Seminar Assignment Winter 2023/2024
The central registration for all computer science seminars will open on September 22nd.
This system is used to distribute students among the available seminars offered by the CS department. To register for any of the seminars, you have to register here until October 24th 23:59 CET. You can select which seminar you would like to take, and will then be automatically assigned to one of them on October 26th.
Please note the following:
- We aim to provide a fair mapping that respects your wishes, but at the same time also respects the preferences of your fellow students.
- Experience has shown that particular seminars are more popular than others, yet these seminars cannot fit all students.
- Please only select seminars if you are certain that you actually do want to complete a seminar this semester. If you have already obtained sufficient seminar credits, or plan to take other courses this semester, feel free not to choose any seminars. Students who drop out of seminars take away places from those, who might urgently need a space or are strongly interested in the topic.
- We encourage those students who wish to take a seminar this semester, to select their preferences for all available seminars, which eases the process to assign students that do not fit the overly popular seminars to another, less crowded one. So if you are serious about taking a seminar this semester, please select at least three seminars (with priority from "High" to "Low").
- If you urgently need to be assigned to a seminar in the upcoming semester, choose at least five seminars (with priority from "High" to "Low"). We will then guarantee that you will be assigned to a seminar (yet not necessarily one of your choice).
- If you are really dedicated to one particular seminar, and you do not want any other seminar, please select the "No seminar" as second and third positive option. However, this may ultimately lead to the situation that you are not assigned to any seminar. Also, choosing "No seminar" as second/third option does not increase your chances of getting your first choice.
The assignment will be automatically performed by a constraint solver on October 25th. You will be added to the respective seminars automatically and be notified about this shortly thereafter. Please note that the assignment cannot be optimal for all students if you drop the assigned seminar, i.e., make only serious choices to avoid penalty to others.
Seminars
Accelerating Applications with Specialized Hardware by Antoine Kaufmann
In this seminar, you will learn about various aspects of hardware acceleration, primarily through a series of hands-on projects. The projects focus on different aspects of the complete system, such as modifying an application to take advantage of an accelerator, implementing necessary software components to interact with the hardware, optimizing the interface between software and hardware, and even implementing (parts of) a hardware accelerator. Overall, the main goal is for you to completely understand the full picture of all system components involved and their interactions. The focus is on the principles and fundamental constraints rather than specific tools or applications.
Kick-Off Meeting: Mon, 30. Oct, 13:15-14:00, E1 5 Room 105 MPI-SWS Building
More Details: https://cms.sic.saarland/hwaccel_23/
Requirements: This seminar is open to senior Bachelor, Masters, and doctoral students. Ideally, students should have already taken courses on system architecture and operating systems. Bachelor students must have passed the basic courses on Programming 2 or equivalent. Proficiency in C(++) programming (including low-level aspects such as pointers and memory management) and UNIX development tools (e.g., shell, make, gcc, gdb) is strictly required to pass this course. Basic Verilog knowledge will also be required for the projects, but this can be acquired or refreshed as part of the project.
Places: 15
Advanced Theory of Secure Messaging by Cas Cremers
Millions of users benefit from modern secure messaging apps. Since recent years, these messaging apps can offer much stronger security properties than they ever did before, for example using the Signal protocol library (as used by WhatsApp, Signal's app, and Facebook's secret conversations) and others.
In this seminar, we will explore some of the theory behind the design of these protocols through a selection of relevant research papers in this area, which is a highly active current research direction. This will include papers on the novel security properties met by these protocols, the subtle edge cases of security in group messaging, as well as recent works that show the achieved security might be lower than expected.
Most (but not all) of the papers are technical, and include security models and proofs; however, for the presentations, understanding the proofs will not be mandatory. Having attended advanced cryptography or protocol analysis lectures can be beneficial, but is not strictly required.
More background on the instructor can be found here: https://cispa.saarland/group/cremers/index.html
Requirements: Students should have completed the core Security or core Cryptography lectures, ideally both.
Places: 10
Advanced Topics in Diffusion Modeling - From Theory to Implementation by Verena Wolf, Gerrit Großmann
Diffusion models have recently transformed the field of generative deep learning. This seminar will explore this vibrant area of research.
This seminar is targeted at students who already have a background in deep learning (theoretical and practical) and are keen to dive deeper into probabilistic diffusion and related concepts such as stochastic processes and normalizing flows.
Experience with diffusion models will be beneficial, but not required.
While we do not aim to provide comprehensive coverage of the topic, we have selected certain papers that we find exceptionally engaging, fun, and thought-provoking.
The seminar is equally divided into two segments: a classical part, where students present a concept based on a research paper, and a practical part, where students develop a tutorial notebook inspired by the ideas from that paper.
Organization:
- The seminar takes place Fridays from 14:15 to 16:00
- More information is available at: https://mosi.uni-saarland.de/lectures/23_2_deep_diffusion/
Requirements: General background knowledge and practical experience in deep learning are strongly recommended. Experiences with diffusion models will be helpful but are not necessary.
Places: 12
Advances in AI for Autonomous Driving by Matthias Klusch, Andreas Nonnengart, André Meyer-Vitali, Christian Müller
This seminar (AI4AD) is concerned with advances in autonomous driving research from the perspective of AI. In particular, we will take a closer look at selected techniques and systems of (neuro-symbolic) AI for autonomous driving with focus on pedestrian behavior prediction and collision-free navigation, and discuss their strengths and weaknesses.
Seminar webpage: https://www.dfki.de/~klusch/AI4AD-seminar-ws23
The topic list is already available from July 27, 2023 on, please check the seminar webpage for more information!
Seminar date and location: Tuesdays 16:15 - 18:00, on-site (DFKI, Bldg D3.2, Room "Turing 2") or virtually via zoom.
The introduction meeting with topic assignments takes place on Tuesday 31.10.2023, 16:15 - 18:00.
Requirements: This seminar aims primarily at advanced master students in Computer Science or Data Science and AI who preferably hold a B.Sc. degree in this or related field. Good knowledge in AI (introductory course on AI covering symbolic knowledge representation and reasoning, machine or deep learning, automated planning) is required. Strong interest in autonomous driving research and neuro-symbolic AI.
Places: 10
An Extravaganza of Algorithmic Models by Sebastian Brandt, Alexandre Nolin
For many a student, the first encounter with algorithmics may suggest that algorithms are well-behaved sequential creatures with full access to the input, living in a static environment, and otherwise privileged. However, the theory of algorithms goes well beyond this particular paradigm: there is an abundance of different kinds of (often less privileged) algorithms, including distributed algorithms, parallel algorithms, online algorithms, streaming algorithms, dynamic algorithms, and many, many more. In this block seminar we want to take a tour through the world of algorithmic models, with a special focus on graph problems. Note that this seminar will focus on theory (e.g., algorithm design and analysis), not implementation.
