OFAI is delighted to announce its 2023 Winter/Spring Lecture Series, featuring an eclectic lineup of internal and external speakers.
The talks are intended to familiarize attendees with the latest research developments in AI and related fields, and to forge new connections with those working in other areas.
Lectures will take place on Wednesdays at 18:30 Vienna time. All lectures will be held online via Zoom; in-person attendance at OFAI is also possible for certain lectures. Attendance is open to the public and free of charge. No registration is required.
Subscribe to our newsletter or our RSS feed, or bookmark this web page, to receive further details for the individual talks.
18 January 2023 at 18:30 CET (UTC+1)
Martin Trapp (Aalto University)
Leveraging Connections Between Deep Architectures and Bayesian Nonparametrics
Deep architectures have become an integral part of modern AI systems. However, despite their computational benefits, concerns about their robustness and missing interpretability often limit their applicability in scenarios where trustworthiness is of importance. This motivates the Bayesian approach to deep learning, which aims to encode prior knowledge into the model. However, defining Bayesian priors for deep architecture is often challenging. In this talk, I will discuss how drawing connections to Bayesian nonparametric priors can help in (i) encoding conservative behaviour into deep learning models, (ii) understanding priors over deep tractable models, and (iii) developing tools for uncertainty quantification in computer vision tasks.
1 February 2023 at 18:30 CET (UTC+1)
Due to unforeseen circumstances, we have had to postpone Winfried Lechner's talk, originally scheduled for 1 February 2023, until Wednesday, 8 March at 18:30 CET (UTC+1). See the new entry below for the talk details.
15 February 2023 at 18:30 CET (UTC+1)
Benjamin Mako Hill (University of Washington)
Balancing Open Participation and Information Quality in Wikipedia Using Machine Learning
Peer produced information goods like free/open source software and Wikipedia are both increasingly important and increasingly under threat. This talk will describe how Wikipedia has sought to balance its commitment to open editing and its desire to allow participation from unvetted and anonymous users with its need to maintain high information quality in its articles. I will focus on the way that a set of ML/AI systems developed by the Wikimedia Foundation allow scholars to measure the value of contributions from anonymous users and the surprising way that these systems can also be used by the Wikipedia community to shape editing behavior. I will argue that use of these ML/AI systems can both improve the efficiency of moderation work while also making moderation actions more fair to anonymous contributors who are the source of substantial vandalism by reducing reliance on social signals and making norm violations by everyone else more visible.
1 March 2023 at 18:30 CET (UTC+1)
Michael Pucher (OFAI)
Synthesizing Dialects, Faces, Singing Voices, Songbirds, and Famous Dead Actors
During the last decades statistical parametric speech synthesis has significantly improved the quality and flexibility of speech synthesis systems. This development started with hidden Markov models (HMM) and then another big step of improvement in acoustic modeling and vocoding was made with deep neural networks (DNN). In this talk I will present a range of applications of statistical parametric speech synthesis that we have investigated. In the field of acoustic speech synthesis I will show how dialect interpolation can be realized, which allows for the generation of in-between language varieties. In audio-visual speech synthesis joint audio-visual modeling and visual control will be presented. In singing speech I will describe our work towards an opera style singing synthesis system that is trained on high quality opera singing data. A model for synthesis of singing birds will be presented that can control bird songs by symbolic input sequences. Finally, I will present a DNN-based synthesizer of a famous Austrian actor that we have built from audio book data, and that was used in a theater play. I will conclude my talk with an outlook on the future of speech synthesis technologies, remaining technical and possible societal challenges.
8 March 2023 at 18:30 CET (UTC+1)
Winfried Lechner (University of Athens)
Natural Language Semantics and Music
Both human language and music can be modeled as discrete, combinatorial systems. Lerdahl and Jackendoff (1983) demonstrated that applying the methods of formal linguistics to the study of (tonal) music provides a strategy for exposing homologies between these two systems. But while syntactic properties of music have attracted a considerable amount of interest (Rohrmeier 2011; Granroth-Wilding and Steedman 2014; see Rohrmeier & Pearce 2018 for an survey), meaning-related aspects have by and large remained understudied (an exception is Schlenker 2019, 2022). In part, this might be due to the widely held belief that unlike natural language, music neither has a lexicon of atomic form-meaning pairs (Katz and Pesetsky 2011) nor employs compositional mechanisms to recursively derive complex meanings. I will explore some consequences of these two assumptions by making explicit what they entail for the study of music from a linguistic perspective.
15 March 2023 at 18:30 CET (UTC+1)
Niels Taatgen (University of Groningen)
The Skill-based Method of Modeling Human Intelligent Behavior
Humans are capable of performing many novel tasks with little or no instruction, contrary to machines. Unfortunately, research in cognitive science and psychology pays very little attention to this. Because of this, many cognitive models focus on exhaustively explaining data from single experiments, but ignore the question where that knowledge originates from, and how it can be reused in other contexts. A possible solution is to assume people have a set of cognitive skills that they can recombine to carry out tasks. The idea is analogous to the idea that words in a language can be combined in many different ways to create new meaning. I will demonstrate this idea using the PRIMs cognitive architecture, which is derived from the ACT-R architecture, for example with the Attentional Blink task. The model of that task only consists of skills taken from models of other tasks.
