Open to all early career researchers. Get to know what your fellow Ph.D. students are working on!
Time: 30.04.2025 at 4:00 pm
Place: Room 1.22 in house 9.
Details are announced via mailing list "SIAM-chapter-list", or ask Giuseppe or Martin.
Speaker: Talha Ersoy
Title: Geometry of Learning – L2 Phase Transitions in Deep and Shallow Neural Networks
Abstract:
When neural networks (NNs) are subject to L2 regularization, increasing the regularization strength beyond a certain threshold pushes the model into an under-parameterization regime. This transition manifests as a first-order phase transition in single-hidden-layer NNs and a second-order phase transition in NNs with two or more hidden layers. We investigate a frame- work for such transitions by integrating the Ricci curvature of the loss land- scape with regularizer-driven deep learning. First, we show that a curvature change-point separates the model-accuracy regimes in the onset of learning and that it is identical to the critical point of the phase transition driven by regularization. Second, we show that for more complex data sets additional phase transitions exist between model accuracies, and that they are again identical to curvature change points in the error landscape.
Open to all early career researchers. Get to know what your fellow Ph.D. students are working on!
Time: 20.03.2025 at 4:00 pm
Place: Room 2.22 in house 9.
Details are announced via mailing list "SIAM-chapter-list", or ask Giuseppe or Martin.
Speaker: Maria Mannone
Title: Brain-Network Alteration in Disease: A Physical Journey
Abstract:
The damage caused by a neurological disease to the brain connectome can be formalized through the action of a suitable operator, the “Krankheit-Operator” (in short, K-operator). K acts on the connectivity matrix of the time-series correlations of a healthy brain and yields a diseased brain, or, acting on an already-diseased brain, gives the time evolution of the disease. Thus, the idea of an operator acting on an observable, derived from physics, can help model damage occurring in some neuropsychiatric and neurodegenerative diseases.
The presented specific forms of the K-operator concern Alzheimer-Perusini’s disease, Parkinson’s disease, epilepsy, schizophrenia, and long Covid. The visual inspection of the most prominent values of the K-operator can help identify those pairs of brain regions where to focus attention, deepening the analysis with recurrence plots. While the examples focus on functional connectivity, some research developments can also include structural connectivity, which is investigated with tractography.
Open to all early career researchers. Get to know what your fellow Ph.D. students are working on!
Time: 29.01.2025 at 4:00 pm
Place: Room 1.22 in house 9.
Details are announced via mailing list "SIAM-chapter-list", or ask Franziska or Martin.
Speaker: Chloé Rouyer
Title: Introduction to Reinforcement Learning
Abstract:
Reinforcement Learning is the science of sequential decision-making and can be used to dynamically learn complex tasks for example in robotics or with self-driving cars. Instead of hard-coding the learner's behaviours, which can be too complex to formulate, the learner is placed in an environment, has to interact with it and is given rewards when performing good actions. These rewards are then used to shape future behaviours. The key challenge of reinforcement learning is to choose a behaviour-learning strategy that adapts to the uncertainty of the environment and the rewards.
In this talk, we consider a simple reinforcement learning framework and discuss strategies to optimize decision-making with noisy reward sequences.
Dear fellow Studis,
to honor and further revitalise the valuable tradition of bi-annually student chapter conferences, MESIGA, the hosting SIAM and GAMM student chapters at Uni Potsdam and the Berlin Universities are happy to invite you to the next iteration of this wonderful event from 11.03. to 13.03.2025. After an excellent event at Uni Trier in 2023, we want to match their spirit of hospitality for another great event. We hereby invite all Uni Potsdam students with a keen interest in the very broad topic to join us.
All required information can be found here: Information and Registration
Open to all early career researchers. Get to know what your fellow Ph.D. students are working on!
Time: 27.11.2024 at 4:00 pm
Place: Room 1.22 in house 9.
We are excited to announce that for our last SIAM student chapter activity, we will be having two closely related talks on the highly active research field: data-driven model order reduction!
Afterwards we will gather for a small social event, which means cookies, games and hanging out!
Details are announced via mailing list "SIAM-chapter-list", or ask Franziska or Martin.
Speaker: Yevgeniya Filanova
Title: Structure-preserving inference of mechanical systems
Abstract:
The construction of mechanical models based on experimental or simulation data is becoming increasingly important in engineering practice. On the one hand, the mechanical models identified from experimental measurements can be used for analysis and simulation. On the other hand, the models used in modern simulation software often need to be further reduced and reconstructed from the simulation results because the original system operators are difficult or impossible to access. Moreover, the identified mechanical system should be able to represent the system behavior for different load cases and time schemes, which means that the original system properties and structural characteristics should be preserved. The recent non-intrusive methods, such as operator inference, often either do not satisfy the latter condition or do not provide a complete system representation due to the specific second-order ODE structure of the governing system of equations. In our work, we propose a methodology for identifying and reducing mechanical systems from data by solving an optimization problem using the capabilities of machine learning techniques. We are able to identify all system operators and preserve their original mathematical properties by including a reasonable parameterization of the system operators in the optimization problem. For linear mechanical systems, this ensures stability and interpretability of the resulting surrogate model. Numerical experiments are performed using simulation data from finite element software.