The seminar will take place as a block course in the spring break. Each student has to present one or two assigned paper(s). Each presentation is followed by a discussion lead by the presenter. Besides presentation and participation in the discussion, the grade will depend on written deliverables. More information can be found on the seminar webpage: https://cms.cispa.saarland/eam2324/
In particular, please check there whether the dates fixed for the seminar work for you. It might be possible to change the dates later if everyone involved agrees, but we cannot guarantee that.
Requirements: There are no formal prerequisites for this seminar, but a general interest in graph theory and designing/analyzing algorithms as well as a basic understanding of probabilities and algorithmic analysis (e.g., O-notation) will be helpful.
Places: 12
Aspects of Quantitative Program Verification by Benjamin Kaminski, Tobias Gürtler, Lena Verscht
Are you passionate about logic, verification, semantics, and alike? Are you tired of thinking black-and-white, true-or-false? Then come and study the more nuanced quantitative formal program verification! In quantitative verification, properties are not just true or false. Instead, we verify quantities like runtimes, error probabilities, beliefs, etc.
Topics which we will cover include:
- Probabilistic programming (a currently trending modeling paradigm in machine learning),
- The geometry of neural networks
- Incorrectness logic (the latest creation of the former chief formal methods researcher at Facebook)
- The flow of quantitative information through programs
- Worst-case execution times
- Verification of heap-manipulating programs
- and many more
The seminar website can be found here:
Requirements: The most important requirement: You should really really like math and/or logic. This seminar covers very theoretical work.
The following courses are mandatory and/or recommended.
Mandatory: Programmierung 1, Programmierung 2, Grundzüge der Theoretischen Informatik
Recommended: Semantics; Introduction to Computational Logic; Automata, Games and Verification;
Ideal: Verification
Places: 12
Block Seminar: Intelligent Systems for Supporting Human Learning by Tomohiro Nagashima, Man Su
Intelligent systems (e.g., Intelligent Tutoring Systems) have been developed and empirically evaluated to support human learning in a variety of domains (e.g., STEM). Recent advancement in AI technologies has allowed such systems to become even more effective through, for instance, adaptive tutoring (e.g., adaptive problem selection, and affect detection). In this block seminar, we will review papers on intelligent systems that support human learning and discuss future opportunities for further enhancing such systems. We will also pretend to be secondary-school learners and use some of the established intelligent systems to get hands-on experiences.
Learning objectives: By completing this seminar, you will be able to discuss the benefits and limitations of various intelligent systems designed for supporting human learning. You will also be able to generate new research and development ideas for the future of intelligent learning systems.
See more information here: https://tomonag.org/seminars/
Requirements: The seminar is open to both master’s and bachelor’s students in all departments across UdS campus. The seminar doesn’t require any prerequisite knowledge and is therefore open to students with any background, but knowledge or experience in one or more of these fields will help: Cognitive Science, Learning Sciences, and Human-Computer Interaction.
Active participation in class discussions and activities is expected and encouraged. I rather want to design learning experiences in this seminar with students – your active participation is the key to the success of the seminar!
Places: 12
Block Seminar Machine Learning for Natural Language Processing by Dietrich Klakow
Deep learning is the predominant machine learning paradigm in natural language processing (NLP). This approach gave huge performance improvements across a large variety of natural language processing tasks and in many other areas of computer science. We will have a more specific theme and topics that will follow.
For more information see:
Places: 8
Brain-Computer Interaction by Maurice Rekrut, Tobias Jungbluth, Johannes Ihl & Antonio Krüger
Brain-Computer Interfaces (BCI) have been a widely researched topic for the last 20 years. BCIs make use of brain activity to create controls for computers or machines in general. The most common approach to establish such a connection between computer and brain is to measure the electrical activity of the brain at the scalp surface, via Electroencephalography (EEG). Those EEG devices are non-invasive, comparably cheap and can nowadays even be used with dry electrode caps, which makes them easy to setup. Within this seminar you will learn about different techniques and types of BCIs and develop your own Brain-Computer Interface with a dry electrode EEG-headset in a group project.
The seminar consists of 3 stages.
In the first stage you will do a basic literature research on a topic/task that we provide to you. Each of those tasks covers a different topic in Brain-Computer Interaction. Those topics will be presented in the Kickoff meeting, including a short introductory lecture on different BCI techniques, providing the necessary background. After the presentation you can indicate your interest on one of those topics and we will try to assign you to a group according to your interest. Afterwards you will receive basic literature regarding this topic and perform a literature research on the state-of-the-art in this field.
In a second step you will create a concept and implementation plan to solve the given task. This implementation plan should be based on state-of-the-art methods that you researched on in the first step and will be presented to the seminar participants and tutors in a 20-minute presentation. The talk should introduce the task/problem you were working on, include the most relevant studies in this field, what impact those studies and their results have on your work and finally your implementation plan that you concluded on. The implementation plan will be discussed in the group and if accepted you can proceed to the implementation.
In the third stage you will implement the BCI-application with your group in the DFKI labs. We will provide you the headsets and other required hardware. In order to give you all an introduction to the headset that you will use (https://www.unicorn-bi.com/ ) there will be a tutorial on the software architecture and general information on how to get started with the headset. Afterwards you will continue with the implementation and at the end of the seminar each group will give a short demonstration of the implemented prototype and submit a summary of the results as report.
For more information about the seminar, visit our website:
https://umtl.cs.uni-saarland.de/teaching/winter-2023/2024/brain-computer-interaction.html
Requirements: No formal requirements, however, basic knowledge in Python programming, Machine Learning and Signal processing might be helpful.
Places: 15
Bridging Language in Machines with Language in the Brain by Mariya Toneva
Current natural language processing (NLP) models (e.g. ChatGPT, GPT-4, etc.) have impressive capabilities, but how closely do they actually align with the capabilities of the only system that truly understands complex language--the human brain? In this seminar, we will review work that studies the existing alignment between the representations of language constructed by NLP models and the representations of language in the human brain obtained from brain imaging devices, as humans and models process the same language input. We will discuss the reasons for existing alignment, and some of the established remaining gaps. We will additionally review works that aim to bring NLP models closer to the human brain. Lastly, students will have the opportunity to propose and complete related projects.
For more information, see: https://bridge-ai-neuro.github.io/
Places: 8
Complexity of Games by Daniel Marx
Popular games hide computationally hard, sometimes even undecidable, algorithmic questions such as whether a player has a winning strategy – and some argue that this makes these games popular in the first place. In this seminar, we study (generalizations) of popular games, such as board, card and video games, for example: Hanabi, Chess, Minesweeper and Portal. They might be zero, one or multi-player games, and interesting problems to study range from whether there is a forced win, to simply whether there is a legal move. These problems often lead us to complexity classes beyond NP such as to PSPACE.
Requirements: A solid background in algorithms and complexity. You should be familiar with NP and reductions to show hardness.
Places: 10
Computer-Assisted Proofs in LEAN by Bartholdi, Pernak
As mathematical research advances, researchers become more and more specialized, and the mathematics they produce becomes more and more complicated to verify.
