Paper Review: Automated Hoarding for Mobile Computers Reviewed by: Jim Cai

overratedbeltAI and Robotics

Nov 25, 2013 (3 years and 11 months ago)

67 views

Paper Review:

Automated Hoarding for Mobile Computers

Reviewed by: Jim Cai




The paper proposed a
n

automated
hoarding algorithm
SEER
that takes semantics of
human behaviors into account in determining which files should be hoarded.
It first
introduces l
i
fetime semantic distance of two (not simultaneously opened) files
to be

number of intervening files open between opening these two files.
It then explains
a
gglomerative clustering
algorithm

that
starts from each data p
oints being individual
clusters, gradu
ally combining

clusters with shared
-
neighbor criterion.

The proposed
SEER
algorithm uses lifetime semantic distances between files to measure the
correlations between files.
It then feeds the distance data to

agglomerativ
e clustering
algorithm
to perform h
oarding predictions.

In later sections the

paper also presents

and
attempts to solve

several real
-
world challenges
that could results low performance of
SEER
such as meaningless activities created by find program in UNIX, shared libraries
that are accesse
d by most programs
etc.

The paper finally compared the experimental
results between LRU
-
type of hoarding and SEER and conclude that the later has
outperforms the former in terms of miss
-
free hoard size.


The paper explored

on a novel approach of automatic
hoarding problem.
One renovation
is that it incorporated the user behaviors as an aid in determining the relationship be
tween
files. One can anticipate

that by quantitatively measuring such relationship, it is
easier

to
know which files should be clustered

in one group. Another contribution of this paper is
it provided such a quantitative measure


the semantic distance between pair of files.


SEER is based heavily on assumptions that there is strong semantic locality in user
behavior when accessing files
.
However, this assumption won’t always hold and there is
no control mechanism for irrational users

and buggy programs
.

The algorithm would
have ended up hoarding unnecessary files or missed hoarding critical files due to
malicious user behaviors.


In t
he results section, the paper presented a measurement scale called “miss
-
free hoard
size”. It considered each file equally critical such that missing any of these would be
severe enough to stop the program from operating properly.
However, in reality not e
very
file
in a project
is equally important and missing one critical
library
file cannot be
compared to missing a non
-
critical file
, such as a documentation file
.