1. Consider the quasi-identifier {job,birth,postcode}.
(a) What is the highest k for which this data is k-anonymous. 2-anonymous
(b) Describe one possible privacy attack on this data Homogeneity attack – If I know from other background information that someone lives in postcode 4350 and was not born in 1975, I can conclude they have HIV.
2. Consider the quasi-identifier {gender,date of birth,zipcode}. Apply generalisation to the following table to make it 3 anonymous.
Copyright By PowCoder代写 加微信 powcoder
One possible answer. We would first get rid of name attribute before doing any further processing, we then just work with the quasi identifiers and sensitive (attribute)
3. Consider the quasi identifier {Age,Zip} for the table below.
(a) What is the highest k for which this data is k-anonymous? 3-anonymous
(b) What is the highest l for which this data is l-diverse? 2-diverse
(c) Describe one possible privacy attack on this data One possibility: Background attack. For example, if I know that someone aged 48-55 in postcode 92*** doesn’t have Cancer, I can conclude they have Obesity
4. In the context of providing differential privacy:
What is global sensitivity G? What is the privacy budget k?
How does the G/k ratio affect the noise level?
Global sensitivity is evaluating the maximum possible change in query output due to a presence of a single record. The privacy budget de- termines how close the query result for a database with the record is expected to be compared to query result for a database without the record. For smaller k, or larger global sensitivity, more noise will be added to the query result.
5. Consider a survey that collects two values from the respondents, e.g., marital status and sex.
Consider a query that takes the survey database as input and outputs a pair of counts (CountNumberFemale,CountNumberMarried). How much can adding or removing an individual affect the output? What is the global sensitivity?
Consider a query that takes the survey database as input and outputs the quadru-
plet of counts (CountMaleMarried,CountMaleSingle,CountFemaleMarried,CountFemaleSingle). How much can adding or removing an individual affect the output? What is the
global sensitivity?
In the first case
Adding or removing any individual can affect the count of each col- umn by maximum 1. The maximum difference a single record can make query is 1+1=2
In the second case
Adding or removing any individual can affect the count of each col- umn by maximum 1. If it affects the count of some column by one, then it will affect the counts of other columns by 0 (since the columns are mutually exclusive, you can’t have a 1 in both columns)
The maximum difference a single record can make To F is thus 1+0+0+0=1
Other Previous Exam Questions
Exam 2018 – Question 4
University X is planning to build a recommender system for its students. Based on subjects they have enrolled in, the system will recommend new subjects they might consider studying in the future.
A table showing a fragment of the data input to the system is below. The columns correspond to the codes of all the subjects in the University handbook. Rows correspond to students and whether they have enrolled in a subject (“Yes” if they have previously enrolled and “-” otherwise). The dataset covers the period 2010-2017, with 100,000 students and 3000 subjects. It is proposed that subject recommendations should be made using user based collaborative filtering.
PersonName Sub ject1
Subject2 Subject3 Subject4 Subject5 … – – – – … Yes – Yes – … Yes – – – … … … … … …
Explain three challenges for making this recommendation approach effective.
-cold start – how to handle new students with no history of subjects, eg. starting in 1st year
-how to avoid over-recommending popular subjects and under recommend- ing rare ones
-how to incorporate domain knowledge about degree requirements and com- pulsory subjects
Need to make three plausible points. Each point needs to be clear. Points should be distinct from each other
Exam 2019 – Question 3
(a) (1.5 mark) Are these likely to have come from the same dataset? Give two reasons why or why not.
(b) (1.5 marks) Consider a dataset with 10000 rows and 500 features. Give three reasons why we might want to apply PCA while analysing the dataset.
Part a: Yes.
Upper cluster in VAT corresponds to upper cluster in scatter plot
Centre cluster corresponds on VAT to large cluster in the middle of the dataset.
Centre cluster less well defined = higher spread in centre cluster on scatter plot
Well defined section at bottom of centre cluster on VAT corresponds to points around (0,0)
Bottom cluster on VAT corresponds to right cluster on scatter plot Part b
Computational efficiency
Reduce noise
Allow visualisations
Exam 2019 – Question 2b
(3 marks) Consider the following regular expression:
int[1-9]+=[A-Z][a-z]*
Which of the following terms would matched by this regular expression (write down all that apply):
int2=A int3=Abc int44=Azz
程序代写 CS代考 加微信: powcoder QQ: 1823890830 Email: powcoder@163.com