The possibility to formalize and check proofs thanks to computer programs is thus more relevant than ever. What’s more, tremendous progress in recent years make it so that formalizing actual research level mathematics is possible, and formalizing student level mathematics is accessible to students.
In this seminar, students will practice with the LEAN proof assistant (https://leanprover-community.github.io/).
We meet weekly on Zoom, and discuss informally: each student gets a chance to speak, to explain the work they have done in previous weeks, and to plan ahead.
See the course's page to obtain the zoom link.
https://www.math.uni-sb.de/ag/bartholdi/cap/
Requirements: Students may obtain up to 8 credit points, by formalizing a theorem or new definitions, thus contributing to mathlib, the library which gathers all mathematics that has already been formalized in LEAN.
Topics suitable to both MSc and BSc students will be offered, no prior knowledge is required.
Places: 10
Data and Society by Ingmar Weber, Annika Haß, Till Koebe
From finding a mate to booking a holiday, our lives are increasingly mediated by online platforms. Digital traces left by these interactions provide opportunities to study societal phenomena while creating challenges around the responsible use of data. In this seminar, students will learn how computational methods and machine learning can be applied to study society through such data.
The first part of the seminar will familiarize students with existing work in computational social science with each week focused on a topic such as “love” or “food” and methods to quantify it. The second part of the seminar will be about projects in which students are asked to quantify a societal phenomenon of their choice using computational methods.
The overall course performance will be based on (i) overall course participation, (ii) assigned paper presentations, (iii) literature review and “project pitch” (prior to in-depth work), (iv) written project report, and (v) final project reports.
Apart from learning about interdisciplinary research and applications of machine learning, students will also learn research skills such as how to read and discuss papers, how to plan a project, how to present their work, how to write a scientific paper, and how to work in teams.
Students can take this course as either a seminar or a proseminar. Bachelor students taking the course will have a reduced load in the first part of the course.
Future details at https://cms.sic.saarland/das_23/. Timing is Tuesdays, 10am-noon.
Requirements: The project-based element of the seminar will require some programming and data analysis experience. Beyond that there are no formal requirements, though a desire to engage in interdisciplinary discussions and an interest in studying societal processes is a must.
Places: 7
Differential Privacy: Mathematical Foundations and Applications in Machine Learning by Franziska Boenisch; Jilles Vreeken
Since machine learning (ML) becomes increasingly prevalent in sensitive areas, such as healthcare and finance, it is crucial to ensure privacy. This seminar is centered around the mathematical framework of differential privacy, a current gold standard for privacy protection. Differential privacy, in simple terms, ensures that the outcome of a data analysis remains roughly the same when a single data point in the underlying private dataset changes.
Throughout the seminar, we will delve deep into the core principles of differential privacy, dissecting its precise definition and exploring various relaxations that allow for adaptable privacy guarantees. We will also dive into the practical aspects, examining the mechanisms and strategies used to achieve differential privacy ML. Furthermore, we will study different mechanisms for privacy accounting, a critical aspect of quantifying and validating the level of privacy afforded by differential privacy mechanisms. In the concluding section of our seminar, we will take a closer look at the practical implementation of differential privacy in state-of-the-art foundational models, such as large language models. Understanding how to apply differential privacy to these cutting-edge models is crucial for addressing privacy concerns in advanced ML systems.
Requirements: This seminar is open to senior Bachelor, Masters, and doctoral students. Ideally, students should have a solid background in mathematics through the base lectures, and at least a basic understanding of deep learning.
Each student will present a topic during the seminar hours in the form of an oral presentation. In addition, each student will read the relevant papers for the other students’ presentations, and hand in a seminar paper at the end of the semester.
Places: 12
Engineering of Interactive Systems with Generative AI by Marie Muehlhaus, Artin Saberpour, Jürgen Steimle
Today, interactive systems are typically designed by hand: Storyboards are drawn by hand, code and hardware is implemented manually, and the evaluation is conducted through time-consuming user studies. With the rise of Generative AI (GenAI), e.g., Large Language Models, Generative Adversarial Networks, Diffusion models, or Transformer-based models, this may change. In this seminar, we will investigate if and to what extent the manual design process of interactive systems could be improved and streamlined by GenAI models. This involves, for instance, the following considerations: How can GenAI models be used to assess real-world requirements, create personas, and identify stakeholders? How can they help in the actual design phase of the interface, for implementing software or hardware? And can they be used to evaluate the quality of an implemented outcome? In this seminar, we aim to find answers to these timely questions by conducting practical projects. We aim to identify opportunities, best practices and current limitations, to ultimately come up with principled recommendations as to how GenAI can speed up, improve or even replace the manual design process.
The seminar comprises a combination of readings, presentations, and a group project in which students will first familiarize themselves with the basic concepts of Generative AI. In small teams, students will then re-design how we engineer interactive systems by systematically investigating and reporting on ways to alter the design process of interactive systems with Generative AI.
Requirements: - For the registration, we require you to submit a brief motivation statement that elaborates on why you want to take this class and what relevant projects and courses you have taken before.
- It's highly recommended to have passed Interactive Systems and/or HCI.
- It's advantageous but not strictly required to have some background in topics such as Artificial Intelligence, Generative AI, and/or Large Language Models.
- This is a Master-level seminar, which can also be taken by Bachelor students from the 5th study semester and above.
Places: 15
Eye tracking: experimental investigations in language comprehension and machine learning approaches to data analysis by Iza Škrjanec, Margarita Ryzhova, and Vera Demberg
The present seminar consists of two blocks.
In the psycholinguistic part, we will focus on how eye-tracking measures can be used to test which cognitive processes are involved in reading and what individual differences can be inferred on the basis of different reading behaviors. The collection of papers is primarily focused on lexical access and syntactic complexity in sentence and paragraph reading.
The machine learning part focuses on the processing aspects of modeling reading data, as well as on the recent eye-movement-based classification approaches.
In the first lectures, we will introduce eye-tracking experimental methodology and measures that are typically collected, and present a list of papers that will be covered in this seminar.
Key info:
- The seminar takes place on Fridays, 12:15-13:45
- The first lecture is on 03.11 where we will present the papers that will be discussed in the seminar
- The papers will be distributed among the students in the second lecture (10.11)
- Every week one student presents a paper except for the first 3 lectures, where we will provide some background on eye-tracking, psycholinguistics, and machine learning concepts
- More information to come in MSTeams (a link to be provided)
Requirements: A general interest in eye-tracking and psycholinguistics is expected. As you sign up for the seminar, please also fill out the following form: https://forms.office.com/e/qdkDAqkbWn If you plan to attend the seminar, filling out this form is mandatory (deadline is October 15). We will use your answers to adjust the context of the first introductory lectures.
Places: 5
Eye tracking for intelligent systems - a pratical research seminar [Blockseminar] by Anna Maria Feit, Michael Feld, Michael Barz, Amr Gomaa
You have learned about eye tracking and its potential for reading the user's mind? Looking into someone's eyes can tell you a lot about their state of mind, their intentions or their next actions; whether they are daydreaming or concentrating, whether they understand what they are reading or whether they are going to eat that last cookie.
