Prof. Dr. Felix Naumann

For our first meeting in this semester, we will discuss organisational issues.

• Introduction of new members 
• Travel and travel funding
• Semester curriculum, including assignments for the remaining open slots 
• Research skills curriculum 
• PhD regulations at DEF
• Open questions 

Venue: L.E-03

Paraskevi Machaira

Title: REX: A Realistic Time-Dependent Model for Multimodal Public Transport

Abstract: We present the non-FIFO time-dependent graph model with REalistic vehicle eXchange times (REX) for schedule-based multimodal public transport, along with a novel query algorithm called TRIP-based LAbel-correction propagation (TRIPLA) algorithm that efficiently solves the realistic earliest-arrival routing problem. The REX model possesses all the strong features of previous time-dependent graph models without suffering from their deficiencies. It handles non-negligible exchanges from one vehicle to another, as well as supports non-FIFO instances which are typical in public transport, without compromising space efficiency. We conduct a thorough experimental evaluation with real-world data which demonstrates that TRIPLA significantly outperforms all state-of-the-art query algorithms for multimodal earliest-arrival routing in schedule-based public transport.

Juliana Schneider

Title: Hypothesis tests for image embeddings in n-of-1 trials.

Abstract: N-of-1 trials are multi-crossover clinical trials in individual persons, where a trial participant alternates between receiving a treatment or intervention and none / an alternative / placebo. We develop a combined modelling and testing approach to enable the use of multimodal outcome data in n-of-1 trials. We present a use case for images as outcome in a trial on acne crème for skin care. Embeddings are created by a Variational Autoencoder, then reduced to one-dimensional values by Principal Componant Analysis. Statistical tests on these resulting univariate data are compared in a first attempt to estimate the individual treatment effect. We outline ideas on how to incorporate time dependencies, carryover in effects and confounding in subsequent tests.

Venue: F.E-06

Eshant English

Title: Ferumal Flow: A Kernelised Normalising Flow

Abstract: Normalising Flows are generative models with invertible architecture. The invertibility in flows constrains expressiveness requiring a large number of parameters and innovative architectural design to produce good results. Most works on flows use neural network-based transformations, and the use of other transformations is rather understudied. We developed a novel flow architecture, Ferumal Flow, integrating Kernels into Normalising Flows. We show kernel-based flows can provide competitive or better results than neural net-based flows whilst being parameter efficient. We also demonstrate that our method works significantly better in low data regimes, where neural net-based architectures struggle to produce good results.

Venue: F.E-06

Prof Dr. Tilmann Rabl 

Prof Dr. Holger Karl

Title: Good Scientific Practice

Abstract: Good Scientific Practice is the necessary basis for the research community. In this talk, we will discuss major aspects of research integrity for PhD students. We will look at common pitfalls for young researchers and solution strategies. We will link this to the DFG Codex, which is basis for regulations regarding good scientific practice at all research institutions receiving DFG funding in Germany.

Venue: L.E-03

Prof Dr. Patrick Baudisch

Title: How to get into a faculty position before/after your PhD

Abstract: So when you get a Turing award, what will it have been for?” You have never  been asked this question? Well, you might be soon—and you better have a good answer, too. Because this is a question some schools ask as part of their job interviews for faculty positions. 

So, if you should consider staying in research past your PhD, You might want to learn more about how to score a position. There is lots to teach and learn. In this first of the talks, I will focus on the actual interview. 

Venue: F.E-06

Alejandro Sierra

Title: Efficient Ultrafine Typing of Named Entities

Abstract: Ultrafine named entity typing (UFET) refers to the assignment of predefined labels to entity mentions in a given context. In contrast to traditional named entity typing, the number of potential labels is in the thousands and one mention can have more than one assigned type.Previous approaches either depend on large training datasets, or require inefficient encoding of all input-type combinations. Therefore, there is a need for investigating the efficiency during training and prediction of entity typing models in the ultrafine-grained setting, considering its distinctively bigger search space, compared to the coarse- and fine-grained tasks.
To efficiently solve UFET, we propose Decent, a lightweight model that encodes, using a pretrained language model, the input sentences separately from the type labels. Additionally, we make use of negative oversampling to speed up the training while improving the generalization of unseen types. Using an openly available UFET dataset, we evaluated the classification and runtime performance of Decent and observed that training and prediction runtime is orders of magnitude faster than the current state-of-the-art approaches, while maintaining a competitive classification performance.

