Back to top

sebis and Bayerischer Rundfunk launch NLG system for automated basketball news

The chair of Software Engineering for Business Information Systems (sebis) and the AI + Automation Lab of the Bayerischer Rundfunk (BR) have developed an NLG system for the automated generation of news reports about matches in the highest German basketball league. The system is based on the open source surface realiser Simple-NLG, that was developed at sebis.

The BR has published an article with details about the project (in German).

  nlg nlp

Sebis Chair joins Bloxberg Consortium

We are very happy to announce that our chair has joined the Bloxberg consortium.

The bloxberg infrastructure is a secure global blockchain established by a consortium of leading research organizations to provide scientists with decentralized services worldwide. The bloxberg Consortium aims to fosters collaboration among the global scientific community, empowering researchers with robust, autonomous services that transcend institutional boundaries. For example, with consented transactions on the bloxberg infrastructure, research claims need not be limited to one institution alone, but can be confirmed by the whole trusted network.

The sebis chair set up and maintains a node within the network, contributing to the ongoing stability and success of the Bloxberg network.

Find out more here: Bloxberg Website

Press statement: Max Planck Digital Library

   No tags assigned

Paper titled Modeling aspects of the language of life through transfer-learning protein sequences published at BMC Bioinformatics Journal

The paper titled "Modeling aspects of the language of life through transfer-learning protein sequences" was pusblished at BMC Bioinformatics Journal.

Abstract:

Background One common task in Computational Biology is the prediction of aspects of protein function and structure from their amino acid sequence. For 26 years, most state-of-the-art approaches toward this end have been marrying machine learning and evolutionary information. The retrieval of related proteins from ever growing sequence databases is becoming so time-consuming that the analysis of entire proteomes becomes challenging. On top, evolutionary information is less powerful for small families, e.g. for proteins from the Dark Proteome.

Results We introduce a novel way to represent protein sequences as continuous vectors (embeddings) by using the deep bi-directional model ELMo taken from natural language processing (NLP). The model has effectively captured the biophysical properties of protein sequences from unlabeled big data (UniRef50). After training, this knowledge is transferred to single protein sequences by predicting relevant sequence features. We refer to these new embeddings as SeqVec (Sequence-to-Vector) and demonstrate their effectiveness by training simple convolutional neural networks on existing data sets for two completely different prediction tasks. At the per-residue level, we significantly improved secondary structure (for NetSurfP-2.0 data set: Q3=79%±1, Q8=68%±1) and disorder predictions (MCC=0.59±0.03) over methods not using evolutionary information. At the per-protein level, we predicted subcellular localization in ten classes (for DeepLoc data set: Q10=68%±1) and distinguished membrane-bound from water-soluble proteins (Q2= 87%±1). All results built upon the embeddings gained from the new tool SeqVec neither explicitly nor implicitly using evolutionary information. Nevertheless, it improved over some methods using such information. Where the lightning-fast HHblits needed on average about two minutes to generate the evolutionary information for a target protein, SeqVec created the vector representation on average in 0.03 seconds.

Conclusion We have shown that transfer learning can be used to capture biochemical or biophysical properties of protein sequences from large unlabeled sequence databases. The effectiveness of the proposed approach was showcased for different prediction tasks using only single protein sequences. SeqVec embeddings enable predictions that outperform even some methods using evolutionary information. Thus, they prove to condense the underlying principles of protein sequences. This might be the first step towards competitive predictions based only on single protein sequences.

 

Resources:

https://bmcbioinformatics.biomedcentral.com/articles/10.1186/s12859-019-3220-8

https://www.gauss-centre.eu/news/research-highlights/article/high-performance-computing-and-artificial-intelligence-decode-the-language-of-life-in-proteins/

 


'ProPerData - A process model to support GDPR compliance' published as technical report

Abstract

The General Data Protection Regulation (GDPR) has changed the perception towards privacy and data protection worldwide. Passed in 2016 and in force since 2018, the regulation has been a steady part of the academic and practical discourse over the past years. However, companies still struggle with the task of becoming compliant, mainly because of the large interdisciplinary scope and the overall complexity of the regulation. Once established, maintaining GDPR compliance in an accelerating business environment remains a challenge.

With this report, we present ProPerData, a process model for the protection of personal data. It addresses software developers and enterprise architects of large organizations and aims to provide a structured overview of the GDPR and a clear definition of responsibilities.

ProPerData is organized along 11 tasks that are derived from the GDPR. 16 work units of ProPerData are assigned to the tasks and executed by ProPerData stakeholders. We account for 6 resources that support the work units and 13 work products that result from them. The work units take place at one or more of the 10 stages or events of ProPerData.

 

Huth, D., Matthes, F.: "ProPerData - A process model to support GDPR compliance". Technical Report. Technical University of Munich, Munich 2020.

 


Document Automation Tool Study is Published

We are happy to announce that our survey about document automation tools for the legal domain has been published.

The survey evaluates 13 document automation tools in accordance with requirements, which were identified with respect to the legal domain.

Please find the result of our survey here.

We would like to sincerely thank the 13 vendors who participated in our study and granted us access to their tools for conducting the study.