Now it's time to put theory into practice and discover the transformative potential of eye tracking technology for creating intelligent interfaces and enhancing user experiences.
In this hands-on research seminar you will work in groups of 3-4 people on a practical project of your choice based on scientific papers. You can set out to advance research by analyzing the gaze behavior of people or develop new innovative applications that make use of explicit or implicit gaze input. Whether you want to focus on empirical research or practical applications, in either case this seminar gives you hands-on experience in eye tracking development and in conducting empirical research studies.
To ensure that students are able to focus on the practical implementation, we offer this seminar as a block course in March 2023. To prepare for this execution phase, we will meet 2-3 times during the semester to help you develop your idea, plan your project, and pitch it to your classmates. Throughout the seminar, each group is supervised by a teacher. Regular check-ins and peer discussions will ensure progress and foster a collaborative environment.
The seminar will culminate in a research showcase, where students present their
projects and findings.
Key Info:
- Groups of 3-4 students
- 2-3 in-person meetings distributed over the WS 23/24 on Mondays, 2-4 PM (attention mandatory)
- practical implementation during all of March 2024 with regular in-person meetings with your project group, your teacher, or all students.
Requirements: This seminar is intended for students who have already learned about eye tracking in other seminars or courses. Thus, you must already:
- be familiar with eye tracking technology
- bring theoretical foundations of how eye tracking is used for interactive applications or user research
You must specify your knowledge and how you obtained it in the motivation text below.
Places: 15
Hot Topics in Data Networks by Anja Feldmann
While the Internet has started as a research effort, it has consistently evolved throughout the decades into the largest commercial network. Today, many research efforts focus on understanding the Internet's structural trends, optimizing its packet delivery, and laying the foundations for its future developments. In this seminar, you will receive a closer look at bleeding-edge research published at the top conferences in our domain. Discussing state-of-the-art approaches and recent findings from a broad range of network-related topics with your peers and instructors will provide you with a deep understanding of your assigned topic. Preparing the accompanying research survey and topic presentation will not only strengthen your academic writing but also help you to present your future work in a more accessible and structured way.
Requirements: Participants should have successfully participated in "Data Networks" or an equivalent course.
Places: 20
Individual differences in language processing by Jia Loy, Alexandra Mayn, and Vera Demberg
It has long been observed in psycholinguistic research that group-level patterns can obscure variation that exists at the level of the individual. More recently, there has been growing interest in investigating the source(s) of this individual variability (e.g. cognitive, linguistic, environmental, or other factors), and how they come into play in the context of language processing.
In this seminar, we will review papers that take an individual differences approach to the study of language processing. We will examine the role of individual differences across a range of psycholinguistic phenomena, the reliability and limitations of various measures used, and the overall utility of an individual differences approach to language research. In the process, we will explore how studying these differences can provide insight into broader questions relating to the mechanisms that underlie language processing.
Requirements: No prerequisites, but most of our papers deal with psycholinguistics phenomena, so students should at least have an interest in that area.
Places: 6
Interpolation and Approximation Methods for Visual Computing Algorithms by Michael Ertel und Prof. Dr. Joachim Weickert
Interpolation has been introduced to determine a continuously-defined function from given discrete data, often the function is constructed in such a way that the function agrees perfectly at the given measurement points. However, the given data is not always reliable and approximating the perturbed data with a smooth function instead can be beneficial. The function can then be used to approximate for example the derivative in a certain point.
In the meantime the application field has expanded to surface reconstruction, kernel learning, estimation of image structures and numerical solution of partial differential equations to solve, to name a few. In this seminar, we firstly cover basic concepts and move on to forefront research in the field.
For more information visit the website https://www.mia.uni-saarland.de/Teaching/iamvca23.shtml
Requirements: The seminar is for advanced bachelor or master students in Visual Computing, Mathematics, or Computer Science. Basic mathematical knowledge (e.g. Mathematik für Informatiker I-III) and some knowledge in image processing and computer vision is required.
Places: 12
Language Models at the Intersection of Cognitive Science and Software Engineering by Prof. Dr. Sven Apel, Alisa Welter, Kallistos Weis, Christof Tinnes, Anna-Maria Maurer, Christian Hechtl, Sebastian Böhm
While Machine Learning has revolutionized numerous fields, its capabilities remain largely unattained in software engineering. What's the reason? Inspired by breakthroughs in other sectors, this seminar offers a thrilling exploration into the convergence of Language Models, Software Engineering, and Cognitive Science.
We will scrutinize a broad spectrum of existing contributions and potential future developments, spanning from code generation and model completion to effective collaboration in large-scale projects. The seminar will also illuminate current research in cognitive science, specifically focusing on human code comprehension.
More information: https://cms.sic.saarland/se_seminar_ws23_24/
Kick-Off Meeting: 2.November 12:00-14:00
The seminar takes place Thursdays from 12:00 -14:00 (4 appointments in sum)
Requirements: This seminar is open to Bachelor and Master students. Ideally, students should have already taken courses in Machine Learning, Deep Learning and Software Engineering. But this is not required.
Some foundational knowledge acquired through courses such as Programming 2, the Software Lab, or similar, would be beneficial.
Places: 9
Machine Learning Security Reproducibility by Lea Schönherr
The intersection between security and machine learning can be viewed from two perspectives: The security of machine learning algorithms and systems, e.g., adversarial examples and poisoning attacks. Second is the use of machine learning methods to improve and analyze the security of a system, e.g., malware detection or decompilation. In this seminar, we will cover recent publications from both sides by reading and summarizing the state-of-the-art on these two topics and performing an artefact evaluation of their code to verify and comprehend the practical implementations of the latest scientific publications.
The seminar is structured into two parts. In both parts, you will work in groups of two:
- you will write a short survey paper on the main topic of your assigned paper.
- you will evaluate the code of the paper during an artefact evaluation.
Additional information can be found at https://cms.cispa.saarland/mls_2324/
Requirements: This seminar aims primarily at master students in Computer Science or related fields. Previous experience in machine learning and computer security is beneficial.
Places: 12
Microarchitectural Weird Machines: Theory and Practice by Jan Reineke
Weird machines are computational artifacts where machines are operating outside of their intended specification. The concept of weird machines has originally been introduced as a theoretical framework to understand the existence of exploits for security vulnerabilities. A particular class of weird machines are microarchitectural weird machines, where artifacts of a processor's microarchitectural implementation give rise to unintended "weird" behavior. For example, recent work has shown how speculative execution and caches can be combined to emulate logic gates and to combine these logic gates to create a Turing-complete weird machine that operates entirely in "microarchitectural space" and is thus invisible at the traditional hardware-software interface.
The goal of this seminar is to study microarchitectural weird machines from both a theoretical and a practical perspective. We will start by reading recent literature on the concept of weird machines and microarchitectural weird machines. In the second part of the seminar, we will explore microarchitectural weird machines in a practical and playful manner. In small teams, students will be (a) reproducing existing microarchitectural weird machines from the literature, and (b) experimentally evaluating these weird machines in novel ways.