Speaker: Martin Nicolaus
Title: Learning stochastic reduced order models from data: a nonintrusive approach
Abstract:
A nonintrusive model order reduction method for bilinear stochastic differential equations with additive Gaussian noise is proposed. A reduced order model (ROM) is designed in order to approximate the statistical properties of high-dimensional systems. The drift and diffusion coefficients of the ROM are inferred from state observations by solving appropriate least-squares
problems. The closeness of the ROM obtained by the presented approach to the intrusive ROM obtained by the proper orthogonal decomposition (POD) method is investigated. Two generalisations of the snapshot-based dominant subspace construction to the stochastic case are presented. Numerical experiments are provided to compare the developed approach to POD.
Open to all early career researchers. Get to know what your fellow Ph.D. students are working on!
Time: 30.10.2024 at 4:00 pm
Place: Room 1.22 in house 9.
Details are announced via mailing list "SIAM-chapter-list", or ask Franziska or Martin.
Speaker: Jan Szalankiewicz
Title: Filtering SPDEs with Spatio-Temporal Point Process Observations
Abstract:
We develop the mathematical framework for filtering problems arising from biophysical applications where data is collected from confocal laser scanning microscopy recordings of the space-time evolution of intracellular wave dynamics of biophysical quantities. In these applications, signals are described by stochastic partial differential equations (SPDEs) and observations can be modelled as functionals of marked point processes whose intensities depend on the underlying signal. We derive both the unnormalized and normalized filtering equations for these systems, demonstrate the asymptotic consistency and approximations of finite dimensional observation schemes respectively partial observations. Our theoretical results are validated through extensive simulations using synthetic and real data. These findings contribute to a deeper understanding of filtering with point process observations
and provide a robust framework for future research in this area.
Open to all early career researchers. Get to know what your fellow Ph.D. students are working on!
Time: 31.07.2024 at 4:00 pm
Place: Room 2.22 in house 9.
Details are announced via mailing list "SIAM-chapter-list", or ask Franziska or Martin.
Speaker: Joanna Zou (MIT)
Title: Path-space error bounds for the statistical learning of stochastic differential equations
Abstract:
In physical systems whose equation of motion takes the form of a stochastic differential equation (SDE), the drift function of the SDE often involves expensive calculations which render numerical simulation to be cost-prohibitive. The goal in surrogate modeling for SDEs is to learn an approximate drift function such that the surrogate process replicates observable quantities of interest of the system, such as statistics over measures of the stochastic process, at a fraction of the cost. This work examines the use of variational bounds to define loss functions in the statistical learning of SDE models which provide error guarantees in transient and long-time statistics. In particular, we show that error in path-based observables are upper bounded by a term which depends on the variance of the observable quantity and pathwise Kullback-Leibler divergence. We show this bound is computable using Girsanov’s theorem and relate it to information metrics such as the relative entropy rate (RER) and pathwise Fisher information. We discuss directions for using the bound to derive an appropriate weighted norm in the training loss function for learning the surrogate SDE.
Please note that the seminar takes place on the 19.06. instead of 26.06.!
Open to all early career researchers. Get to know what your fellow Ph.D. students are working on!
Time: 19.06.2024 at 4:00 pm
Place: Room 1.22 in house 9.
Details are announced via mailing list "SIAM-chapter-list", or ask Franziska or Martin.
Speaker: Luci Fumagalli
Topic: Sequential Monte Carlo squared to simultaneous localize and map autonomous vehicles
Abstract:
The Simultaneous Localization And Mapping (SLAM) problem arises in autonomous driving when a vehicle has no direct information about its surroundings. In this situation, the vehicle has to autonomously create a representation of its environment (Mapping) and locate itself (Localization) in order to be able to navigate, which we refer to as the SLAM problem.
In this talk, we rephrase the SLAM problem as a filtering problem and explore a probabilistic algorithmic solution, the Sequential Monte Carlo Squared (SMC2) filter. The SMC2 filter uses observations from the vehicle's sensors to find the solution to the SLAM problem. We will also sketch the mathematical justification for why the SMC2 filter works.
Please note that the seminar takes place on the 22.05. instead of 29.05.!
Open to all early career researchers. Get to know what your fellow Ph.D students are working on!
Time: 22.05.2024 at 4:00 pm
Place: Room 1.22 in house 9
Details are announced via mailing list "SIAM-chapter-list", or ask Franziska or Martin.
