Maximilian Pensel

About me

Nice to meet you! I am a passionate computer scientist and I have recently finished my PhD in theoretical computer science. Nonetheless, my affection for complex problems and the search for cognitive challenge has merely taken an excursion into the world of theory.

I have discovered my love for the digital world in my first computer science class in high school in Berlin, Germany. Immediately, I found myself not finishing exercises to the point, but setting my own, more challenging goals and exceeding the requirements manifold. My personal interest and enthusiasm has lead me through the path of studying computer science in Dresden, Germany. With a particular affinity for artificial intelligence, I have dedicated my graduate studies to the field of knowledge representation and reasoning at the chair for automata theory of the elite university TU Dresden.

In knowledge representation and reasoning we use formal logic (e.g. propositional logic, first-order logic, Description Logics) to represent explicitly known facts and relations and then devise and analyze efficient strategies to reason over this knowledge, i.e. to derive implicitly entailed information. I studied a form of non-monotonic reasoning for the (also in industry) increasingly popular family of Description Logics. Such reasoning allows to derive conclusions based on a set of facts that can be automatically retracted in case the discovery of new facts contradicts the original conclusion. This strategy is essential in the way humans reason with their own set of facts or beliefs. Clearly, agents based on artificial intelligence can also benefit from the capability to model default behavior and react appropriately in light of contradicting observations.

The formal approach has certain advantages over the stochastical methods that are so very popular nowadays (e.g. machine learning). It is well-behaved, easily reproducible, and most importantly, it allows for formal verification of a systems behavior, something that is vital in many realistic scenarios where reliability or even safety is of the highest priority, e.g. guiding a course of action in medicine or steering an autonomous vehicle from point A to point B.

Whenever I am not sitting at a computer working, following my own projects, self-educating about modern development technologies or methods, or contemplating new theoretical advancements, you will likely find me in a local climbing gym, spending time with my family or home-cooking fresh and healthy meals.

-- Knowledge is everywhere.

Projects

Ever since I came into contact with programming I have been following my own interests, ideas and projects. Be it out of interest for a new technology or programming language/paradigm, or out of personal benefit to myself and the people around me.

OpenWebScraper

I am currently developing an open source PyQt frontend for the scraping and web crawling framework Scrapy. In addition to rudimentary crawling functionality, the tool will also contain features to support a quantitative analysis of the text data obtained from running Scrapy. The initial extent of features is tailored to the needs of the researchers participating in the german DFG research project Die Konstruktion organisationaler Identität und der Einfluss von Geschichte at the TU Ilmenau.

Technologies	Python, Pandas, Scrapy, PyQt
Status	Early development
Development	Since 07/2019

GitHub (GUI) GitHub (Scrapy Wrapper)

ELastiq

ELastiq is a quantitative Description Logic reasoner, providing relaxed answers to instance queries based on a formal calculation of concept similarity and its comparison to a predefined threshold value.

Technologies	Java, OWL
Status	Finished
Development	2015

Theoretical Paper System Description GitHub

Simi-Framework

Simi is a Java framework for constructing concept similarity measures for Description Logic concepts. This implementation is called a framework in the classical sense, it itself has no immediately executable functionality but can be used, modified or extended to construct custom concept similarity measures. The framework is practically employed in a cooperation project between the TU Paderborn and the TU Dresden, providing a new form of so-called congenial benchmarks, evaluating the Description Logic concept representation of SPARQL queries by means of concept similarity.

Technologies	Java, OWL
Status	Finished
Development	2015

Theoretical Paper

HyperDiff

HyperDiff is a Java tool for computing the logical difference between Description Logic Ontologies. This tool prototypes a novel approach to compute the difference between two \(\mathcal{EL}\) TBoxes based on their representation as hypergraphs. In very brief terms, the logical difference of two sets of terminological axioms is the set of all conclusions that can be drawn from one set of axioms, but not the other. While this set is infinite in general, it can be described by a finite set of primitive witnesses. Those witnesses can be determined by computing mathematical simulations between two hypergraphs with a recursive algorithm.

Technologies	Java, OWL, Elk
Status	Finished
Development	2014

Theoretical Paper

Project Time Manager

I developed the Project Time Manager as an exercise to learn both C# and GUI programming. It is a rudimentary tracking system for work-time and accomplishments, not unlike a time stamp clock. You can create projects and track the time of individual work-sessions per project, while attaching notes of accomplishments to every session.

Technologies	C#, WDT, XML
Status	Beta, stopped development
Development	2011

Publications

Before and during my PhD I have coauthored several scientific publications, from system descriptions of practical implementations to theoretical contributions in the field of knowledge representation and reasoning. Below is a simple list of my publications, for potentially more detailed lists you can consult the following resources.

DBLP Google Scholar TU Dresden

2020

A Lightweight Defeasible Description Logic in Depth - Quantification in Rational Reasoning and Beyond.
In Künstliche Intelligenz - Special Issue: Ontologies and Data Management

2019

With Axel-Cyrille Ngonga Ngomo, Felix Conrads and Anni-Yasmin Turhan:
Congenial Benchmarking of RDF Storage Solutions.
In the 10th International Conference on Knowledge Capture

2018

With Anni-Yasmin Turhan:
Reasoning in the Defeasible Description Logic \(\mathcal{EL}_\bot\) - Computing Standard Inferences under Rational and Relevant Semantics.
In the International Journal for Approximate Reasoning (Special Issue on Defeasible and Ampliative Reasoning)
With Anni-Yasmin Turhan:
Computing Standard Inferences under Rational and Relevant Semantics in Defeasible \(\mathcal{EL}_\bot\).
In the International Workshop on Description Logics 2018
With Franz Baader and Oliver Fernandez Gil:
Standard and Non-Standard Inferences in the Description Logic \(\mathcal{FL}_0\) Using Tree Automata.
In the 4th Global Conference on Artificial Intelligence
With Franz Baader and Pavlos Marantidis:
The Data Complexity of Answering Instance Queries in \(\mathcal{FL}_0\).
In the Companion of the The Web Conference 2018

2017

With Anni-Yasmin Turhan:
Including Quantification in Defeasible Reasoning for the Description Logic \(\mathcal{EL}_\bot\).
In the 14th International Conference Logic Programming and Nonmonotonic Reasoning
With Anni-Yasmin Turhan:
Making Quantification Relevant Again - the Case of Defeasible \(\mathcal{EL}_\bot\).
In the 4th International Workshop on Defeasible and Ampliative Reasoning

2015

With Andreas Ecke and Anni-Yasmin Turhan:
ELastiq: Answering Similarity-threshold Instance Queries in \(\mathcal{EL}\).
In the 28th International Workshop on Description Logics

Theses

Dissertation:
A Lightweight Defeasible Description Logic in Depth - Quantification in Rational Reasoning and Beyond.
Supervised by PD. Dr.-Ing. Anni-Yasmin Turhan and Prof. Dr. Gerhard Brewka
Submitted on June 26, 2019
Disputation on October 28, 2019
Graded summa cum laude
Master Thesis:
An Automata-Based Approach for Subsumption w.r.t. General Concept Inclusions in the Description Logic \(\mathcal{FL}_0\).
Supervised by Prof. Dr.-Ing. Franz Baader
Graded 1.0
Bachelor Thesis:
Automated Tuning of SAT Solvers for Solving Hidokus.
Supervised by Prof. Dr. rer. nat. habil. Steffen Hölldobler, Dr. Christoph Wernhard, Dipl.-Inf. Norbert Manthey and Dipl.-Inf. Hau Nguyen Van
Graded 2.3

Contact me

If you have any inquiries regarding my projects, publications or otherwise, please do not hesitate to send me a mail. Perhaps you are interested in, need support for, want to use or continue working on one of my projects for which the sources are not publicly available, or you want to discuss any of my (or your) research interests. In any case, I am happy to be of service.

As of August 2019 I am also offering my capabilities as a freelancing developer, consultant and data-scientist. Whether you need another productive member in your development team or require small but individual solutions to your IT problems, be sure to contact me. For my full curriculum vitae, feel free to check my Xing profile.

Xing