Some relevant literature:
- Weird machines, exploitability, and provable unexploitability
Thomas Dullien
- Mismorphism: The Heart of the Weird Machine
Prashant Anantharaman, Vijay Kothari, J. Peter Brady, Ira Ray Jenkins, Sameed Ali, Michael C. Millian, Ross Koppel, Jim Blythe, Sergey Bratus, Sean W. Smith
- The ghost is the machine: Weird machines in transient execution
Ping-Lun Wang, Fraser Brown, Riad S. Wahby
- Computing with time: Microarchitectural weird machines
Dmitry Evtyushkin, Thomas Benjamin, Jesse Elwell, Jeffrey A. Eitel, Angelo Sapello, Abhrajit Ghosh
- The Gates of Time: Improving Cache Attacks with Transient Execution
Daniel Katzman, William Kosasih, Chitchanok Chuengsatiansup, Eyal Ronen, Yuval Yarom
- Optimization and Amplification of Cache Side Channel Signals
David A. Kaplan
Requirements: Bachelor of Science in Computer Science or a related discipline.
Alternatively: Passed all mandatory courses from the first four semesters of our Bachelor's programs in CS, Cybersecurity, DSAI, etc.
The seminar is also open to PhD students.
The practical part of the seminar will involve low-level programming. Thus some experience in C/C++ or other low-level languages would be highly beneficial.
The timing of the course and the kick-off time will be decided based on the students' availabilities.
Places: 12
Modern Hashing and Filtering Algorithms by Jens Zentgraf and Sven Rahmann
Amidst the verdant halls of this learned gathering, we shall embark on an expedition of discovery, unveiling the secrets of modern hashing and filtering algorithms. Behold! Like precious jewels of knowledge, we shall delve into the realm of new minimal perfect hash algorithms - the illustrious BBHash, the enigmatic RecSplit, the venerable PTHash, and their companions of mystery and might. Furthermore, our journey shall lead us to the domain of modern filters, where we shall encounter the Hierarchical Interleaved Bloom Filters, the elusive XOR Filters, the resplendent Binary Fuse Filters, and more.
With diligence and craftsmanship akin to that of the Dwarven smiths, you, dear students, shall take on the task of reimplementing and honing these wondrous algorithms. In the crucible of benchmarking, the truth of their prowess shall be revealed, as you strive to replicate the sacred results bestowed upon us. Fear not, for in the end, the fruits of your labor shall be distilled into a succinct seminar thesis, a tome of knowledge encapsulating your findings. And lo, when the appointed hour arrives, you shall take the stage as bard and scholar, regaling your peers with a captivating presentation, enlightening them with the wisdom you have gleaned from the cryptic world of algorithms.
[transformed by ChatGPT from our original description in the style of Tolkien]
Requirements: Basic knowledge of algorithms and data structures
Places: 8
Monte Carlo Ray Tracing by Philipp Slusallek, Qingqin Hua, Pascal Grittmann
Monte Carlo ray tracing is a popular technique to render realistic images. It is used for movies, architecture, video games, and product design. This seminar looks at a broad range of methods to make rendering via Monte Carlo ray tracing more efficient. We will look at rendering algorithms, from a basic path tracer to photon mapping, at sampling and material models, Markov chains and path guiding, and more. Each student will give a presentation on an assigned paper, write a short summary, and reproduce the key idea in a simplified setting.
Requirements: Basic understanding of computer graphics is assumed, for example through our Computer Graphics core lecture.
Places: 15
New Developments in PETS by Wouter Lueks
Technology is everywhere. We carry extremely powerful computers, also known as phones, everywhere we go. We interact with friends, family and colleagues via online platforms, and a large fraction of the information we consume comes from online platforms. On the one hand, digitalization brings enormous benefits. On the other hand, it makes it almost much easier to violate user's privacy, to surveil large fractions of a population, and sometimes even to control or influence what people do and think.
In this seminar, we will look at privacy-enhancing technologies, digital means that can help counteract this reduction in privacy caused by increasing digitalization. We will read and discuss new and seminal papers to learn about new techniques and ideas in the field of privacy-enhancing technologies. You will learn how to critically analyze and present existing research papers, as well as how to write your own small paper.
Organization. Roughly each week, we will discuss a different topic. In particular:
- Each student will prepare a presentation for their assigned topic based on the assigned (lead and follow-up) papers
- For two topics (that students do not present) they will write a review and prepare questions for the follow-up paper
- Students write a short seminar paper based on their assigned topic.
All students are expected to attend and participate in the discussions.
More information can be found here: https://cms.cispa.saarland/newpets23/
Requirements: The course has no formal requirements but preference will be given to Master students in Computer Science and related fields. A basic understanding of security and cryptography (as taught for example in CySec1/CySec2 or the Security course) is essential to be able to follow the material in this course. Having taken the Privacy-Enhancing Technology class will really help, but is not essential.
Places: 10
On the Practice of System and Network Engineering by Anja Feldmann, Tobias Fiebig
In this block-seminar (in the lecture-free period between Winter semester 2023/2024 and Summer semester 2024), we will go into the practical side of system and network engineering, covering all relevant aspects:
From the humans involved to the bits of underlying technology.
# Learning Objectives
By the end of this course, students will be able to:
- Describe best practices in system administration, from organizing work over service operation practices to operational management.
- Assess a given case and identify shortcomings in the situation
- Propose approaches for given cases based on identfied shortcomings or insufficiently implemented best-practices
- Select best-practices for given tasks (e.g., service deployment, change-management, efficacy improvement, budgeting) and apply them to a given case
- Describe the interdisciplinary nature of system administration and identify how it manifests in given cases
# Learning Activities
The seminar will follow a flipped-classroom model.
Prior to the course, a reading list pairing sessions with chapters will be provided.
Students are expected to read the material before the session.
During the session, the focus will be on hands-on activities and case discussions in the group.
# Reading Material
The seminar is based on:
The Practice of System and Network Administration
Volume 1: DevOps and other Best Practices for Enterprise IT
ISBN 9780321919168
Students are expected to read this book.
