Arjen's bibliography

First? paper that articulates the position that persistence should be orthogonal to type.

The original System R implementation of a relational database management system

Describe the JACOB video retrieval system, using feature descriptions of the keyframes.

CORBA-based middleware infrastructure designed to assist applications with strong synchronization and coordination requirements.

Paper about AMIS, with a central role for the ideas of Mirror.

not read yet

The standard document on OO-DBMSs.

Illustrates the wide scope of building digital libraries. Note: the article is rather vague, addressing too many issues; this way, it clearly illustrates the need of further abstraction, in which databases may play an important role

A describtion of the Virage approach to image retrieval. It confirms the basis of all approximate retrieval is a Feature extraction function and a similarity function.

Retrieving stored video mail using voice recognition and information retrieval techniques. Best paper award.

A firm mathematical treatment of pattern recognition, and the application of neural networks for this task. Good intro on applying Bayesian inference for this purpose.

Explains in detail the extendibility of the Monet database system. Preferred reference about ADT's and Monet.

An introduction to the book, with a.o. a classification of metadata.

L2 cache affects the performance of a join implementation!

Uses 2900 images of Corel photo CD for evaluation; each CD has 100 images of some class.

Monet and GIS applications

Eric Brown's thesis on improving InQuery's performance

Database systems must become more open to retain their relevance as a technology of choice and necessity. Openness implies not only databases exporting their data, but also exporting their services. This is as true in classical application areas as in non-classical (GIS, multimedia, design, etc).

An overview of all literature of learning in probabilistic networks. Very good companian to the more technical Heckerman tutorial. Especially the general introduction to Bayesian networks is very understandable.

Explains all the tricks applied in the monet database kernel to handle objects using MOA.

Survey on this matter

Compound documents in the inference network retrieval model

About Codd's manifesto

Guest editor of special issue; describes motivation for choice systems discussed, and general motivation of development of OO and OR DBMSs.

OQL reference

Part of MeasTex framework

Demo at http://www.ctr.columbia.edu/videoq. Not just static scenes in retrieval, but also actions. Small collection for testing.

An overview of object-orientation in databases, identifying four streams in research, and discussing what should happen next.

Part of MeasTex framework

About the vertical decomposition as in Monet

An invited project review about the development of OQL, discussing the design of the language in detail, including some previous constructs that were removed, and why, and an evaluation of what is good in OQL (and wrong in SQL).

A new algorithm and systematic evaluation is presented for searching a database via relevance feedback.

Paper that presents the RM and its merits.

A perfect example of people thinking multimedia retrieval is completed as soon as we have an extensible database system

I do not have the original paper...

(just in case any existing latex source might "break": this reference used to have key cox.)

Check this reference - part of MeasTex framework

Choose from a set of distributed databases which is most likely to contain an answer.

Some results on logical imaging (in small collections); extended version available from Glasgow website.

Deductive databases, the fixpoint algorithm, and mapping of deductive database programs to SQL programs. IN 516.34 D012

Date's standard textbook on DBMSs.

A detailed study of the impact of objects and type theory on the relational model, including a comprehensive proposal for type inheritance. It identifies clearly where ODMG and alike make (possibly) big mistakes. It reminds you of why we developed databases in the first place.

INQUERY on top of Rdb, using cooperative indexing. See also gu93 and vasanthakumar96.

IN 516.35 d087

This may very well be the same article as [dittrich91]; I do not have a copy of this one...

This seems to me the eldest reference mentioning structural versus behavioural object-orientation.

Discusses the issues in providing OO functionality in databases, clearly written by a database person and not a programming language one. This may have appeared previously in 1986, as dittrich86

Text search in SQL databases, SQL/MM2 extensions, tight coupling of text-retrieval search engines with database through user defined functions in DB2.

The retrieval model and its embedding in Moa

A demo of the Mirror DBMS and its image retrieval prototype

Using a database approach with algebraic query optimization to perform IR querying

Discusses Paivio's dual coding theory

Standard for binding SQL to Java

A short overview of the language features in SQL99.

2nd edition of elmasri89. Standard textbook for databases, good reference for things like data abstraction, data models, etc.