Venue: F.E-06

Martin Boissier

Title: Research Tools and Hacks

Abstract: For the next meeting on Wednesday 7 June, we would like to get input from all of you.
We are trying to collect tools and hacks that help you with your research. Such tools include, e.g., Overleaf shortcuts, Latex tricks/templates, Python libraries, and more.

We kindly ask you to send us your tools and hacks by Tuesday at the latest. This can be a single slide, text, or a short video.
We will prepare a slide deck with all the tools and hacks and ask everyone to briefly introduce their item(s) (in ~1 minute).

Please send your contribution(s) to martin.boissier@hpi.de.
 

Venue: F.E-06

Prof Dr. Felix Naumann

Title: Writing and receiving scientific reviews

Abstract: Scientific reviews are among the most important instruments to advance science. They serve to check and validate scientific results before publication and they serve to give important feedback to authors to improve their work. In this talk, we will explore best practices for writing and receiving scientific reviews in computer science, including strategies for providing feedback that is both helpful and actionable. We will also discuss the importance of maintaining a culture of helpfulness and respect within the scientific community. Receiving negative reviews can be a frustrating experience, but it is important to approach them with an open mind and a willingness to learn from constructive criticism, while keeping in mind that not all reviewers' feedback may be relevant or helpful.

Attendees are welcome to send in sentences from reviews they received that were particularly frustrating or particularly helpful (max one sentence for each category). I will try to include and discuss these examples: felix.naumann(at)hpi.de, deadline June 12.

Venue: L.E-03

Hendrik Rätz

Title: Are LLMs Becoming Conscious? (Discussion)

Abstract: Large Language Models (LLMs) such as ChatGPT are becoming increasingly more powerful and show exceptional conversational abilities, which might seem human-like. Naturally, this raises the question if these models are still only machines or if there's "more" behind them. Last year a developer at Google sparked controversy around this topic after conversing with Google's Chatbot LaMDA. He publicly claimed that this LLM is conscious and should be treated as such.
This talk will try to explain how much truth there is behind these claims and what could be possible indicators for consciousness. In addition, I want to discuss with you if we could see sentient AI in the near future.

Venue: F.E-06

Lawrence Benson

Title: Writing Less Platform-Specific SIMD Code in Databases.

Abstract: Modern query engines often use SIMD instructions to speed up query performance. As these instructions are heavily CPU-specific, developers must write multiple variants of the same code to support multiple target platforms such as AVX2, AVX512, and ARM NEON. This process leads to logical code duplication, which is cumbersome, hard to test, and hard to benchmark. In this paper, we make the case for writing less platform-specific SIMD code by leveraging the compiler’s own platform-independent SIMD vector abstraction. This allows developers to write a single code variant for all platforms as with a SIMD library, without the library’s redundant layers of abstraction. Clang and GCC implement the platforms’ SIMD intrinsics on top of their own abstraction, so code written in it is optimized for the underlying vector instructions by the compiler. We conduct four database operation microbenchmarks based on code in real systems on x86 and ARM and show that compiler-intrinsic variants achieve the same or even better performance than platform-intrinsics in most cases. In addition, we completely replace the SIMD library in the state-of-the-art query engine Velox with compiler-intrinsics. Our results show that query engines can achieve the same performance with platform-independent code while requiring significantly less SIMD code and fewer variants.

Russa Biswas

Title: Embedding based Link Prediction for Knowledge Graph Completion.