Furthermore, the following material is suggested as further reading:
Kaur, M., Parkin, S., Janssen, M., & Fiebig, T. (2022). “I needed to solve their overwhelmness”: How system administration work was affected by COVID-19. Proceedings of the ACM Human-Computer Interaction (Proc. CSCW 2022), 6, CSCW2. doi:10.1145/3555115
Holzbauer, F., Ullrich, J., Lindorfer, M., & Fiebig, T. (2022). Not that Simple: Email Delivery in the 21st Century. In USENIX ATC ’22, USENIX Annual Technical Conference. Carlsbad, CA, USA: USENIX Association. Retrieved from https://www.usenix.org/conference/atc22/presentation/holzbauer
Kaur, M., Ramulu, H. S., Acar, Y., & Fiebig, T. (n.d.). “Oh yes! over-preparing for meetings is my jam :)”: The Gendered Experiences of System Administrators. Proceedings of the ACM Human-Computer Interaction (Proc. CSCW 2023). https://pure.mpg.de/rest/items/item_3477431_1/component/file_3477432/content
Fiebig, T., & Aschenbrenner, D. (2022). 13 Propositions on an Internet for a “Burning World.” In TAURIN+BGI ’22, ACM SIGCOMM 2022 Joint Workshops on Technologies, Applications, and Uses of a Responsible Internet and Building Greener Internet. Amsterdam, The Netherlands: ACM. doi:10.1145/3538395.3545312
Fiebig, T., Gürses, S., Gañán, C. H., Kotkamp, E., Kuipers, F., Lindorfer, M., … Sari, T. (n.d.). Heads in the Clouds? Measuring Universities’ Migration to Public Clouds: Implications for Privacy & Academic Freedom. Proceedings on Privacy Enhancing Technologies Symposium (Proc. PETS 2023), 2023(2). https://pure.mpg.de/rest/items/item_3480830_3/component/file_3482426/content
Dietrich, C., Krombholz, K., Borgolte, K., & Fiebig, T. (2018, October). Investigating system operators' perspective on security misconfigurations. In Proceedings of the 2018 ACM SIGSAC Conference on Computer and Communications Security (pp. 1272-1289). https://dl.acm.org/doi/pdf/10.1145/3243734.3243794
# Assessment
Assessment will take place via two presentations held by students. Both presentations contribute equally to the grade.
Requirements: Participants should have a sufficient working knowledge of the Internet and how digital infrastructure in general works, e.g., by having successfully participated in "Data Networks", an equivalent course, or by demonstrating relevant prior practical experience.
Places: 35
Passwordless Authentication on the Web by Katharina Krombholz, Maximilian Golla
The seminar will give students a deep understanding of typical passwordless user authentication schemes enabling them to reason about their usability, deployability, and security properties.
Starting with how passwords are used, why they are not secure, and how one can reinforce them, the seminar will shed light on secure and usable alternatives to passwords on the Web. We will review papers that study usability issues, misconceptions, security guarantees, or report about real-world deployments of passwordless authentication solutions like passkeys, hardware security keys, and legacy solutions like single sign-on or magic links.
Students will read and summarize various recent scientific publications on passwordless user authentication solutions. During a hands-on experience students are asked to present one passwordless authentication system, reason about its usability benefits and need to explain how this authentication system might fail in practice.
Requirements: Many key concepts in computer security come up in user authentication, from cryptography, over usability, to Web standardization processes. This seminar aims primarily at master students in Computer Science or related fields, as it requires a basic understanding of the full stack of how security is built in real systems.
Places: 12
Privacy of Machine Learning by Yang Zhang
Machine learning has witnessed tremendous progress during the past decade, and data is the key to such success. However, in many cases, machine learning models are trained on sensitive data, e.g., biomedical records, and such data can be leaked from trained machine learning models. In this seminar, we will cover the newest research papers in this direction.
Requirements: Students are required to have basic knowledge of machine learning.
Places: 12
Pruning deep neural networks for lottery tickets by Rebekka Burkholz
Deep learning has achieved major breakthroughs in a variety of tasks. Yet, it comes at a considerable computational cost, which is exaggerated by the recent trend towards ever wider and deeper neural network architectures. Instead, many problems can be solved with the help of extremely sparse neural network architectures but finding and training them is a non-trivial task. According to the recent lottery ticket hypothesis, such sparse architectures can be identified by pruning large randomly initialized neural networks.
In this seminar, we will present recent algorithmic advancements in this direction, gain theoretical insights into the existence of lottery tickets, identify open problems, and discuss common challenges in the quest for winning lottery tickets.
Course organization:
-Kick-off meeting in the first or second week of the semester.
-We will have a block course in the spring break (date and time tbd at kick-off meeting) with one presentation and discussion of a topic per person.
-Note that some (light load of) written deliverables will be due before the block course meeting and a seminar paper one week after the meeting.
More information can be found here:
https://cms.cispa.saarland/pdl23/
Requirements: The course has no formal requirements but preference will be given to Master students in Computer Science and related fields with prior knowledge in deep learning and a convincing motivation.
Places: 12
Quantum Computer Vision and Machine Learning (QCVML) by Vladislav Golyanik
Computer vision (CV) studies the automatic processing of visual and spatial information in the form of 2D images, depth maps, 3D point clouds, and different combinations of those (possibly along with other sensory signals). The long-term aim of computer vision is inspired by the capabilities of the human visual system (HVS) to come up with intelligent and high-level interpretations of the observed scenes. It manifests a tight intertwining between CV and machine learning (ML): While modern CV strongly relies on ML techniques, developments in ML are often driven by challenging CV problems.
A quantum computer (QC) is a computing machine which takes advantage of quantum effects such as quantum superposition, entanglement, tunnelling and contextuality to solve problems notoriously difficult (belonging to challenging complexity classes such as NP) for a classical computer. Thanks to the exponentially increasing investment in the technology, quantum computers are gradually moving from the realms of theory towards actual devices. Albeit restricted, experimental realisations of numerous quantum algorithms have demonstrated improved computational performance bringing the community closer to the desired supremacy in recent years.
The aforementioned premises of quantum computing and annealing has led to the popularisation of quantum computer vision (QCV), where researchers started to port existing computer vision problems into forms amenable to quantum computation. However, re-framing the existing problems in the context of this new computing paradigm is not trivial. For instance, existing literature relaxes most of the (discrete) combinatorial search problems to continuous ones, whereas quantum annealing is great at optimisation on discrete binary variables. Hence, oftentimes we are required to revise the problem formulations at hand altogether. As one of the most massively data-producing disciplines, computer vision could become a particular profiteer from QC. This seminar series will cover advanced research topics that cross the boundaries between the fields of Computer Vision, Machine Learning and Quantum Computing.
Seminar's web page: https://4dqv.mpi-inf.mpg.de/teaching/QCVML2023/index.html
Requirements: The seminar targets computer science students (mathematics and physics students are welcome to attend). Knowledge of linear algebra and optimisation (B.Sc. level) and some experience with visual computing (e.g., accomplished courses/seminars in image processing and computer graphics) is a requirement. While the seminar assumes an interest in learning quantum computational paradigms, knowledge of quantum computing and quantum physics is not a prerequisite (the necessary notions will be conveyed and acquired during the seminar).
Places: 10
Quantum Cryptography by Nico Döttling, Anne Müller, Riccardo Zanotto, Jesko Dujmovic
In the mid 1970s, the late Stephen Wiesner circulated a manuscript with a revolutionary idea: He proposed to leverage the laws of quantum mechanics, specifically the famous uncertainty principle, to devise a "quantum money scheme". In such a scheme banknotes are digital quantum information, which can be transmitted at the speed of light between users, but have the remarkable property that they are encoded in such a way that they are "unclonable". That is, a user cannot spend a banknote and keep a copy of it at the same time. This is impossible using classical information alone, as classical information can be copied arbitrarily. This work was eventually published in 1983 and was the igniting spark that initiated the field of quantum cryptography. Just like with Wiesner's original proposal, the general goal of quantum cryptography is to achieve functionalities or security notions that are impossible to achieve classically. In this seminar we will look at seminal works in quantum cryptography, such as the Bennett-Brassard quantum key distribution scheme or the BCMVV'18 quantum testing scheme.