Earlier version of the invited paper in PODS'98

Invited paper. The title says it all!

Demonstrates document-space relevance feedback processing, in a true probabilistic indexing model. That is, the probabilities are the real probabilities as observed in the training data, not some estimate based on vague measures. Also provides a good explanation of the independence assumptions and the formula manipulation underlying most probabilistic retrievel models.

Describes not just the techniques, but also a very small evaluation procedure

Good review of reasons to apply Bayesian networks in IR. They discuss a model to represent dependent topics in an efficient manner, and show a minor improvement in precision-recall. Good paper to study after reading Turtle and Croft.

A detailed identification of the problems associated to using words in systems. From computing reviews: `This is an excellent paper for anyone interested in interface design where users' choice of words is involved, especially those who believe that vocabulary is not a problem.'

An application of the weighting paper in ICDT to IR query weighting.

Audio retrieval based on techniques from speaker identification research. Evaluated against the Musclefish data set.

Part of MeasTex framework

An easier to find introductory paper into the QBIC approach to image indexing and retrieval.

Good paper on weighting. It is very clearly written, and the result is nice. Using this approach, weighting can be added to any scoring rule. A research issue might be to consider how it relates to probability updates in the inference network model.

Good description of the use of Link Matrices. The tech report is even better though. Gives probabilistic interpretation of pnorm operators from vector space model.

More recent description of the PIC matrices and its implementation; still same mistake...

An introduction to the problems of multimedia retrieval through associative retrieval, defining some terms like media objects vs. content-based metadata.

For everybody who is serious about high-quality object-oriented development

They thoroughly collected a data set of 21000 frames from 9 movies, with 200 ground-truth cuts, to evaluate a set of color spaces and frame difference measures for scene segmentation algorithms.

Survey of efficient query evaluation techniques, also pointing out that OO and extensible DBMS-s discourage efficient query evaluation...

Very good description of the real implementation of INQUERY and its probabilistic reasoning model. Research note: compare these with causal independence models (noisy-or etc.) from Heckerman.

Still need to get a copy; discusses a.o. the richness of media objects compared to non-media objects.

Evaluation of color indexing schemes against varying recording situations.

The counterexample given in this paper is wrong. So, Cox's theorem still stands.

Reviews many different multimedia and hypermedia document models

A good description of several causal independence models for Bayesian inference networks, and the associated network rewrites.

Nice tutorial on probability and decision theory in AI.

type extensibility in the starburst system

A very good tutorial to learning approaches and Bayesian network technology

First introduction of Djoerd's LMM (go, Djoerd, go!).

Good overview of video data management.

Describes query optimalization in a heterogeneous environment, on a level above the component data collections.

Something to reference when listing media types.

A clear presentation of Cox's theorem and its impact.

Good explanation of the use of probabilities for uncertainty modelling in information systems, with three cases and the basic mathematical model. IN 516.34 U060

A description of the news-on-demand application as a demonstrator for the Informedia tools.

Describes the relation between inference networks and Phrasefinder

Combination of multiple sources in spoken document retrieval. Best paper award on SIGIR 1996.

The most referenced survey of query optimization in (relational) database systems.

An extensive overview of the limitations of record-based representations of information, with many examples of business applications in which a data model with just records can be trouble.

Discusses in a very quick overview the notion of open implementation and the problems of blackbox abstraction.

Reflections on ODMG by the UniSQL director.

Explains the definition of a new join called fitness join, and discusses its implementation as theta-join.

Very good undergrad level book about information retrieval. A little too much from a system viewpoint, not very good at conceptual level, and attempts to deal with too many aspects varying from file structures through compression to security issues.

I do not have this book...

Overview of ObjectStore, mentions the impedance mismatch

Overview of Starburst

Probabilistic reasoning in a RDBMS. Theoretically quite good, starting on an algebra approach with similar equivalences as relational algebra. Implemented on Dbase IV, not really integrated in the Database Management System though.

Texture based image retrieval. Evaluation on Brodatz set.

The MoCA video abstracting system

Relational databases and text retrieval; some examples of SQL queries. Ignores the benefits of an algebraic system.

