UCR CS 260: SEMINAR IN COMPUTER SCIENCE

1
UCR CS 260 SEMINAR IN COMPUTER SCIENCE 

• 940 a.m. - 1100 a.m.
• CHUNG 139
• Winston Chung Hall (CHUNG)
• Eamonn KeoghComputer Science Engineering
  Departmenteamonn(AT)cs.ucr.edu

2
Format of the Class

• Very informal
• We will take turns to present and discuss papers
• You will be graded on your presentation, and on
  your participation

3
Presenting a paper

• You can request the original slides from the
  author (for a conference paper).
• Write a very polite email to first author, but CC
  all authors.
• You need to do this well in advance.
• You need to augment the slides, and make it clear
  what are your words/analysis
• Plan to speak for 25 min (not counting
  questions/discussion)

4
Prepend anything we need to know before you start
discussing the paper. At a minimum, where/when
the paper was published and its impact Us a
different font color to make it clear these are
your words
Append a discussion of the paper Us a different
font color to make it clear these are your words
5
Your discussion of the paper

• Is the paper
• Original? (including original compared to
  authors previous work)
• Well written?
• Well motivated?
• Well illustrated?
• Are the experiments
• Forceful
• Fair to rival methods
• Reproducible
• Does the algorithm require many parameters, if so
• Is it explained how to set them?
• Are we shown how sensitive the results are to
  them?

6
Your discussion of the paper

• What are the limitations of the work?
• Are they clearly stated?
• What are the assumptions of the work?
• Are they clearly stated?
• Is there a good literature review? If not, what
  is missing?
• Do you think a simpler method might have worked
  as well as the proposed method?

7
Your discussion of the paper

• Did the paper garner interest?
• Consider Eamonn J. Keogh, Jessica Lin, Ada
  Wai-Chee Fu HOT SAX Efficiently Finding the
  Most Unusual Time Series Subsequence. ICDM 2005
  226-233

It has 276 citations It was expanded into a
journal paper (which has 54 citations)
8
Your discussion of the paper

• Did the paper garner interest?

It is the most cited paper, of that conference in
that year
9
You need to understand the paper

• I know you did not write the paper, and you are
  not responsible for the content.
• However you cannot simply show someone else
  slides and shrug you shoulders if asked
  questions.
• You should try to anticipate any questions that
  an audience member might ask, and research an
  answer.
• So, to present a paper, you might have to read 3
  or 5 other papers. http//videolectures.net/

10
You need to understand the paper

• Hint many papers have videos of their
  presentation online. Often you can hear the
  audience questions
• http//videolectures.net/

11
Lier aide d'interprétation II
Sur cette figure, la couleur des flèches à
l'intérieur de la liaison de poisson pour les
couleurs des flèches de la série chronologique.
Cela nous dit exactement comment nous allons
partir d'une forme à une série de temps
Figure 4 I) A l'intuition visuelle de la
conversion d'une forme bidimensionnelle à une
"série temporelle" unidimensionnelle. Ii) deux
formes qui sont similaires à l espace forme sera
également similaire dans la forme de séries
chronologiques. III) Ici, nouscomparer un poisson
(Priacanthus arenatus) à partir d'un Cubain de 40
ans manuscrit à un poisson connexe (Priacanthus
de hamrur) à partir d'un manuscrit publié en 1899.
Minimalisme aide Dans ce cas, les numéros sur
l'axe X ne veut rien dire, ils sont supprimés
12
Linking helps interpretability II
In this figure, the color of the arrows inside
the fish link to the colors of the arrows on the
time series.
This tells us exactly how we go from a shape to a
time series.
Note that there are other links, for example in
II, you can tell which fish is which based on
color or link thickness linking.
Minimalism helps In this case, numbers on the
X-axis do not mean anything, so they are deleted.
13
Where to find Papers?