Abstract: Knowledge Graphs (KGs) are the most widely used representation of structured information about a particular domain consisting of billions of facts in the form of entities (nodes) and relations (edges) between them. Besides, the KGs also encapsulate the semantic type information of the entities. The last two decades have witnessed a constant growth of KGs in various domains such as government, scholarly data, biomedical domains, etc. KGs have been used in Machine Learning based applications such as entity linking, question answering, recommender systems, etc. Open KGs are mostly heuristically created, automatically generated from heterogeneous resources such as text, images, etc., or are human-curated. However, these KGs are often incomplete, i.e., there are missing links between the entities and missing links between the entities and their corresponding entity types. This talk focuses on addressing these two challenges of link prediction for Knowledge Graph Completion (KGC)

(i) General Link Prediction in KGs that include head and tail prediction, triple classification, and

(ii) Entity Type Prediction.

My work explores that contextual information about the entities obtained via random walks, textual entity descriptions, and under-treated features in a KG such as Wikipedia Categories are essential for improving the latent representation of the entities for KGC. The talk will also cover some interesting scopes for future research involving exploiting the proposed models in domain-specific KGs such as scholarly data, biomedical data, etc.

Venue: F.E-06

Rainer Schlosser

Title: Pricing Agents for R(e)commerce Markets

Abstract: Many sales applications have become highly dynamic and competitive. However, deriving successful strategies to estimate demand and to automatically update prices in the presence of competition is challenging. In this talk, we look at pricing strategies from (i) an analytical, (ii) a practical, and (iii) an AI perspective. Besides theoretical models and practical applications, we study self-learning agents applied to sustainable recommerce markets, which allow to trade in and resell used products.

Venue: F.E-06

Prof Dr. Boris Glavic

Bio: Boris Glavic is an Associate Professor in the Department of Computer Science at Illinois Institute of Technology leading the IIT 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.

Title: Uncertainty-Annotated Databases - A Lightweight and Expressive Model for Uncertain Data

Abstract: Uncertainty arises naturally in many application domains due to data entry errors, sensor errors and noise, uncertainty in information extraction and parsing, ambiguity from data integration, and heuristic data preparation & cleaning techniques. Analyzing uncertain data without accounting for its uncertainty can create hard to trace errors with severe real world implications. The database community has proposed incomplete and probabilistic databases as principled models for uncertain data. However, the class of queries that can be answered efficiently over such databases is severely limited. In this talk, I will introduce the audience to uncertainty-annotated databases (UA-DBs), a light-weight and practical uncertain data model. UA-DBs accommodate a wide range of uncertain data sources, can approximate many existing incomplete and probabilistic data models as well as can approximate the uncertainty arising naturally in many data cleaning techniques, and, most importantly, query answering over UA-DBs is efficient for a very large class of queries. Well-known hardness results for incomplete and probabilistic query processing imply that achieving efficient query evaluation over uncertain data requires some form of relaxation or approximation. In the case of UA-DBs, the answers to a query compactly encode an under-approximation of the query’s certain answers (all facts that are guaranteed to be true) and an over-approximation of its possible answers (all facts that could be true). In addition to presenting UA-DBs, I will also briefly introduce the implementation of UA-DBs in Vizier, a notebook system developed at IIT, SUNY Buffalo, and NYU.

Venue: L.E-03

Prof Dr. Bert Arnrich

Title: How to write a good research plan?

Abstract: A good research plan outlines the goals and nature of the doctoral thesis and the responsibilities of the doctoral student. Usually, the research plan has to be submitted to the supervisor within the first twelve months of enrollment. There is no strict format for the research plan.  According to ETH Zurich, a good research plan contains motivation, research theses, related work, research contribution, approach, evaluation, dependencies, risk and contingency plan, work plan, publication plan, and if applicable a course plan.  This presentation highlights the important components of a good research plan on concrete examples.

Venue: HS3

Prof Dr Ralf Herbrich

To register for the lecture please follow this link: https://hpi.de/open-campus/registrierung/2023/inaugural-lecture-prof-dr-ralf/

Time: July 26, 2023 at 4pm (CEST)

Venue: HS1 on campus I