A user-interface to specify the combination of several agents, each creating a representation of multimedia data. Important in the design is the open architecture.

IN 516.35 M43

Reports about the MITRE corporation's work on video retrieval. They use an adaption of Text-tiling, and evaluate on a data set of 17 1/2 hours. They only hand-segmented program boundaries, not topic boundaries. They did notice that the program changes that they missed did have visual cues, so a combined algorithm might work better.

Cobra approach to IR. Very nice object-oriented approach, however, little possibility for query optimization and batch processing, and exposing too much of the indices at the physical level to the schema level.

The main description of the FourEyes learning agent for Photobook

The best reference for learning machine learning

Different types of relevance in IR. [ Another motivation for IR in databases, cause these attributes are typically meta information about the documents, that can be modelled in the logical structure of the document. ]

Segmentation of images using texture flow analysis, texture based retrieval, additional learning of similarity measure using neural network, all for the Alexandria Digital Library project at UCSB.

Texture based image retrieval, using Gabor filters. Evaluation on Brodatz set.

..

Overview paper of Cobra; see the thesis for more info.

Very nice paper, about Cora online c-s research library

The overview reference for the FourEyes system

Good introduction or overview of MARS system, first system to apply IR techniques for multimedia retrieval.

modeling documents, like logical vs. content structure, etc.

Marjo's user study on journalists at a Finnish news paper agency.

Describes the approximate retrieval techniques introduced in the QBIC project

Image retrieval op Monet

Demonstrates a significant performance increasement on a standard clustering algorithm by set-orientation.

Overview of several projects at National University of Singapore, especially FACEit and STAR, and the general process of multimedia retrieval.

A critique on the assumption that shared-nothing = all.

Bible of Java programming with CORBA.

Harold's MSc work on feature extraction for similarity search on music.

Also in Readings in database systems, 3rd edition. This paper discusses the role of theory in databases. It also discusses the role of theory in computer science in general, and draws on Kuhn's philosophy of science.

A very extensive overview of imprecision in literature in Databases and AI. Very good introduction into all aspects of reasoning with imperfect information

Text segmentation, using an algorithm based on the LCA approach by Xu.

The ideas underlying Foureyes: inference and learning algorithms for model selection.

Short note on evaluation

Provides a cluster-based probability model, a proof that in infinity it converges to the real probability, a comparison with the EM algorithm and experiments with a.o. classification of images

Orderings in relational algebra. Interesting basis to consider LIST in MOA.

EL e91 r43

They evaluate against the performance of QBIC and measure how much better their method is on the same amount of results

Also TR-MARS-97-08.

MARS and relevance feedback: similarity matching toolkit.

Relevance feedback in MARS (again).

Middleware, application servers, tp monitors

A mathematically better paper about probabilistic inference networks for IR; clearly based on the Turtle and Croft ideas, but better consistent with probabilistic retrieval models.

Reference for relevance feedback in query-space. Both adapting the weighting scheme and adding terms to the query are discussed.

Referenced by Mark, have not read it yet

Nona-trees to search segmented images. Uses wavelet transform. Evaluation on Brodatz collection.

Some experiments on a limitly annotated, huge corpus.

Potential explanation where the logs in INQ's ranking formula originated.

Composition of color in image is taken into account. Measure performance with one query to retrieve the 83 lions from a 3100 image database. Also use a synthesized database for evaluation purposes.

Color-based retrieval in VisualSEEK, Columbia University

Clustering method to improve retrieval, using judgements of domain experts.

Cluster algorithm based on wavelets.

This article discusses information systems as a scientific discipline.

On predator's implementation

A typical user study of multimedia retrieval, however, with a data set of only 7 MS Windows objects.

Relationship of similarity as used in image databases with similarity as studied in cognitive psychology research. Evaluation against Brodatz texture collection.

Mainly deductive databases and multimedia data. Do not address the kind of topics I work on, but focus on modelling of complex objects and query languages over these objects.

Overview of Postgres

A bundle of coordinated articles on metadata for multimedia retrieval; too much introduction, and the theory has clearly not matured, but useful as an overview.

An example of increased effiency using an algebraic approach, in a different domain than `normal' databases.

An explanation of what Stonebraker calls object-relational systems. The concepts are illustrated using Illustra, the commercial version of Postgres. This reference has expired by the second edition of this book, which differs significantly.

The second edition is far more elaborate. It discusses OR-DBMSs in more detail, and architectural issues and optimization are discussed more thoroughly.

Reviews the aspects of imperfect information, as well as frequentist versus subjectivist probabilities, and fuzzy vs. possibilistic vs. probability based approaches. IN 516.34 U060

Proposed task-based evaluation method for CBR techniques.

A reference to the Viper system; they think Mirror

Semantic networks etcetera. Introduction/overview previous research, some relations with logic and lists some key articles about semantic networks and its problems.

NF2 models and its application to IR in databases.

Texture modeling with an autoregressive model. Evaluated on a 135 element texture database with 15 texture classes.

The implementation of the POSTGRES extensible database system

The standard reference for the NF2 data model.

The theory underlying the development of Moa. A convincing argument for rewriting at the calculus level; and, a thorough review of algebraic approach to OO query processing.

A good overview with a focus on techniques.

One of the first (academic) prototype implementations of a relational database management system

Referenced by Mark, have not read it yet

Lists arguments against databases as monolithic pieces of software; argues that database research results must be adapted such that these can be applied in different more component-wise developed systems.

A comparison, among which the vector space retrieval model expressed as an inference network retrieval process

Another explanation of the inference network model for information retrieval. IN 516.34 U060

A survey of techniques for concurrency control, to improve upon performance degradation observed in standard locking.

Presents the ANSI/SPARC model

WK 517.3 T13

The thesis describing the details of INQUERY and some experiments and possible extensions of the framework implemented. Very good insights in information retrieval with inference networks, sometimes a little chaotic though.

An implementation of INQUERY on an RDBMS. No extensibility, just for multi-user, storage, etcetera.

Erik's MSc work, on CORBA, databases and multimedia, developing an improved version of the work described in my IDEAS'98 paper. The Spie'98 paper also covers Erik's work (at a less detailed level).

Hierarchical color-based approach, using wavelet decomposition. Evaluation on four queries against some 3400 image database.

A Bayesian method for scene segmentation, that also is shown to provide a basis for the classification of these segments in genres. The evaluation uses the manual classification of the Internet movie database.

Cool results, applying the Bayesian framework to CBR techniques. It really works!

A first draft of his ideas on logic and IR, formalizing some aspects of information retrieval for use in a logical model.

The PERL bible, covering Perl 5 with extensions.

Parallel query processing in PRISMA

STAR trademark retrieval system, evaluate with user-defined order of expected similarity ranking, for two queries; the ordering is created by 10 users, based on a voting procedure.

Very good introduction, especially into why text retrieval is computationally expensive, Zipf law, etc., includes code; much focus on computational issues in IR and the benefits of compression to handle the indices.

The famous Wordnet thesaurus

The ETH speech indexing system. Describes their test collection.

The ContIndex indexing structure is introduced, that sort of combines classification with indexing. The necessity to combine feature spaces into multimodal feature measures is discussed. The index structure is not much more than the classification tree itself; a neural net is used for clustering. The classes are defined by the user (supervised classification); measures are combined using unsupervised training. [The difference with Mirror is that they first require the user to provide a manual classification; we try to learn this classification only while querying.]

INQUERY's best model for relevance feedback information, called LCA.

Fractal-based clustering of image features for texture based retrieval. Evaluate on Brodatz and MIT's VisTex.

Video segmentation techniques and a very compact evaluation.

Overview article on video segmentation and retrieval techniques, as used by the National university of Singapore.

An extensive overview of ranking formulas known in experimental IR with an evaluation of a large subset of all possible combinations.

Causal independence in inference networks. Nice examples of noisy adder (lottery) and noisy OR (alarms), directly applicable to explaining our model. Also describes a SPI based inference algorithm, Variable Elimination, and a comparison with cluster-tree based methods.