures ◮ Tables ◮ Floats ◮ Cross-references ◮ Table of contents and other automatic lists
◮ Presentation slides are available on
◮ BibTeX is a plain text citation database format ◮ You can import references from publishers websites, Scopus,
◮ You can use the program JabRef to manage the database ◮ You can use numbered citations or author-year (with the
◮ With a bibliography style, the citation list is formatted
◮ Use pdfLaTeX+MakeIndex+BibTeX in TeXworks to compile
@ARTICLE{Aran2011, author = {H.C. Aran and S. Pacheco Benito and M.W.J. Luiten-Olieman and S. Er and
and R.G.H. Lammertink}, title = {Carbon nanofibers in catalytic membrane microreactors}, journal = {Journal of Membrane Science}, year = {2011}, volume = {381}, pages = {244--250}, number = {1--2}, } @BOOK{Shah1979, title = {Gas--liquid--solid reactor design}, author = {Y. T. Shah}, publisher = {McGraw-Hill}, year = {1979}, }
AT
227 90Th+
Speech Recognition Techniques for a Sign Language Recognition System
Philippe Dreuw, David Rybach, Thomas Deselaers, Morteza Zahedi, and Hermann Ney Human Language Technology and Pattern Recognition, RWTH Aachen University, Aachen, Germany
Introduction
◮automatic sign language recognition system ◮necessary for communication between deaf and hearing people ◮continuous sign language recognition, several speakers, vision-basedapproach, no special hardware
◮large vocabulary speech recognition (LVSR) system to obtain a textualrepresentation of the signed sentences
◮evaluation of speech recognition techniques on publicly available sign languagecorpus Automatic Sign Language Recognition (ASLR)
◮similar to speech recognition:temporal sequences of images
◮important features ◮hand-shapes, facial expressions,lip-patterns
◮orientation and movement of thehands, arms or body
◮HMMs are used to compensate timeand amplitude variations of the signers
◮goal: find the model which bestexpresses the observation sequence
Video Input Feature Analysis Global Search: argmax wN 1Experimental Setup Database
◮system evaluation on the RWTH-BOSTON-104 database ◮201 sentences (161 training and 40 test sequences) ◮vocabulary size of 104 words ◮3 speakers (2 female, 1 male) ◮corpus is annotated in glossesProblems
◮26% of the training data are singletons ◮simple sentence structure ◮one out-of-vocabulary (OOV) words with whole-wordmodels Differences in Comparison to ASR
◮simultaneousness ◮signing space ◮environment ◮speakers and dialects ◮coarticulation and movement epenthesis ◮silence ◮whole-word models and sub-word unitsSystem Overview Visual Modeling (VM)
◮related to the acoustic model in ASR ◮HMM based, with separate GMMs, globally pooled diag. covariance matrix ◮monophone whole-word models ◮pronunciation handlingLanguage Modeling (LM)
◮according to ASR: LM should have a greater weight than the VM ◮trigram LM using the SRILM toolkit, with modified Kneser-Ney discounting withinterpolation Features
◮appearance-based image features: forbaseline system
◮thumbnails of video sequence frames(intensity images scaled to 32x32 pixels)
◮give a global description of all (manualand non-manual) features proposed in linguistic research
◮manual features: ◮dominant hand tracking: handposition, hand velocity, and hand trajectory features
40 60 80 100 120 140 20 40 60 80 100 hand positions ut eigenvectors vi 60 70 80 90 100 11050 60 70 80 90 100 hand positions ut eigenvectors viFeature Selection and Model Combination Feature Selection
◮concatenation ofappearance-based and manual features
◮sliding window for contextmodeling
◮dimensionality reduction byPCA and/or LDA
manual features head tracking scaling PCA / LDA hand tracking feature nation combi− non−manual features ut ut−2 ut−1Model Combination
◮log-linear combination ofindependently trained models
◮profit from independentalignments (e.g. performing well for long and short words)
◮profit from different featureextraction approaches
