Presentation 11/06: A Passive Approach to Sensor Network Localization
Su Kim
Title: A Passive Approach to Sensor Network Localization
Published: In Proceedings of 2004 IEEEmSJ International Conference on
Intelligent Robots and Systems, September 28. October 2,2004, Sendai,
Japan
Please reply to this message in order to write your critique and
summary.
swaroo...@asu.edu
In the paper, the author uses gradient descent to maximize l. l
depends on three events, S, E and T. So, when you find gradient
descent, do you partially differentiate the expression wrt all three
and equate it to zero? Also, this would either give a maxima or a
minima. Is this why the authors state in the end that the gradient
descent may get stuck in local minima?
I am not sure of my reasoning, so kindly correct me if I have stated
something incorrect..
regards,
Swaroop
Ayan Banerjee
that is what is done.
You are write in your reasoning of why the algorithm gets stuck at
local maxima or minima. However there is less chance of their being a
minima as the random variable R is Normally ditributed.
Aravind Krishna
from Swaroop
jun shen
Paper Overview:
Localization is of great interest in sensor network. The paper
presents a kind of novel mechanism for localization. The most
important advantage of the new idea is that the cost of infrastructure
can be greatly saved. Actually, infrastructure is no longer needed. So
the mechanism is called as passive localization. A sensor can locate
its own position by listening to noises in the environment. The paper
proves the correctness of the proposal by setting up a Bayesian
Network and adopting gradient descent to get the maximum likelihood of
the position of the node. Besides, the paper gives the result of
simulations to support the authors' claim.
Key points in critique:
Strength:
1. The passive approach is cost-effective. With this approach, no
infrastructure is needed any longer.
2. The mathematical proof is well-done.
3. The author introduces machine learning into the localization
computation.
Weakness:
1. The error range of the localization can be significantly large if
sound is heard by few nodes.
2. The paper shows no comparisons with any other anchor-based
approaches. Therefore the claim of the authors that passive approach
can work almost on par with anchor-based approach is poorly supported.
3. The inherent weakness of gradient descent is not addressed. The
algorithm of gradient descent is very time-consuming in some cases.
4. The impact of the mobility of nodes is not mentioned. Apparently,
the mobility or movement of the nodes can influence the accuracy of
the localization result.
Improvement Suggestions:
1. Manifold (one machine learning technology) can be employed to
overcome the above weakness 1.
2. One effective way to raise the accuracy of the localization result
is hill climbing search with random start.
On Nov 8, 3:14 pm, "Su Jin Kim" <sujin...@gmail.com> wrote:
> Updated version of swaroop's paper critique
>
> critique_sound.pdf
> 5KDownload