29 March 2023 at 18:30 CEST (UTC+2)
Due to illness, we have had to postpone Stacy Marsella's talk, originally scheduled for 29 March 2023, until Wednesday, 17 May at 18:30 CEST (UTC+2). See the new entry below for the talk details.
12 April 2023 at 18:30 CEST (UTC+2)
Christoph Scheepers (University of Glasgow)
The “Crossword Effect” in Free Word Recall: A Retrieval Advantage for Words Encoded in Line with their Spatial Associations
According to the perceptual symbol hypothesis (Barsalou, 1999), word concepts trigger mental re-enactments of perceptual states and actions. While many studies have shown how word concepts modulate sensori-motor responses, it is less well known how sensori-motor actions influence access to word concepts in memory. Here, we investigated how well English words with strong horizontal or vertical associations are retrieved from memory dependent on how they are presented during encoding (i.e., horizontally or vertically printed). Initial pre-testing of 129 candidate words yielded 43 words with a strong horizontal association (e.g., floor, beach, border, etc.) and 51 words with a strong vertical association (e.g., tree, crane, bottle, etc.). These were quasi-randomly compiled into 160 ‘crossword arrays’, each containing 5 horizontally and 5 vertically printed items drawn from the horizontal association word set, as well as 5 horizontally and 5 vertically printed items drawn from the vertical association word set. The main experiment (160 participants) was preregistered on OSF and was introduced to participants as “testing how word arrangements affect subsequent mathematical problem solving”. There were three experimental phases: (1) in the encoding phase, each participant studied a uniquely generated crossword array for ca. 2 minutes; (2) in the following distractor phase, they had to solve simple mathematical equations for 1 minute; (3) in the final (surprize) free recall phase, they were asked to write down as many words as they could remember from the encoding phase. Dependent variables were likelihood of correctly recalled words and retrieval ranks of correctly recalled words in the recall list. Results showed no appreciable effects in retrieval rank, but a clear interaction (p < .001) between word association and word presentation in the likelihood of correct word recall: vertical association words, in particular, were reliably more likely to be recalled correctly when they were presented vertically (i.e., in line with their spatial association) than when they were presented horizontally during encoding. Implications for the perceptual symbol hypothesis will be discussed.
26 April 2023 at 18:30 CEST (UTC+2)
Paolo Petta (OFAI)
Due to unforeseen circumstances, we have had to postpone Paolo Petta's talk, originally scheduled for 26 April 2023. A new date will be announced later.
10 May 2023 at 18:30 CEST (UTC+2)
Hannes Fellner (University of Vienna)
Digital Advances on the Ancient Silk Road
From the 2nd century CE on, communities and monasteries developed along the trade routes of the ancient Silk Road in and around the Tarim Basin in today’s Xinjiang Uyghur Autonomous Region in the People’s Republic of China. These were centres of writing, copying, translating, and transmitting texts similar to the monasteries in medieval Europe. The old Indo-European languages Sanskrit, Tocharian, and Saka written in a Central Asian variant of the Indian Brahmi script – Tarim Brahmi – were the major languages in use in the Tarim Basin in the first millennium CE. In contrast to the writing traditions in medieval Europe, the ones on this part of the Silk Road are not well understood, mainly due to the fragmentary status of texts. I this talk, I will address recent efforts of making these languages and the Tarim Brahmi script digitally accessible and operable for philological, palaeographic, and linguistic research in the framework of the FWF-START project “The characters that shaped the Silk Road – A database and digital palaeography of Tarim Brahmi”. In the project, the text witnesses are linked to their digital facsimiles on the character level using Transkribus. All data concerning the texts is combined in an XML database and published through a web application. This allows to determine which text was written by whom, when, where, and how in order to
- trace the evolution of Tarim Brahmi and its adaptation to the different languages
- reveal the relationship between script types, languages, and genres
- categorize countless text fragments that are so far unidentified (regarding language, provenance, date, genre etc.)
- potentially (re)combine scattered fragments belonging to the same manuscript leaf
- and, of course, to better understand literacy and writing culture in the Tarim Basin
17 May 2023 at 18:30 CEST (UTC+2)
Stacy Marsella (Northeastern University and University of Glasgow)
Engineering the Impact of Emotion on Human Behavior
Computational models of human behavior are used in a wide range of artifacts. At a large scale, social simulations are being used, for example, to explore people’s response to a natural disaster. At a medium-scale, models of human decision-makers are being used to study social technical systems such as the pharmaceutical drug supply networks. At the individual scale, work on human-robot and human-agent interaction seeks to facilitate interaction by giving artificial agents models of their human partners. At the extreme of modeling individual human behavior, virtual replicas of humans are being crafted, facsimiles of people that can engage people in face-to-face interactions using the same verbal and nonverbal behavior people use. The designs of these various models heavily leverage psychological theories and data. Psychology and the social sciences, in turn, use these computational artifacts as means to formulate, test, and explore theories about human behavior. In this talk, I will first give a brief overview of my group’s work in social simulation, social technical systems, HRI and virtual humans. Then I will exemplify the synergy between psychology and the engineering of these artifacts from the perspective of my group’s work on developing and applying computational models of emotion.