. . . . . . extraction feature Recognition appearance−based spatio−temporal patch−based body−parts model combination ut ut−2 ut−1Experimental Results
20 25 30 35 40 50 100 150 200 250 300 WER [\%] LDA feature dimensionality window size = 3 window size = 5 window size = 7 20 25 30 35 40 50 100 150 200 250 300 WER [\%] PCA feature dimensionality window size = 3 window size = 5 window size = 7 10 20 30 40 50 60 70 80 100 200 300 400 500 WER [%] LM scale zerogram unigram bigram trigramFeatures
frame intensity (w/o pronunciations) 1024 54.0 frame intensity (w/ pronunciations) 1024 37.0 frame intensity (w/ pronunciations + tangent distance) 1024 33.7 PCA-frame 110 27.5 PCA-frame, hand-position 112 25.3 PCA-frame, hand-velocity 112 24.2 PCA-frame, hand-trajectory 112 23.6 model-combination 2x100 17.9 Example Results Correct Examples IX-1P FIND SOMETHING-ONE BOOK IX-1P FIND SOMETHING-ONE BOOK JOHN FISH WONT EAT BUT CAN EAT CHICKEN JOHN FISH WONT EAT BUT CAN EAT CHICKEN LOVE JOHN WHO LOVE JOHN WHO JOHN BUY YESTERDAY WHAT BOOK JOHN BUY YESTERDAY WHAT BOOK Incorrect Examples MARY VEGETABLE KNOW IX LIKE CORN MARY VEGETABLE KNOW IX LIKE MARY JOHN IX GIVE MAN IX NEW COAT JOHN IX WOMAN NEW COAT LIKE CHOCOLATE WHO JOHN LIKE CHOCOLATE WHO JOHN [UNKNOWN] BUY HOUSE JOHN FUTURE NOT BUY HOUSE RWTH-BOSTON-104 Database Corpus Statistics Training Test sentences 161 40 running words 710 178 frames 12422 3324 vocabulary 103 65 singletons 27 9 OOV
LM Perplexities LM type PP zerogram 106.0 unigram 36.8 bigram 6.7 trigram 4.7 Database is publicly available Conclusion
◮LVSR system is suitable for vision-based continuous sign language recognition ◮many of the principles known from ASR can directly be transfered ◮important for ASLR: temporal contexts, pronunciation handling, languagemodelling, and model combination
◮outlook: connection of recognizer output to a statistical machine translationsystem achieved promising translation results Lehrstuhl f¨ ur Informatik 6 - Computer Science Department - RWTH Aachen University - Aachen, Germany Mail: dreuw@cs.rwth-aachen.de WWW: http://www-i6.informatik.rwth-aachen.de
ln(I/nA) Peaks ln(I/nA) z/Å −2 2 −2 2 −5 5
Senior Research Analyst, Strategy W F New Montgomery Street San Francisco, CA
+1-415-839-6885 x6753
+1-415-882-0495
✉ dtaraborelli@wikimedia.org hp://nitens.org/taraborelli
Areas of specialization
Cognitive science • Perceptual psychology • Philosophy of science
Areas of competence
Social computing • Human-computer interaction
Education
PhD in Cognitive Science (highest honours) École des Hautes Études en Sciences Sociales, France
: Roberto Casati, CNRS.
MSc in Cognitive Science École des Hautes Études en Sciences Sociales, France
: Pierre Jacob, CNRS; Kevin O’Regan, Université Paris .
MA in Philosophy of Science (highest honours) Università di Pisa, Italy
: Guglielmo Tamburrini, Università di Pisa; Adriana Fiorentini, CNR.
Licenza in Philosophy (highest honours) Scuola Normale Superiore di Pisa, Italy.
Appointments held
. Research
-present
Wikimedia Foundation Senior Research Analyst Strategic Development Team.
-present
University of Surrey Research Associate (-) Research Fellow (-) Centre for Research in Social Simulation, Department of Sociology.
-
University College London Marie Curie Fellow Chater Lab, Department of Psychology.
-
École des Hautes Études en Sciences Sociales Research Assistant Institut Jean-Nicod (CNRS-ENS-EHESS).
◮ Compile often ◮ Write neatly: indent code inside environments ◮ The Not So Short Introduction to L
AT
◮ For math, see this Short Math Guide for LaTeX ◮ Read package documentation (e.g. Windows Start - Run -
◮ Detexify is a nice website (also available as app) to find the
◮ For questions: http://tex.stackexchange.com ◮ No L
AT
◮ Typographic comparison L
AT
◮ The Not So Short Introduction to L
AT
◮ wikibook on L
AT
◮ John Lees-Miller’s L
AT
◮ Engelbert Buxbaum’s L
AT
◮ Rob J. Hyndman’s L
AT
◮ Daviid Reid’s L
AT
◮ Peter de Waal’s L
AT
◮ The Elements of Typographic Style by Robert Bringhurst