The seminar proceeds as follows
- In the first meeting we will assign topics on a first-come-first-serve basis
- In the second phase students prepare a report and presentation, reporting to their assigned supervisor
- Towards the end of the semester there will be a block seminar where every participant needs to attend and present their topic
- At the end of the semester a written report on the assigned topic needs to be submitted.
- The final grade will be determined based on both the presentation and the report
Requirements: - Attended and passed the core lecture cryptography
Places: 8
Reinforcement Learning with Large Language Models by Adish Singla, Goran Radanovic
The course will provide an overview of recent advancements in research that study reinforcement learning (RL) with large language models (LLMs). The course material will comprise research papers covering three different perspectives: (a) utilizing RL in training LLMs, (b) utilizing LLMs in training RL agents, and (c) foundation models for sequential decision-making. This course will familiarize participants with state-of-the-art techniques that bridge RL and LLMs.
The course will consist of three components as follows:
(1) Research papers: During the first half of the semester, students will be assigned about 6 research papers and have to write a short report for each paper. These reports will be due during the semester, about one report per week.
(2) Project: During the second half of the semester, students will work on an implementation project.
(3) Final presentation: At the end of the semester, students will give a 25 mins presentation for one of the papers and the project.
There will be no weekly classes. We will schedule office hours where students can receive feedback on their projects.
Requirements: There are no specific course prerequisites; however, please note the following points:
(1) It would be beneficial for students to have some background in topics such as artificial intelligence, reinforcement learning, natural language processing, or software engineering.
(2) We are requesting you to provide a short motivation letter explaining the reasons why you are interested in taking this seminar. Please mention any relevant project(s) you have done or relevant course(s) you have taken. You can provide this information in the text box below.
Places: 12
Seminar Legal Tech und eJustice by Prof. Dr. Sorge
Studierende erhalten während der Vorbesprechung ein Thema und müssen eine Seminararbeit hierzu anfertigen. Innerhalb des gewählten Themas sind die Studierenden in der Schwerpunktsetzung grundsätzlich frei.
Vor der Abgabe der fertigen Seminararbeit ist die Einreichung eines Abstracts nebst einer vorläufigen Gliederung vorgeschrieben. Die Besprechung der Abstracts findet verpflichtend statt, um inhaltlichen Missverständnissen entgegenzuwirken und den Studierenden bereits ein Feedback durch die Betreuer zu ermöglichen.
Danach ist die Abgabe eines „Preprints“ der Seminararbeit erforderlich. Daraufhin findet ein Peer Review der abgegebenen Preprints durch die weiteren Teilnehmer des Seminars statt. Jedem Studierenden werden hierfür 3-4 Paper anderer Studierender zugeteilt, welche von ihm/ihr begutachtet werden. Schließlich finden – in der Regel kurz vor Ende der Vorlesungszeit – die Vorträge statt. Abschließend werden die fertigen Seminararbeiten abgegeben.
Allgemeines: Es gilt der Grundsatz „Tiefe vor Breite“. In der Regel werden breit formulierte Seminarthemen gestellt. Teil Ihrer Aufgabe ist es, selbständig die Literatur zu sichten, sich einen Überblick über das Thema zu verschaffen und sodann einen Schwerpunkt (oder ggf. auch zwei, falls es sich im Einzelfall anbietet) zu setzen. Sie sollen zeigen, dass Sie sich in ein Forschungsthema aus Ihrem Fachgebiet einarbeiten und Sachverhalte aus diesem Forschungsthema vollständig durchdringen können. Das zwingt Sie, viele ebenfalls interessante, angrenzende Aspekte wegzulassen. Ein wesentlicher Teil der Schwierigkeit der Aufgabe ist die Entscheidung, was Sie weglassen können (oder müssen) und welche Aspekte für das Verständnis Ihres Themas wesentlich sind.
In der Einleitung geben Sie eine Motivation und ordnen ihr Thema ein; ggf. bietet sich ein weiterer Abschnitt zu verwandten Arbeiten an, um den Stand der Forschung darzustellen. Dann sollten Sie aber in die Tiefe des von Ihnen gewählten Schwerpunkts gehen. Den Abschluss bildet ein kurzes Fazit.
Abstract: Einarbeitung in das Themengebiet, Literaturrecherche und Studium erster wissenschaftlicher Veröffentlichungen. Auf dieser Grundlage Konkretisierung des gewählten Themas bzw. Wahl des Schwerpunkts innerhalb des Themas sowie die Entscheidung über einen sinnvollen Aufbau. Das Abstract ist eine kurze und aussagekräftige Beschreibung des konkretisierten Themas auf mindestens einer Seite (aber nicht mehr als 1,5 Seiten). Eine erste Gliederung sollte mit dem Abstract eingereicht werden.
Preprint: „Vorversion“ der Seminararbeit, wobei der aktuelle Stand ersichtlich sein muss. Hierbei muss es sich um keine bereits vollumfänglich fertige Arbeit handeln, jedoch muss der Inhalt und das Ziel der Arbeit sowie die Auseinandersetzung mit der einschlägigen Literatur erkennbar sein. Literaturverzeichnis, Gliederung und Schwerpunkt sind notwendige Bestandteile.
(Peer) Review: Feedback, welches sich inhaltlich auf das gesamte Paper erstrecken muss (Quellen, Aufbau/Gliederung, inhaltliche Nachvollziehbarkeit, klare Schwerpunktsetzung sprachliche/grammatikalische Schwächen etc.). Dieses muss aus mind. 400 Zeichen bestehen und soll die Möglichkeit einer ersten Einschätzung der Arbeit geben. Unschädlich ist, dass Juristen Themen der Informatik oder umgekehrt bewerten.
Vortrag: 20 Minuten mit anschließender Diskussion (30 Minuten bei Vorträgen von Zweier-Teams). Die Vorträge sollten sich an ein Publikum aus dem eigenen Fach (also Informatik bzw. Jura) richten, aber ohne spezielle Kenntnisse im Thema des Seminars verständlich sein. Idealerweise sind die Vorträge so gestaltet, dass Fachfremde (also z.B. Juristen bei Informatikvorträgen) die grundlegende Problemstellung erfassen können; eine Verständlichkeit des ganzen Vortrags für Fachfremde ist aber nicht erwartet.
Die Gesamtnote ergibt sich aus dem Preprint, den verfassten Reviews zu den Ausarbeitungen der anderen Studierenden, dem Vortrag und der finalen Seminararbeit.
Die Nichtabgabe des Abstracts, des Preprints, der Reviews oder der finalen Seminararbeit sowie ein Nichterscheinen zum Vortragstermin bzw. den Vortragsterminen führt zum Nichtbestehen des Seminars.
Termine:
Vorbesprechung: Montag, 30.10.2023 / 18.00 Uhr (s.t.)
Abgabe Abstracts: 19.11.2023
Abgabe Preprint: 07.01.2024
Abgabe Reviews: 17.01.2024
Vorträge: 05.02 und 06.02.2024 (Anwesenheit an beiden Tagen wird vorausgesetzt)
Finale Abgabe der Seminararbeiten: 25.02.2024
Requirements: Es wird erwartet, dass Teilnehmer in der Lage sind, Vorträgen in deutscher Sprache zu folgen.
Places: 7
Static Program Analysis Lab by Andreas Zeller + Jordan Samhi
How can one reason about program code? In this advanced seminar, we study several seminal approaches to static code analysis and debugging and implement them all (at least in a basic way). Our set of techniques includes:
* Control Flow Analysis
* Data Flow Analysis
* Symbolic Reasoning
* Points-to Analysis
* and more!
The general process will be as follows: Each week, you get 1-2 reading assignments and write an abstract about them. We may also ask you to give an (ungraded) five-minute short presentation to kick off the discussion and improve your presentation skills. Having discussed the approach, you have another week to finish a programming assignment (using Python and Jupyter Notebooks). in which you implement the respective technique on top of the given code.
At the end of the seminar, you give a 15-20 minute presentation on one of the techniques, including experiments you designed and conducted. We will determine your final grade from your abstracts (10%), your programming assignments (30%), and the final presentation (60%).
For further details, see https://cms.cispa.saarland/static2324/
Requirements: This seminar requires creativity and ambition. Experience with programming languages and logic reasoning is a plus. Prior knowledge in automated testing, debugging, and software engineering (esp. from earlier courses) will be beneficial. In your motivation, please mention relevant projects and courses you have taken along with your grades.
Places: 10
Systems for Large (Language) Models by Krishna Gummadi + Laurent Bindschaedler + Vedant Nanda + Till Speicher
In this course, we will go over various systems advances that can enable training billion parameter transformer models, such as GPT-3 and LLaMa. Such “foundation models”, trained on internet-scale data have enabled breakthroughs in generative AI such as ChatGPT. However, working with these models is often considered impossible without access to industry scale GPU clusters, primarily due to the sheer size of these models. We will go over advances in systems that democratize research on such large scale models and make working with such models possible even with modest resources. By the end of this course, you will learn how to train and finetune state-of-the-art large models that contain billions of parameters and cannot fit into a single GPU’s memory. While we will primarily work with transformer based language models, the focus will be on learning general principles that can be broadly applied to any large model.
For more details, please check out the course webpage: https://sysllm.mpi-sws.org.
Requirements: The course will require being able to grasp involved concepts and quickly implement and test out ideas in code and thus requires a good level of familiarity with deep learning frameworks such as PyTorch, as well as libraries such as Huggingface, Lightning, and DeepSpeed. Familiarity with C/C++/CUDA is necessary. Additionally, this course also requires some level of research maturity, since we will be critiquing some state of the art papers, with the goal of identifying possible research projects in this nascent space of systems for LLMs.
Places: 10
Techniques for Building Scalable and Robust Web Applications by Jens Dittrich
In this seminar, we will learn about various techniques that you need to know to make web applications scalable and robust.
Requirements: sound knowledge of the content from the Big Data Engineering course, i.e. you passed that course or a comparable course
Places: 7
The Web Security Seminar by Giancarlo Pellegrino, Cristian-Alexandru Staicu, Aurore Fass, Ben Stock
The Web Security Seminar will teach students to present, analyze, discuss, and summarize papers in different areas of Web security. The seminar combines a reading group with (almost) weekly meetings and a regular seminar, where students will write a seminar paper.
Each student will get a topic assigned, consisting of a lead and a follow-up paper. The student will present the follow-up paper in a 20-minute presentation followed by a 10-minute Q&A. Afterwards we will all discuss the lead paper as a reading group. All students must read the lead paper and, before each session, must submit a summary with strengths and weaknesses.
Finally, each student will write a seminar paper on the topic assigned to them, for which the two papers serve as the starting point.
The seminar slot takes place on Mondays from 10:15 to 12:00.
CMS: https://cms.cispa.saarland/websecsem_wise2324/
Places: 12
Trustworthy Machine Learning by Adam Dziedzic, Franziska Boenisch, and Mario Fritz
The deployment of machine learning in real-world systems necessitates methods to ensure trustworthy AI. This course explores research at the intersection of machine learning, security, and privacy. This course provides a comprehensive overview of techniques to build robust and trustworthy machine learning models, with a focus on neural networks. We will examine seminal work on defending against adversarial attacks, detecting out-of-distribution inputs, and adapting models to distribution shifts. We will analyze privacy-preserving collaborative learning methods that enable multiple parties to jointly train models without exposing private data or models. To protect intellectual property, we will study approaches for thwarting model stealing attacks and establishing ownership of models. Special attention will be given to watermarking techniques for large language models and defending against data reconstruction attacks on foundation models. Throughout the course, we will discuss outstanding challenges and future research directions to make machine learning more robust, private, and trustworthy.
Every class, we will discuss several papers. The papers for a given class will have a common theme. At the beginning of the quarter, students will be assigned roles which will rotate every week. There are three roles:
1. The Presenter: This person presents a paper and takes the lead in answering the questions posed by The Questioners.
2. The Questioners: This group is responsible for preparing a list of 4–5 discussion questions about the papers to be discussed in class. For a given week, The Questioners must prepare their questions during the preceding week, and send them to the rest of the class by 5pm Friday. This means that The Questioners must read all the papers for their assigned week several days in advance of the actual discussion sessions. We suggest aiming to read the papers by the end of the day on Thursday, to allow at least one day to discuss possible questions.
3. The Observers: During a discussion, this group will take notes on a shared document. These notes are not meant to be a transcription of what is being said in the discussion; they should capture the major take-away points of the discussion, as well as any issues The Observers feel should be discussed in more depth. The Observers should also search for additional resources, or answers to unresolved questions, on the Internet during the discussion itself.
These roles do not preclude anyone in the class from participating in the discussion. A member of The Observers can jump in when a question is posed, and a presenter can pose a new question on the fly.
Requirements: The course presumes a basic understanding of machine learning. This seminar is open to senior Bachelor, Masters, and Doctoral students. Through seminal and recent papers, students will survey the emerging literature across research communities investigating these issues. The class aims to inspire new research directions and applications. Lectures, slides, and research papers comprise the course materials - no textbook is required. By engaging with the latest work in this rapidly evolving field, students will be prepared to advance trustworthy machine learning. Each student will present a paper during the seminar hours in the form of an oral presentation. In addition, each student will read the relevant papers for the other students’ presentations, and hand in a seminar paper at the end of the semester.
Places: 12