Bilan de la deuxième journée de la conférence AXMEDIS
Prises de notes
Suite .. et fin.
Et oui, la troisième et dernière journée de la conférence s'achève. Je pense qu'un petit bilan s'impose, tant sur les idées intéressantes présentées durant ces trois, que les différents et bien sur
AXMEDIS. Pour tout ça, c'est la section Future Work.
Mercedi matin (Industrial papers session)
9h50) Interactive Contents Protection in Digital Terrestrial Television
Outline
A) Interactive Contents Protection in Digital Terrestrial Television
B) MHP: Middleware for interacive television
security in MHP
code obfuscation
C) expermential Tests & Results
D) Conclusions & future works
A) Objective
Which DRM?
- Video & Audio : CAS (Conditional Acccess System)
- Interactivity?
Protection of interactive contents & effects of the solution in term of perf.
CAS
- keys & billing system
- Encryption with CSA
B) MHP
- API, JVM, Security Model ; middleware between JVM & application
> PRF: Permission Request File: restricted access
- MHP-PKI: a backbone to protect contents
> Authentication, Integrity & Authorization
> No protection against reverse engineering
> copyright protection (?) => code obfuscation
Obfuscation
several obfuscant transformation that can be classified
- layout transformation: remove comments, scramble identifiers, chaange formatting
> one way, impossible to reverse
- data transformation: data structure
- control transformation: application logic harder to follow
- reventive transformation:
C) Tests
- MHP: MHP Teste
- Obf: DashOPro ; Klassmaster
- Decoders: ADVI-iCam 2000T
- Basic Test (remotecontrol, ...), Graphic Test, Network Test
- 18Mbps ; 1Mbps of bandwith for interactivity
- Test:
> functionality (obfuscated vs normal)
> resilience: strenght of the code protection
> cost
Results
- functionality: fully functional
- layout transformation: size reduced of 20%
- obfuscation-decompilation: human understanding is diffucult
> decompiler not able to revuild the correct tree
(> ... & anti obfuscator?)
- graphic test:near form the original (rectangs / oval, lines, text /sec animatation FPS) (max:4% of loss)
10h20) A framework for digital watermarking next generation media broadcasts
Issues
- Creators of contents must be compensated for the use of their works
- in some juridiction,
- rights agencies are responsible for:
> monitoring the use of copyright
> calculation notificiation of royalties
> colllection & redistribution of royalties
>existng solution: manual
Actors
- Rights Entitiy
- Monitoring Agency
- Broadcaster
General Framework
PKI point of view
- Centralized PKI (common trust)
Watermarking process
- Filestream (FS) => Hash Sum (HS)
HS + digital Signature => Nyberg Rueppel Algorithm => DSHS
- DSHS+FS => watermaking
- DSHS => Hash Table
Monitoring agency can compare hashtable information & retreived watermaking (DSHS => DS +HS)
- DS: who?
- HS: check in hashtable
watermaking of content for each user
- security = dissuasion
- file can be used no matter what happen
Watermarking security
- possible to alterate the watermark?
- illegal distribution:
someone stole the watermarked file
> comm. chanel
> at user
=> user is accused
10:55) dContentWare Project
A) Project Assumptions & primary goals
- Research on semantic web & multimedia distribution technologies
- needs also of protection (DRM)
B) dContentWare Technical Infrastructure
- JeromeDL: digital Library ; semantic web & social semantic collaborative filtering
- AXMEDIS
=> mix the two of them... dContentWare Portal
C) dContentWare Business Model
- working in JermoeDL
> RDB => RDF by D2RQ mapping specification
> Semantic querries on top fo sesame 1.7 JeromeDL repository
> NL queries ; RDF queries
- Bookmarking & annotation
SSCF: Social Semantic Collaborative Filtering
SIOC: Semantically Interlinked Online Community
D) Business Model
- for working of jeromeDL interoperability & axmedis integration
- prototyping use case
> source of digital contents: Hystorical Archive of Laterza
> Digital Contents Migration: laterza RDB to JeromeDL/Sesame RDF via D2RQ
> Domain Taxonomy Specification: Laterza domain taxonomy / Jonto Taxonomy
E) Conclusion
- modern digital lib : from to distriution centers to konwledge promoters of digital contenst
- dCW: way to overcome interoperatbility problem ; innovative business using AXMEDIS
11h24) Telecare & Telemedia with DRM support
Issues
Evolution in customers reqs
> open user's issues
> possible available app
> new solution
Market situation
> DTT in 2012
> low cost CPE
> Data Link
> revolution; interactivity
STB Evolution
- Guide line in italian DTT: HDBOOK
> Video H264 & audio AAC
> LAN wired / wireless connectivity
- HD-Book: no specs about IPTV
> identify DMA (Digital Media Adapter)
> Elsagdatamat ARIES STBs cakked hybrids
- Elsag Datamat solution
> new generation boxes to work on DTT transmission
> based on ST7/109 processor & possible of NAS (Network attached Storage)
> ...
- Evolving architecture
> MHP architecutre
- available coding
Where it works
> Rai.it, DTT Toscana MHP, ..
> deployment issues (specific cases, different types of users)
Mercedi Après-midi
14h) Evaluating Recommender Systems
A) Motivations
- 93% information : digital format
- 161 hexabytes data
- 1 billin internet users
- 213 queries each day
Recommender systems
content-based approach
collaborative filtering approach
- Techniques
> user & item based
- alogirmths
> memory & model based
- collecting preferences
> explict (ask user) & implict rating
Recommender systems
CNN, last.fm, igoogle, amazon.com, ...
