[Apologies for multiple postings]


While deep neural networks have proven their predictive power in many
tasks, there are still several domains where a single deep learning
network is not enough for attaining high precision, e.g., prediction
of subjective concepts such as violence, memorability, etc. Late
fusion, also called ensembling or decision-level fusion, represents
one of the approaches that researchers employ to increase the
performance of single-system approaches.

It consists of using a series of weaker learner methods called
inducers, whose prediction outputs are combined in the final step, via
a fusion mechanism to create a new and improved super predictor. These
systems have a long history and are shown to be particularly useful in
scenarios where the performance of single-system approaches is not
considered satisfactory.

The task challenge participants to develop and benchmark late fusion
schemes. This task would allow to explore various aspects of late
fusion mechanisms, such as the performance of different fusion
methods, the methods for selecting inducers from a larger set, the
exploitation of positive and negative correlations between inducers,
and so on.

*** TASK ***
The participants will receive a data set of real inducers and are
expected to provide a fusion mechanism that would allow to combine
them into a super-system yielding superior performance compared to the
highest performing individual system. The provided inducers were
developed to solve two real tasks: (i) prediction of visual
interestingness (int), and (ii) diversification of image search
results (div).

*** DATA SET ***
ImageCLEFfusion-int. The data for this task is extracted and
corresponds to the Interestingness10k dataset. We provide output data
from 33 inducers, representing visual interestingness predictions for
2,435 images.
ImageCLEFfusion-div. The data for this task is extracted and
corresponds to the Retrieving Diverse Social Images Task dataset. We
provide outputs from 56 inducers, representing a total of 123 queries.

*** METRICS ***
Evaluation will be performed using the metrics specific to each
dataset we use, i.e., MAP@10 for the interestingness prediction, and
F1@20 and ClusterRecall@20 for the search results diversification.

- Task registration opens: November 15, 2021
- Run submission: May 6, 2022
- Working notes submission: May 27, 2022
- CLEF 2022 conference: September 5-8, Bologna, Italy

*** REGISTER ***

Liviu-Daniel Stefan, Politehnica University of Bucharest, Romania
Mihai Gabriel Constantin, Politehnica University of Bucharest, Romania
Mihai Dogariu, Politehnica University of Bucharest, Romania
Bogdan Ionescu, Politehnica University of Bucharest, Romania

On behalf of the Organizers,

Bogdan Ionescu



[log in to unmask]

If you don't already have a password for the LISTSERV.ACM.ORG server, we recommend
that you create one now. A LISTSERV password is linked to your email
address and can be used to access the web interface and all the lists to
which you are subscribed on the LISTSERV.ACM.ORG server.

To create a password, visit:


Once you have created a password, you can log in and view or change your
subscription settings at: