Prof. Dr. Felix Naumann

For our first meeting in this semester, we will discuss organizational issues:

-          Introduction of new members 

-          Travel and travel funding

-          Semester curriculum, including assignments for the remaining open slots

-          Research skills curriculum

-          HPI Research Symposium

-          PhD regulations at DEF

-          Open questions

From April 25-29, 2022, the two Research Schools of the Hasso Plattner Institute will host the Research Symposium. Further information can be found here.

Dr. Ralf Herbrich

In this discussion, I will share my experiences of both the typical structure of a scientific conference (week), what are my goals when attending a conference and the implications on what to do before, during and after the conference. I will dive deep into all elements of a scientific conference starting from tutorials, the technical program, poster session, workshops to the opening reception and dinners. The presentation shall only serve as a structure and initiation of a group discussion exchanging all our experiences in attending a conference.

Prof. Dr. Tilmann Rabl

In this session, we will discuss benchmarking and measurement. We will present some techniques for fast performance modelling and how to integrate this into the publication pipeline. Then we will further discuss terminology and some of the pitfalls in benchmarking and performance comparisons.

Prof. Dr. Bernhard Renard

Commonly doctoral students help in advising students working on master theses or projects. This can be very beneficial from a scientific perspective, while it additionally also allows building up key competences for future careers in academia as well as industry. At the same time, advising is a not without challenges. Within the seminar, we will discuss key components and strategies of advising as well as cover important boundaries including mental health aspects to consider.

Dr. Michael Perscheid

Have you ever asked yourself what is expected during your PhD studies? How many papers should I write? Which conferences should I visit? When am I done? Trying to answer these questions, Michael will present a three-phase PhD journey framework (Finding a topic, doing research, writing down) that serves as starting point for discussing expectations in our research school round. With the help of fruitful discussions, we all should get a clearer picture on the way ahead and a more refinement of this framework for guiding upcoming PhD students.

Marcus Pappik

During recent decades, there has been a rich scientific exchange between theoretical computer science and statistical physics. On the one hand, this is because models studied in statistical physics are often very complex. This results in a need for efficient algorithmic methods to simulate them and to compute their properties. Even though statistical physicists were already studying such algorithmic techniques early on, most results were only heuristic in nature. From the perspective of theoretical computer science, this raises the question if those techniques can be turned into rigorous algorithmic results. On the other hand, many computational problems that arise from statistical physics have deep connections to standard questions from theoretical computer science, especially from the field of algorithmic counting complexity. Along these connections, various properties that are studied in physics were shown to be helpful for the construction of efficient algorithms or proofs of computational hardness.
In my talk, I give an overview of this exchange between initially seemingly unrelated areas of research, and I guide you through some of the most celebrated results at the intersection of both fields. At the end, I will finish the talk by giving a short summary of my own contributions so far and give a list of possible future research directions.

Stefan Neubert

Computing a problem solution in the form of small parts instead of one large chunk is key to improve utilization of processing resources. By enumerating partial solutions with little delay, subsequent processing can start its work before all parts of the solution are known. An obvious application for this algorithmic technique is scheduling: As soon as the first jobs in a schedule are fixed, processing units can already start executing whilst later jobs are still being scheduled.
However, the research field of scheduling is vast. In this work-in-progress research talk I will (1) give you a rough idea on how to navigate this field, (2) show how I picked problem variants that seemed interesting, and (3) share my successes, failures and which open problems I’m working on.

Christoph Anders

Objective: Wearable multi-modal time-series classification applications outperform their best unimodal counterparts and therefore hold great promise. In this context, different combinations of physiological signals, like heart-rate-variability and sweat-rate, are utilized. While many measurement techniques record physiological correlates of a persons mental state, electroencephalography directly measures electrical correlates of the human decision-making organ, the brain. Due to various noise sources, different key brain regions and frequency bands, as well as signal characteristics like non-stationarity, many options exist on how to work with electroencephalography data. Some aspects like data pre-processing and classification algorithms depend on the task at hand, and not every approach is suited for every application. Approach: Here, a systematic literature review on mental state classification for wearable electroencephalography is given. Discussing recorded data, frequently used pre-processing techniques, time-series classification models, and reproducibility, publications are set in the context of the current state-of-the-art in time-series classification of mental states. Main results: Five main conclusions are given, covering utilization of available area for electrode placement on the head, most often or scarcely utilized features and time-series classification model architectures, baseline reporting practices, as well as explainability and interpretability of Deep Learning. Significance: This systematic literature review summarizes the current state-of-the-art mental state classification using wearable electroencephalography. Discussions on under-reported aspects relevant to practical applications support future research.
In my talk I will give a brief introduction to the subject area of mental state classification and related work, presenting my first investigation during my Ph.D.-studies so far, a systematic literature review. I will close with a small discussion of open problems and future goals.

