Multimedia Retrieval (HS21) at University Of Basel | Flashcards & Summaries

Select your language

Suggested languages for you:
Log In Start studying!

It looks like you are in the US?
We have a website for your region.

Take me there

Lernmaterialien für Multimedia Retrieval (HS21) an der University of Basel

Greife auf kostenlose Karteikarten, Zusammenfassungen, Übungsaufgaben und Altklausuren für deinen Multimedia Retrieval (HS21) Kurs an der University of Basel zu.

TESTE DEIN WISSEN

What is the retrieval problem?

Lösung anzeigen
TESTE DEIN WISSEN

Given

  • N documents
  • Query Q of user

Return

  • A ranked list of k documents which match the query sufficiently well. Rank with respect to relevance of the document to the query
Lösung ausblenden
TESTE DEIN WISSEN

Why do we need information retrieval?

Lösung anzeigen
TESTE DEIN WISSEN

We are living in a digital world with exponential information growth. Enormous amounts of data. The price of storage drops by a magnitude, every 4 years, but companies spend the same amount of money on storage -> so much more data. 

We should avoid drowning in the big data lake!


Lösung ausblenden
TESTE DEIN WISSEN

Why can't we use low level features for applications such as object recognition or motion detection?

Lösung anzeigen
TESTE DEIN WISSEN

Signal information is too low level and too noisy to allow for accurate recognition of higher-level features such as objects, genres, moods, or names.

Feature extraction based on machine learning abstracts lower level signal information in a series of transformations and learning steps as depicted below. The key ingredient of a learning approach is to eliminate noise, scale, and distortion through robust intermediate features and then cascade one or many learning algorithms to obtain higher and higher levels of abstractions.

Lösung ausblenden
TESTE DEIN WISSEN

​What is imputation of missing values?

Lösung anzeigen
TESTE DEIN WISSEN

Imputation of missing values requires an algorithm to replace (estimate / guess) missing data with
substituted values. The algorithm must provide a
prediction for the missing values.

~ incomplete data ~

Lösung ausblenden
TESTE DEIN WISSEN

If someone is looking for an apartment in Basel with 1BR <1200CHF, which search evaluation would we use?

Lösung anzeigen
TESTE DEIN WISSEN

Recall! We want to see all possible apartments and probably adapt the query if our original query was not helpful (e.g. go outside the city center a bit more, also see apartments that are 1250CHF, etc.)

Lösung ausblenden
TESTE DEIN WISSEN

Why do we do performance evaluation?

Lösung anzeigen
TESTE DEIN WISSEN

We need to understand the performance of an approach to determine the best way of searching.

Lösung ausblenden
TESTE DEIN WISSEN

What is clustering?

Lösung anzeigen
TESTE DEIN WISSEN

Clustering divides a set of inputs into groups.

The groups & number of groups are not known beforehand. (unsupervised learning task)

~ human genetic clustering, market segmentation ~

Lösung ausblenden
TESTE DEIN WISSEN

How are dense assessments generated?

Lösung anzeigen
TESTE DEIN WISSEN

Contestants of a competition crowdsource to rate the relevance of documents that were returned to (other) contestants.

Lösung ausblenden
TESTE DEIN WISSEN

Which search paradigms do we look at in the course?

Lösung anzeigen
TESTE DEIN WISSEN
  • Key-word based search: User enters keywords. Works fine for text (or spoken audio), other media types suffer from semantic gap between keywords and signal information
  • Similarity search: User provides examples of what the result should look like. Have to define "similarity"
  • Combined search: Merges capabilities of attribute based queries (e.g., predicates), keywords,
    and examples to match. We need additional models/algorithms to find overall retrieval order
Lösung ausblenden
TESTE DEIN WISSEN

What measure would we use to compare several search engines on typical queries?

Lösung anzeigen
TESTE DEIN WISSEN

Mean Reciprocal Rank (MRR). We care about the first results more than the later ones, so we have to take ranking into account.

Lösung ausblenden
TESTE DEIN WISSEN

What is density estimation?

Lösung anzeigen
TESTE DEIN WISSEN

Density estimation (probability mass function estimation) is the construction of an estimate of an underlying, unknown probability density function.

(e.g. predict the density underlying an observed histogram)

~ age at death in countries ~

Lösung ausblenden
TESTE DEIN WISSEN

What is semi-supervised learning?

Lösung anzeigen
TESTE DEIN WISSEN

Semi-Supervised Learning is a special case of supervised learning. The algorithm is presented
with features and targets, however, some features or targets are missing (incomplete observation) in the training data. Depending on the task, the algorithm must either complete the missing features or predict targets for newly presented data sets.