• I have given some suggestions on the next page.
• You can also find papers by..
• Looking through the proceeding of major
  conferences, SIGKDD, ICDM, SDM, ICDE, NIPS, VLDB,
  SIGMOD, ICML
• Topic driven (outside of data mining) If you are
  interested in astronomy or insect (or anything)
  there is probably a data mining paper on it.
• Data mining Topic driven I would like to learn
  about DM for/with social networks, active
  learning, anytime algorithms, privacy
  preservation, mobile devices, bloom filters,
  personalization, deep learning, the reasonable
  effectiveness of data, fraud detection,
  anti-terrorism, advertising 
• Person Driven You know of an interesting
  researcher
• I want to see interesting papers, sometimes bad
  papers can be interesting -)
• Please run your idea by me first

14
Suggested Papers Toward an Architecture for
Never-Ending Language Learning.A. Carlson, J.
Betteridge, B. Kisiel, B. Settles, E.R. Hruschka
Jr. and T.M. Mitchell. In Proceedings of the
Conference on Artificial Intelligence (AAAI),
2010 "Predicting Human Brain Activity Associated
with the Meanings of Nouns," T. M. Mitchell, S.
V. Shinkareva, A. Carlson, K.M. Chang, V. L.
Malave, R. A. Mason, and M. A. Just, Science, 320,
1191, May 30, 2008. DOI 10.1126/science.1152876.
Fei Xu, Ravi Jampani, Mingxi Wu, Chris
Jermaine, Tamer Kahveci Surrogate ranking for
very expensive similarity queries. ICDE 2010
848-859 Mingxi Wu, Chris Jermaine Guessing the
extreme values in a data set a Bayesian method
and its applications. VLDB J. 18(2) 571-597
(2009) Jayendra Venkateswaran, Tamer
Kahveci, Christopher M. Jermaine, Deepak
Lachwani Reference-based indexing for metric
spaces with costly distance measures. VLDB J.
17(5) 1231-1251 (2008) Pedro Domingos The Role
of Occam's Razor in Knowledge Discovery. Data
Min. Knowl. Discov. 3(4) 409-425 (1999) Nilesh
N. Dalvi, Pedro Domingos, Mausam, Sumit K.
Sanghai, Deepak Verma Adversarial
classification. KDD 2004 99-108 Smriti Bhagat,
Amit Goyal, Laks V. S. Lakshmanan, Maximizing
Product Adoption in Social Networks. In WSDM
2012. Text Classification from Labeled and
Unlabeled Documents using EM, K. Nigam, A.
McCallum, S. Thrun, T. Mitchell, Machine
Learning, 39, 103-134, 2000. Marie
desJardins, Eric Eaton, Kiri Wagstaff Learning
user preferences for sets of objects. ICML 2006
273-280 Clustering with Instance-level
Constraints Kiri Wagstaff and Claire
Cardie. Proceedings of the International
Conference on Machine Learning (ICML), p.
1103-1110, 2000. Juha Reunanen Overfitting in
Making Comparisons Between Variable Selection
Methods. Journal of Machine Learning Research 3
1371-1382 (2003)
15
Sign up sheet

• Will be on my door.
• You can ONLY sign up in numeric order.
• Just sign up your name, and the first five words
  of the paper title

16
January 2014 January 2014 January 2014 January 2014 January 2014 January 2014 January 2014
Sunday Monday Tuesday Wednesday Thursday Friday Saturday
1 2 3 4
5 6 7 First day briefing 8 9 Sample Presentation 10 11
12 13 14 No meeting 15 16 1) -------- 2) 17 18
19 20 M L King Day 21 3) -------- 4) 22 23 5) -------- 6) 24 25
26 27 28 7) -------- 8) 29 30 9) -------- 10) 31
17
February 2014 February 2014 February 2014 February 2014 February 2014 February 2014 February 2014
Sunday Monday Tuesday Wednesday Thursday Friday Saturday
1
2 3 4 11) -------- 12) 5 6 13) -------- 14) 7 8
9 10 11 15) -------- 16) 12 13 17) -------- 18) 14 Valentine's Day 15
16 17 Presidents' Day 18 19) -------- 20) 19 20 21) -------- 22) 21 22
23 24 25 23) -------- 24) 26 27 25) -------- 26) 28
18
March 2014 March 2014 March 2014 March 2014 March 2014 March 2014 March 2014
Sunday Monday Tuesday Wednesday Thursday Friday Saturday
1
2 3 4 TBA 5 6 TBA 7 8
9 10 11 TBA 12 13 TBA 14 15