FG Datenbank- und Informationssysteme
Refine
Year of publication
- 2021 (3) (remove)
Document Type
Language
- English (3)
Keywords
- Abfragesprache (1)
- Assessed dataset (1)
- CNN (1)
- CQQL (1)
- Clique (1)
- Clustering (1)
- Condition Weighting (1)
- Erklärbarkeit (1)
- Instance weight (1)
- Neural networks (1)
Institute
Experts and crowds can work together to generate high-quality datasets, but such collaboration is limited to a large-scale pool of data. In other words, training on a large-scale dataset depends more on crowdsourced datasets with aggregated labels than expert intensively checked labels. However, the limited amount of high-quality dataset can be used as an objective test dataset to build a connection between disagreement and aggregated labels. In this paper, we claim that the disagreement behind an aggregated label indicates more semantics (e.g. ambiguity or difficulty) of an instance than just spam or error assessment. We attempt to take advantage of the informativeness of disagreement to assist learning neural networks by computing a series of disagreement measurements and incorporating disagreement with distinct mechanisms. Experiments on two datasets demonstrate that the consideration of disagreement, treating training instances differently, can promisingly result in improved performance.
To enable users to influence clusters of a social network by their external feedback, we present an adaptive clustering-based quantitative weighting approach. Intrinsically, Persons in a social network are connected and their homogeneity is reflected based on the similarity of their attributes. But all attributes do not have the same influence on the network and thus may affect to form the network and compromise the quality of resulting clusters. The introduced weighting approach is completely embedded in logic and has the capability of assigning query weights to atomic conditions in user interaction. Hence, the presented system supports users by offering an intuitive feedback formulation without deeper knowledge of the underlying attributes of objects. Experiments demonstrate the benefits of our approach.
Convolutional neural networks are often successfully used for clas-
si cation problems. Usually, a huge number of weights need to be
learnt by use of training data. However, the learnt weights give no
insight how the cnn really works. Thus, a cnn can be seen as a black
box solution. In our approach we develop a method to generate a
commuting quantum query language (cqql) condition from a sample
derived from a given cnn or from training input. The query language
cqql is inspired by quantum logic and its conditions obey the rules of
a Boolean algebra. The evaluation of a cqql condition provides values
from the unit interval [0; 1] and establishes therefore an elegant bridge between logic and a cnn. The underlying assumption is that a condition (a logic expression) gives much more understanding than pure cnn weights. Furthermore, the rich theory of Boolean algebra can be used for manipulating logic expressions. After extracting a cqql condition from a cnn or its training data we can use logic as a way to predict classes alternatively to a cnn.