Lösung ausblenden
  • 45788 Karteikarten
  • 946 Studierende
  • 23 Lernmaterialien

Beispielhafte Karteikarten für deinen Multimedia Retrieval (HS21) Kurs an der University of Basel - von Kommilitonen auf StudySmarter erstellt!

Q:

What is the retrieval problem?

A:

Given

  • N documents
  • Query Q of user

Return

  • A ranked list of k documents which match the query sufficiently well. Rank with respect to relevance of the document to the query
Q:

Why do we need information retrieval?

A:

We are living in a digital world with exponential information growth. Enormous amounts of data. The price of storage drops by a magnitude, every 4 years, but companies spend the same amount of money on storage -> so much more data. 

We should avoid drowning in the big data lake!


Q:

Why can't we use low level features for applications such as object recognition or motion detection?

A:

Signal information is too low level and too noisy to allow for accurate recognition of higher-level features such as objects, genres, moods, or names.

Feature extraction based on machine learning abstracts lower level signal information in a series of transformations and learning steps as depicted below. The key ingredient of a learning approach is to eliminate noise, scale, and distortion through robust intermediate features and then cascade one or many learning algorithms to obtain higher and higher levels of abstractions.

Q:

​What is imputation of missing values?

A:

Imputation of missing values requires an algorithm to replace (estimate / guess) missing data with
substituted values. The algorithm must provide a
prediction for the missing values.

~ incomplete data ~

Q:

If someone is looking for an apartment in Basel with 1BR <1200CHF, which search evaluation would we use?

A:

Recall! We want to see all possible apartments and probably adapt the query if our original query was not helpful (e.g. go outside the city center a bit more, also see apartments that are 1250CHF, etc.)

Mehr Karteikarten anzeigen
Q:

Why do we do performance evaluation?

A:

We need to understand the performance of an approach to determine the best way of searching.

Q:

What is clustering?

A:

Clustering divides a set of inputs into groups.

The groups & number of groups are not known beforehand. (unsupervised learning task)

~ human genetic clustering, market segmentation ~

Q:

How are dense assessments generated?

A:

Contestants of a competition crowdsource to rate the relevance of documents that were returned to (other) contestants.

Q:

Which search paradigms do we look at in the course?

A:
  • Key-word based search: User enters keywords. Works fine for text (or spoken audio), other media types suffer from semantic gap between keywords and signal information
  • Similarity search: User provides examples of what the result should look like. Have to define "similarity"
  • Combined search: Merges capabilities of attribute based queries (e.g., predicates), keywords,
    and examples to match. We need additional models/algorithms to find overall retrieval order
Q:

What measure would we use to compare several search engines on typical queries?

A:

Mean Reciprocal Rank (MRR). We care about the first results more than the later ones, so we have to take ranking into account.

Q:

What is density estimation?

A:

Density estimation (probability mass function estimation) is the construction of an estimate of an underlying, unknown probability density function.

(e.g. predict the density underlying an observed histogram)

~ age at death in countries ~

Q:

What is semi-supervised learning?

A:

Semi-Supervised Learning is a special case of supervised learning. The algorithm is presented
with features and targets, however, some features or targets are missing (incomplete observation) in the training data. Depending on the task, the algorithm must either complete the missing features or predict targets for newly presented data sets.

Multimedia Retrieval (HS21)

Erstelle und finde Lernmaterialien auf StudySmarter.

Greife kostenlos auf tausende geteilte Karteikarten, Zusammenfassungen, Altklausuren und mehr zu.

Jetzt loslegen

Das sind die beliebtesten StudySmarter Kurse für deinen Studiengang Multimedia Retrieval (HS21) an der University of Basel

Für deinen Studiengang Multimedia Retrieval (HS21) an der University of Basel gibt es bereits viele Kurse, die von deinen Kommilitonen auf StudySmarter erstellt wurden. Karteikarten, Zusammenfassungen, Altklausuren, Übungsaufgaben und mehr warten auf dich!

Das sind die beliebtesten Multimedia Retrieval (HS21) Kurse im gesamten StudySmarter Universum

Multimedia

Universidad Autónoma de Madrid

Zum Kurs
Multimedia-Technologie

Hochschule Darmstadt

Zum Kurs
Multimedia

Technische Hochschule Aschaffenburg

Zum Kurs
Multimedia Systeme

Universität Duisburg-Essen

Zum Kurs
Multimedia-Technologie

Hochschule Darmstadt

Zum Kurs

Die all-in-one Lernapp für Studierende

Greife auf Millionen geteilter Lernmaterialien der StudySmarter Community zu
Kostenlos anmelden Multimedia Retrieval (HS21)
Erstelle Karteikarten und Zusammenfassungen mit den StudySmarter Tools
Kostenlos loslegen Multimedia Retrieval (HS21)