To foster collaboration across research groups and to encourage creativity in coming up with research projects, we will be writing and proposing abstracts about fictitious (and future?) research papers. To this end, we will divide up all attendees into teams of three. No team can include more than one member from any one HPI group. Each team will have 30 minutes to create a title and abstract. The topic can be wild, but should cut across the different areas of expertise of the authors. Afterwards, all abstracts will be read, followed by a very brief Q&A. The whole event should be fun and might even lead to actual cooperation in the future.

All good things must come to an end, and so does my time as regular member of the Research School. Thus, in today's talk, I will share the experiences I gathered during that time with you. Rather than being a classical talk, I would like to discuss whether you had similar experiences or whether I missed out on something.

After that I will leave the stage and hand over the sceptre to Aikaterini, my successor in the Research School. She will start her Research School journey by shortly introducing herself.

This is a collaborative session in which we share and discuss tools we use / want to use / can recommend for our research. These tools can for example be your answers to the questions:

• How do you collect your ideas?
• How do you keep track of references and your backlog of "stuff you want to read"?
• How do you manage your data?
• How do you write papers?
• How do you communicate and collaborate with other researchers?
• How do you find related work?
• How do you manage your time?

Research is defined by its quest for novel knowledge, at the same time its inherent freedom faces boundaries with regard to ethical considerations and dual use estrictions. Not all research that is possible should necessarily be conducted. Within this talk I will outline why this question is also of relevance with regard to research in data science and engineering and give examples of possible ethical considerations and dual use conflicts from my own experience. I will introduce current minimum standards as well as procedures for ethics votes and dual use assessments as well as debate further measures.

The majority of current state-of-the-art approaches try to learn disentangled representations by operating on the latent space priors. In contrast, we investigate the effect of regularizing the model architecture itself, by reducing the number of connections between the hidden units of the networks. Dense connectivity increases the number of parameters in a model, and thus its potential expressive power. However, recent works have shown that in practice a large percentage of weights remains unused and can be removed without sacrificing performance.

We argue that sparsely connected features might be desirable when learning disentangled representations and propose techniques for obtaining such models - Random Masking and Learnable L1 Masking. These are then evaluated on standard datasets for measuring disentanglement, along with ablation tests to identify the factors crucial for improving performance.

How to manage your time efficiently? How to motivate yourself? How to avoid distractions?

There is a plethora of time management techniques trying to answer these questions, but who has the time to get to know them all? Luckily, many techniques can be broken down to some key insights. In this seminar I will give a short overview of some of these insights and how to apply them. However, since not all time management techniques work for everyone, you are welcome to share your own insights and experiences.

Applications in the Internet of Things (IoT) create many data processing challenges because they have to deal with massive amounts of data and low latency constraints. The DEBS Grand Challenge 2020 specifies an IoT problem whose objective is to identify the special type of events in a stream of electricity smart meters data.

In this work, we present the Sequential Incremental DBSCAN-based Event Detection Algorithm (SINBAD), a solution based on an incremental version of the clustering algorithm DBSCAN and scenario specific data processing optimizations. SINBAD manages to calculate solutions up to 7 times faster and up to 26% more accurate than the baseline provided by the DEBS Grand Challenge.

Patience, self-analysis, understanding are important qualities that one may forget when reading a response letter. This talk will present some tips to gather a constructive feedback from reviewers and response letters. For each review question, there is a practical exercise that can be used to stimulate objective thinking and finding possible improvements for the submitted work based on the reviews received. Warning: the practice of Zen Reviewing may improve the quality of writing reviews for other works.

In this talk, we will discuss important elements of good scientific practice. We will introduce the DFG's guidelines for good scientific practice and discuss publishing practice. We will also give an overview of the processes and consortia in place at HPI and UP.

Each PhD student will briefly (<5mins) elaborate what they are currently working on, e.g., open problems, solution ideas, proofs, implementation, experimentation.

As long as it is advisable to meet only virtually, we will meet via Zoom. We will use the first session for a quick round of introductions for new members, a quick glance at the schedule for this semester, and to reassign duties.