1) Major Challenges in Non-Interactive Differential Privacy: There are

1) Non-Interactive Differential Privacy
There are two ways by which we can implement the differential data privacy there are interactive
and non-interactive. In non-interactive settings we do not publish the data directly instead an
interface is built through which the users can query the data and there is no directly access. It is
quite similar to k-Anonymity wherein we change the data before publishing it. Once the data is
published the users request query to the database source which has statistical database through a
secure interface like a firewall, then the query is processed and once the requested query result is
obtained it is given back to the user. In non-interactive method we do not reveal the identity of the
user by which we can secure the user’s privacy. The other advantage we have in using noninteractive
setting is we can overcome the implementation delays which we face in the interactive
settings. Therefore, it is preferred over interactive settings as it is even economical than the
interactive settings.
Major Challenges in Non-Interactive Differential Privacy:
There are certain drawbacks of differential privacy we will discuss them as follows:
• When we publish the data, we will have to publish the entire data, if we want to only change
the certain data which has relevant information and publish small amount of data then we
might risk privacy and if we go ahead to publish entire data in case of sparse data the
amount of data is really large and cost is very high.
• In order to provide better utility and privacy the cost will go up. On the other hand, if we
want to publish data by only sampling small amount of data then we risk privacy leak. The
data which might contain actual information is easily identified and might come under
attack and get compromised.
• The other issue we face is even if the security is effectively provided to our original data
the quality of the original data might get affected due to addition of excessive noise to the
original data. In other to provide complete security the quality of the original data tends to
Privacy Issue’s: When we study regarding wearable devices there are certain security concerns
which surround them we are going to state them as follows:
• From a customer point of view if I am using a device e.g.: Fitbit which tracks my daily
health activity I expect important information concerning my health to be given to me
safely without compromising my personal data. These devices come with inbuilt global
positioning tracking system which can track the current location of the user. If there is any
attack on the security system of the device the personal data of the user such as their current
location gets compromised.
• On the other hand, the hospitals or the various health care institutions collect this data and
analyze it. The data should be securely stored and access to this data should be carefully
granted. As the value of this data is really high these health care institute face threats such
as hacking attacks or virus infection into their security systems.
• Also, once the data is collected on the device it has to be transferred to the health care
institutions via a secure channel. If we transfer the data through the Bluetooth then security
of the mobile might be at risk possibly a target to the hacker.
• Usually the data on mobile or other devices is encrypted for safety but in case of wearable
devices there is no encryption of data there are mostly third-party application and some
companies do not follow required security standards due to which customers personal data
might be under the risk.
• The other risk we face is of the signal interception suppose we have a wearable device
which is connected to our personal mobile through the network of any private organization
then it becomes an easy job for the hacker to hack into the security system of the private
organization due to known security flaw in our wearable device.
Candidate Methods to Address the Privacy Issues:
• Before we purchase any wearable device the best way to ensure privacy is by knowing the
security features of the device only if the device meet the security standards we can go
ahead with it also the manufacturer provides important security updates the customers need
to follow them and update their devices in time.
• The data collected by the health care organizations should be used only to treat and provide
good medical attention to the customers. Also, the organizations should have right
infrastructure to support the data which is collected also the network security should be
high such that any security failure are immediately reported.
• We need to establish a secure channel in order to transfer data such that it is securely
transferred to the storage system, because when we transfer the data there is a high risk of
it to be intercepted and leaked therefore a proper channel should be established. Also, it is
suggested that the data can be transferred securely through the physical channel rather than
the wireless medium. The transfer of data should be administered in order to prevent any
• The wearable device should follow design module which supports encryption so that the
data is encrypted and secured. These health care devices should always follow frameworks
which provide encryption to the personal data.
• If the signals from the wearable devices are picked up by any other source other that the
authorized organization then the security alarms must be in place to seize the transfer of
the remaining information and secure the data. Certain specific protocols should be
designed to securely transfer the data such that the risk of interception by the hackers is
References: Collection and Processing of Data from Wrist Wearable Devices in
Heterogeneous and Multiple-User Scenarios.
Benefits of Differential Privacy:
In order to preserve the privacy if we first look into the k-anonymity, it is a technique where
we de-identify or hide the data. In order to hide the data, we group them into similar types of
data and publish large amount of data such that our original data is hidden.
But simply hiding the data is not sufficient we still risk the sensitive information. The hidden
data is still vulnerable in order to overcome this we also need to diverse the data. Therefore,
we move on to the l-diversity approach which is similar to the previous technique but here
the data groups are well defined. It tries to achieve diversity of data by using minimum
amount of data generalization. But in case we have data of high importance it is not enough
to only diverse it. We have to note that it is very important information which is prone to
attack. Hence even if it overcomes the flaws of k-anonymity we still have not achieved the
perfect privacy solution. Taking note of these drawbacks, the differential privacy approach
was introduced which helps in overcoming the above flaws as well as provide better privacy
solution to our data.
In differential privacy the data related to an individual is protected by changing it before it
gets published. It is stronger than the previously mentioned techniques and also can face
threats to the data. The concept is clear when the user request for information the original
data is changed using a random algorithm and the result is delivered. The output is nowhere
related to original data thereby providing security. Also, other advantage is it is not specific
to any type of data, we can say that it is a general approach to provide security to our data.
I have selected the following paper to critique:
A Random Matrix Approach to Differential Privacy and Structure Preserved Social
Network Graph Publishing Faraz Ahmed, Rong Jin and Alex X. Liu Department of
Computer Science and Engineering Michigan State University East Lansing, Michigan,
USA {farazah, rongjin, alexliu}@cse.msu.edu
Basic Idea:
In this paper a technique is developed to provide privacy combining the concept of random
matrix with differential privacy. The attacks on social network platforms have motivated in
this direction. The idea is to project the data as matrix and then perturb the matrix with
random noise. Due to random projection the top eigenvectors are preserved. By random
projection we can achieve differential privacy by small random perturbation. The main aim is
to achieve the best tradeoff between privacy and tradeoff.
• The technique is efficient as there is no data storage involved hence the data storage is
not an issue.
• Along with good privacy the quality of the data is also preserved as we only introduce
small amount of random noise.
• This is considered the best way to handle the large amount of data.
• It is economical as it is cost efficient technique.
• We need a large network or large amount of data it works effectively only for the larger
data sets compared to smaller sets.
• We require more memory utilization as the amount of data is large more memory is
• The time taken for this technique is more as it is a longer process than a normal
differential privacy procedure.
• The data is treated as same even the sensitive data is treated like any ordinary data.