Page 82 - Special Topic Session (STS) - Volume 3

STS515 Jeremiah D. D. et al.
requisites on mathematics and statistics by looking at some DS course content
at a higher level within the undergraduate programme
As explained in Section 3, Machine Learning is a core subject within a DS
programme and serves a good example course for us to examine what
particular prerequisites and core components are necessary for its successful
delivery.
A Machine Learning course can be taught anywhere in the latter half of a
DS programe. In the New Zealand or Australia context, it can be second-year
or third-year “paper”. Regardless, we look at what mathematical and statistical
prerequisites may be necessary to enable effective teaching or learning of key
Machine Learning concepts and algorithms.
Using the famous “top-10 data mining algorithms” (Wu et al., 2008) as a
starting point, we examine the relevant preliminary subject studies as required
by each of the algorithms. In 2006, ACM KDD Innovation Award and IEEE ICDM
Research Contributions Award winners were asked to nominate key algorithms
across all fields of data mining and machine learning, and the nominations
were then voted by hundreds of the ICDM’06/SDM’06/KDD’06 Technical
Programme Committee members, resulted in the top-10 algorithms listed in
Table 3. Here for each algorithm, the relevant background mathematics or
statistics knowledge as required is estimated, and matched to four generic
courses with code and content outlined as follows:
- MATH100: Entry-level algebra and calculus
- MATH200: Linear algebra, discrete mathmatics, optimization
- STAT100: Basic statistics such as probabilities and tests
- STAT200: Statistical inference
The ticks in Table 3 are made in a rough estimation of a normal delivery of
the algorithm. For instance, for the k-nearest neighbour (k-NN) algorithm its
algorithmic operation is introduced and relvance to density estimation is
hinted, but the connection to EM and Bayesian inference is not necessarily
discussed (which then would require STAT200). Also, if a “200” option is ticked
for an algorithm the “100” option will be omitted.
As seen from Table 3, it seems that the top-10 algoirthms can be delivered
effectively without too much mathematical or statistical requisites.
On the other hand, DS is a discipline that undergoes rapid advances. To reflect
the landscape of R&D a decade latter than the top-10, we may have a new list
of key algorithms as given by Table 4. Clearly the algorithms have become
more advanced, bearning complexities that require deeper mathematical or
statistical understanding. Hence we come to the conclusion that both
MATH200 and STAT200 are indispensable cores of a proper DS programme
(that teaches Machine Learning effectively).
Arguably, we can adopt a similar approach to map out the core requisites
of DS programmes by looking at other computing and statistics courses. This

71 | I S I W S C 2 0 1 9