B) Anatomy of the long tail
- total inverntory
- the new growth market
C) Recommendation System Datasets
- MovieLens datasets
explicit rating
- Jester
explict rating ; very heigh densistes
- Netfix
explict rating ; for DVD
- BookCrossing dataset
implict rating; for book
D) Results & discussions
Evaluation Metrics
- Mean Absolute Error: quality of the preidction
- Recall: recommendation quality ; R = Nrn/Nr
- Coverage: neighborhood quality ; C = (Nr - Nrn) / Nr
Comparison:
- Coverage: MovieLens & Jester are equivalent
E) futur work
privacy issue:
user anonymity ; encryption
multicriteria rating systems
exemple:
-Yahoo: Movies!
story, acting, direction & visuals
-CircuitCity (restaurants); ZagatSurvey
14h24) Ontology based matchmaking approach for context aware recommendations
Outline
- Recommendation systems
- ontology base
- matchmaking
- context-aware
EUREKA-Celtic project: MOVIES
A) General view
- context profile
- content profile
- user profile
B) 5 ontologies
user ontology: user as a person, its interest, ...
> interest ctxt: context ontology
> subject of interest: context & content ontology
C) Matchmaking
- goal: compute a matching coefficient between a user and a content
- 2 kinds of match
> inferred from user profile
> explicit ineterest & non interest
for a category ; a specific content or other resources
- matching formullae
> Categories matching
semantic distance between categories
> global matching
D) Conclusion
- Specifites
> semantic reasong
> exprisivity of complex interests
14h51) An evaluation methodology for collaborative recommender systems
- slection of the evalution method: relevant and minimal
- complementary of the first presentation, selection of the dataset
A) Introduction
- not focued on algorithm
- Algo: SVD, cosine item-based, naive bayesian, networks
- performance evaluation methodologies
B) issue
- clear description of the methods used for perf eval & model comparison?
- diff dataset partition & evaluation metrics => divergent results
- netfix: only hold-out & RMSE ; but how netfix is suited?
C) Objectives
- compare diff. algorithms
- how often the user watches the TV => length of user profile
- if the user preferes blockbusters movies or not => user pref vs (un)popular movies
- User Rating Matrix ; URM
Implicit URM
- Error metrics: MSE, RMSE, MAE ; only for explicit datasets
- Accuracy metrics: recall, precision, fallout, F-measure; Implict & Explicit datasets
D) Evaluations
- Hold-out: not adapted for new users
- leave-on-out:
- k-fold partitioning approach
training part : k-1 partitions ; test : 1 partition
Exemple of impact of the algo:
|
RMSE
|
0,9
|
1,6
|
2,7
|
|
Recal
|
1%
|
8%
|
17%
|
|
F-Measure
|
|
|
0,28
|
|
Results
|
bad
|
middle
|
good
|
|
|
netfix
|
cosine
|
SVD
|
Proposed methodology:
a) statiscal analysis
b) optimization of the model paramters
c) computation of the recall for te top N recommendations
E) Recommendation system architecture
- Inputs:
> user rating (uRM)
- batch processing => real-time recommendation => business rule
Real-Time Recommendation: model dependant
Multi-model recommender engine
> best of all as it combines different alg, depending of the ctxt & dataset
16h10) A Standard-Based Approach on the use of Contextual Information for the Adaptation Authorisation
key subjects:
- standard based
- contextual
- adaptation authorisation
A) Ctxt & motivation
- Universal Media Access (UMA)
> ubiquitous access to and consumption of multimedia content
- DRM for governed multimedia scenarios
B) VISNET-II European project
- Themes:
> coding
> processing -> AVSearch
> Security -> DRM
C) Approach to adaptation
- adaptation engine stack (AES)
feedback, exchange of capabilites, adpatation decision
- adaptation decision engine (ADE)
athorisation
- adaptation authorisation (AA)
- context provider (CxP)
D) State-of-the-art standardisation
there is neither agreement on the definition nor on the representation of cotext
exemple:
- Mobile Alliance Forum - User Agent Profil (UAProf)
-W3C - Composite Capabilites / Preferences Profile (CC/PP)
- ISO/IEC - MPEG-7 & MPEG-21 DIA & MPEG-21 for DRM: part 4, 5,6, 15 (Event Reporting) & 19
MPEG-21: adaptation ; part 7
E) DRM & Adaptation
integrate DRM with adaptation:
<r:allConditions>
<dia:permittedDiaChanges>
<dia:ConversionDescription>
..
</></>
<dia:changeConstraints>
..
</>
</r>
Types & representation of context
- MPEG 21 DIA Usage Environements Description Tools
> Natural environement characteristriqucs
Profiling: User Profile, Natural Environment Profile
F) Use of ... for ...
- 1st step: extraction of conditions contained in licencs
- 2nd step: get information from the user, the environment and the system itself (context) to check conditions (-> authorisation)
G) Conclusion
- lack of standardised models for the use of contextual information
- DRM context-based adaptation
- work & morel complex adaptations -> Authorisations Profiles
- concrete scenarios
Future Work
J'espère faire, quand j'aurais le temps :
- un compte rendu digne de ce nom d'AXMEDIS 2008
- un rapport sur les notions intéressantes présentées lors de la conférence. Ce qui signifie un tri des articles & notions présentés !
- un rapport sur les différentes intervenants & point d'intérêts
idem, on verra si cela pourra être public ou non !
Analyse de politiques SELinux
