Sie sind hier

Talks

While probabilistic programming is a powerful tool, uncertainty is not always of a probabilistic kind. Some types of uncertainty are better captured using ranking theory, which is an alternative to probability theory where uncertainty is measured using degrees of surprise on the integer scale from 0 to ∞. Here we combine probabilistic programming methodology with ranking theory and develop a ranked programming language. We use the Scheme programming language a basis and extend it with the ability to express both normal and exceptional behavior of a model, and perform inference on such models.

18.07.2019 - 10:15 h
Frank Schaust

Convolutional Neural Networks werden bereits in vielen Anwendungsbereichen erfolgreich genutzt. Vor allem im Bereich der Bildklassifizierung erreichen Convolutional-Neural-Network-Architekturen gute Resultate. In dieser Arbeit werden gängige Convlutional-Neural Network-Architekturen aus der Bildklassifizierung in einem neuen Aufgabengebiet, namentlich der Vorhersage von Gefechten im Echtzeit Strategiespiel StarCraft II im Bezug auf ihre Performance in der neuen Domäne verglichen. StarCraft II ist in diversen Bereichen des Machine Learning Anschauungsobjekt für unterschiedliche Aufgaben und dient mit seiner Python-Schnittstelle pysc2 als optimales neues Aufgabengebiet.

11.07.2019 - 10:15 h
Oana Dumitrasc

Text generation represents an active research field, especially with generative models. This thesis uses the Generative Adversarial Networks architecture for the purpose of names generation. This architecture was not initially designed to work on discrete data, but thanks to its high performance in other fields (e.g. computer vision) and the advantage of easy data creation, the ongoing research of the community tries to answer the question if the above mentioned architecture is suitable for text (here: name) generation.
The discrete nature of names makes it difficult to leverage the potential of this framework. The Gumbel-Softmax reparameterization technique proposed in the literature is used in this thesis to overcome this drawback.

11.07.2019 - 10:15 h

In the EXCITE project, we have narrowed the gap of the lag between social sciences and other fields in terms of citation data availability. This has been achieved by extracting, parsing and matching references from PDF documents. During the project, we have found that a lot of references are not indexed in any bibliographic index.

04.07.2019 - 10:15 h
Federico Nanni

Political text scaling aims to linearly order parties and politicians across political dimensions (e.g., left-to-right ideology) based on textual content (e.g., politician speeches or party manifestos). Existing models, such as Wordscores and Wordfish, scale texts based on relative word usage; by doing so, they do not take into consideration topical information and cannot be used for cross-lingual analyses. In our talk, we present our efforts toward developing topic-based and semantically aware text scaling approaches.

27.06.2019 - 10:15 h
Ramin Hedeshy

Gaze-based text entry systems have been an important means of communication for people with motor disabilities. Although several dwell-time and dwell-free tools have been developed to facilitate the process of gaze-based text entry, still the typing speed is quite slow and the cognitive load is rather high. Moreover, most previous methods are developed only based on gaze and fixations sequence. However, these methods can result in lengthy amounts of time for typing. Besides, users cannot always perfectly gaze at every key in many cases. In this thesis, we propose TGSBoard an onscreen keyboard that combines the simplicity and accuracy of touch inputs with the speed of eye typing by gaze swiping to provide efficient and comfortable dwell-free text entry.

18.06.2019 - 16:15 h
Teresa Krämer

Subtasks within commonsense reasoning are the derivation of new information from existing knowledge (forward reasoning) and checking whether a statement (conjecture) can be derived from the given knowledge (backward reasoning). In order to perform the latter task, often a small subset of the knowledge is sufficient to prove the conjecture. The difficulty is to identify and select this subset in an automated way. Selection methods like SInE [1] and Similarity SInE [2] can be used to preselect knowledge that might be relevant. However, those methods do not always select enough or the right knowledge to find proof.

13.06.2019 - 10:15 h

Text predictions play an important role in improving the performance of gaze-based text entry systems. However, visual search, scanning, and selection of text predictions require a shift in the user's attention from the keyboard layout. Hence the spatial positioning of predictions becomes an imperative aspect of the end-user experience. In this work, we investigate the role of spatial positioning by comparing the performance of three different keyboards entailing variable positions for text predictions. The experiment result shows no significant differences in the text entry performance, i.e., displaying suggestions closer to visual fovea did not enhance the text entry rate of participants, however, they used more keystrokes and backspace.

13.06.2019 - 10:15 h

In CUTLER project [1] our aim is to assist policy-makers through data (environmental/economical/social) analysis methods. However, Effective data analytics need visualization tools in order to reveal insights and make sense to experts or even users without technical knowledge. For the development of the social data visualization, we follow CUTLER‘s representation interface and integrated framework with KIbana dashboard. Using the framework, I will describe the visualization widgets for tweet, news comments, and event data, by means of different query methods and interactive visualizations such as tag clouds, geographic mapping, temporal characteristics, and sentiment classifications. I will use sample of crawled data for the CUTLER pilot cities to elaborate the visualizations.

06.06.2019 - 10:15 h

Research at WeST now includes a wider range of politics-related topics than before due to several reasons: Firstly, politics drive behaviors on the Web that trouble society as a whole. Secondly, politics influence content on news and social media. Thirdly, democracy online is a challenge of asking what should and what can be regulated. "Politics" in this sense is broader than the level of politicians and includes the audience level. From a data perspective, politics provide a mapping of expected behaviors by groups and positions within text. My talk briefly overviews, with some ongoing snippets, the following application areas: misinformation, partisanship, discourses, platforms, and retro-theories.

23.05.2019 - 10:15 h

Seiten