Prof. Dr. Felix Naumann

Choosing a venue to submit your work, be it a journal or a conference, be it a research paper, a demo paper or anything else, is an important but at times difficult decision. We will discuss the following questions around publication venues and how to choose them:

-          The role of conferences vs. journals (vs. book chapters and others)

-          Conference and journal rankings

-          Predatory conferences and journals

-          Reviewing modes

-          Chairs and editors

-          Venue topics

Prof. Dr. Stephan Mandt

The past few years have seen deep generative models mature into promising applications. Two of these applications include neural data compression and forecasting high-dimensional time series, including video. I will begin by reviewing the basic ideas behind neural data compression and show how advances in approximate Bayesian inference and generative modeling can significantly improve the compression performance of existing models. Finally, I show how neural video codecs can inspire probabilistic forecasting, leading to probabilistic sequence prediction methods with high potential for data-driven weather prediction.

Alejandro Sierra-Múnera

Pre-trained language models (PLMs) have received significant attention, as they allow many natural language processing approaches to achieve state-of-the-art results on a broad range of tasks.
The transformer model, which is the basis of most PLMs, encodes multiple aspects of text into rich contextualized vectors, representing lexical, syntactic, and semantic information captured during pre-training. But some of these representations are kept in the hidden layers and then aggregated to have a single final summarized representation to be used in downstream tasks. The research community has been able to empirically identify the presence of these structures within the pre-trained models, which were not explicitly trained to represent them, suggesting that with large amounts of text and sufficiently rich models, many linguistic structures were inferred in an unsupervised fashion.
In this project, we define a model to train the extraction of a latent space from pre-trained transformer encoders, in which constituency trees are represented in terms of token distances. We validate the potential of this latent space for transfer learning, comparing it with an off-the-shelf syntactic parser.

Anne Radunski

In an unstable and critical time marked by the COVID-19 pandemic, the Russian war against Ukraine, and the climate crisis, imminent changes and constraints are inevitable.
Especially entrepreneurs develop psychological issues due to the fear of the increasing inflation and the unstable economic situation resulting from the ongoing global crises.
However, the role of entrepreneurs in the economy is essential, as they are drivers of innovation for society and create the conditions for future growth. 
Despite this evidence, relatively little is known about effective support programs to address the increasing occurrence of mental illness among different groups of entrepreneurs.
For example, novice and expert entrepreneurs differ in terms of their cognition (e.g., decision-making), but how do they differ regarding their mental health.

In this research project, we aim to address the issue of how the current global crises influence the state of mind of entrepreneurs who have just started their business (novices) and those who have already founded a business (experts).
The present study employs a qualitative approach that involves multiple interviews to explore perceptions of the current global crises through an indirect evaluation process. 
Based on the collected data, annotated transcriptions are used to identify the mentioned emotions, their intensity as well as distinguish whether these emotions are positive or negative. 
Additionally, we use topic modeling to understand the underlying causes of emotions.

Prof. Dr. Boris Glavic

In this talk I will introduce the audience to historical what-if queries, a novel type of what-if analysis that determines the effect of a hypothetical change to the transactional history of a database. For example, "How would revenue be affected if we would have charged an additional $6 for shipping?" We have developed efficient techniques for answering historical what-if queries, i.e., determining how a modified history affects the current database state. Our techniques are based on reenactment, a replay technique for transactional histories. We optimize this process using program and data slicing techniques that determine which updates and what data can be excluded from reenactment without affecting the result. The program slicing techniques developed as part of this work are of independent interest and have potential applications in proving equivalence of and analyzing transactional histories.

Boris Glavic is an Associate Professor in the Department of Computer Science at Illinois Institute of Technology leading IIT's DBGroup. His research spans several areas of database systems and data science including data provenance, data integration, query execution and optimization, uncertain data, and data curation. Boris strives to build systems that are based on solid theoretical foundations.

Mazhar Hameed

CSV files are an excellent format for data exchange due to their flexible standard. However, due to this flexibility and plain text format, these files often contain data with structural issues, such as unescaped quote characters within quoted fields, columns containing multiple value formats, records with different numbers of cells, etc. We refer to records that contain such structural inconsistencies as ill-formed. Consequently, ingesting these files into a host system often requires many prior data preparation steps. Cleaning up the structure of ill-formed records is a task that data scientists have to deal with manually to make data loadable. 

Traditionally, data scientists write custom code to clean up ill-formed records, even before using data cleaning tools and libraries. These tasks are tedious and time-consuming, requiring expertise and frequent human intervention.  To automate this process, we propose TAHARAT, a system that automatically infers row patterns for records in CSV files using a syntactic pattern grammar.  We leverage these patterns both to understand the ill- and well-formedness of individual records in a file and to clean up their structure.