Speaker: Erik Kalz
Topic: The subtle Nature of Interactions in Odd-Diffusive Systems
Abstract:
Odd systems exhibit unusual and novel behavior with no counterpart in normal systems. They are characterized by transverse responses to perturbations. Classical examples would be the Hall effect in conductors and the Magnus effect in fluids. Therefore, in recent years, the physics community has developed an interest in odd physics which arises in the continuous descriptions of matter, such as odd viscosity, odd elasticity, or odd diffusion. Specifically odd-diffusive systems, which, for example, are realized in Brownian systems under Lorentz force, active chiral particles, or diffusing skyrmions are of special interest to us. Particle interactions in these systems result in novel diffusive behavior, the most surprising of which is that collisions enhance the self-diffusion rather than decreasing it. The origin of such a counterintuitive behavior can be traced back to the unusual, non-Hermitian dynamics in odd systems, which are encoded in correlation tensors with non-zero off-diagonal elements, even though the systems are isotropic.
Open to all early career researchers. Get to know what your fellow Ph.D students are working on!
Time: 28.02.2024 at 4:00 pm
Place: 2.28.0.102 (Attention: Change of location!)
Details are announced via mailing list "SIAM-chapter-list", or ask Franziska or Hans.
ATTENTION: This time we have two great speakers working on related fields!
Seminar Title: The Mysterious Motility of Microorganisms
First Presentation
Speaker: Agniva Datta
Topic: The mysterious motion of swimming bacteria: a bridge connecting biology and non-equilibrium physics
Abstract:
From the mesmerizing patterns formed by flocking birds to the remarkable ability of lizards to regenerate their tails, the intricate phenomena observed in living systems at different length scales beckon physicists to unravel their mysteries. These captivating puzzles often find their answers at the interface of nonlinear dynamics and non-equilibrium statistical mechanics. Not only do these solutions shed light on the underlying biology, but they also unveil previously uncharted territories within the black-box of physics far from equilibrium.
In this presentation, I will delve into a case study from my doctoral research: the motion of swimming bacteria. Our study incorporates a combination of experimental analysis utilizing the soil bacterium
Pseudomonas putida and active particle modeling. Specifically, we investigate how the disordered environment (agar) guides the migration patterns of these bacteria, resulting in remarkable motility characteristics like ergodicity breaking. In contrast to E. coli, our research reveals the transient sub-diffusion of bacteria in agar, primarily attributed to intermittent trapping. These findings underscore a dynamic run-and-trap mechanism, with trap times following a power-law distribution. I will discuss the implications of these findings and establish an intricate correlation between the micro-scale navigation of bacteria and their largescale movement in diverse and heterogeneous environments.
Second Presentation
Speaker: Jan Albrecht
Topic: Bayesian Parameter Inference for Biological Tracking Data
Abstract:
In order to understand and predict the motion patterns of microorganisms, robust methods to infer motility models from time discrete experimental data are required. Due to the internal complexity of the organisms and the ensuing quasi-random motion, stochastic models like SDEs are well suited to describe their movements.
Bayesian statistical methods provide a way to efficiently extract information from the trajectory data and provide model parameter estimates together with a measure of uncertainty.
We showcase that Bayesian methods are especially well suited when the models contain additional layers of stochasticity, for example population heterogeneity or temporal dependence of parameters. Furthermore, we demonstrate how challenges that arise when multidimensional dynamics is only partially observed, e.g. second order dynamics, can be addressed.
Fellow SIAM Student Chapter Appreciators and Frequent Guests,
To better get to know our neighbour SIAM chapters in Berlin, discuss shared experiences about studying Mathematics and exchange cool insights about research and organisation in other SFBs, we are happy to invite you to our first shared event:
The Metropolregion Potsdam-Berlin Siam Student Chapter Pub Crawl @ Friedrichshagen (approx. Boxhagener Platz).
The MPBSSCPC@F-hain, for short, will take place on 09.02.2024 starting at Train Station Berlin Ostkreuz (or Ost-X as it is known to regionals) at 6pm. We are planning on having a great time with reasonable cheap drinks, delicious food from all over the area and good company.
But there is more! With some financial support from the SIAM-Chapter treasury, we have some budget for limited free drinks!
To end with our University of Potsdam SIAM Student Chapter President's famous goodbye:
You all look very good today! Have a nice day, bye!
Your SIAM chapter officers,
Giuseppe, Josie, Hans and Franzi
Get to know what your fellow Ph.D students are working on!
Time: 25.10.2023 at 4:00 pm
Place: 2.09.2.22. Details are announced via mailing list "SIAM-chapter-list", or ask Franziska or Hans.
Speaker: Maximilian Kloppe
Topic: A Phase-Field Model of Elastic Surfaces in Flow
Abstract:
The small thickness of membranes, shells and capsules enables their efficient approximation as hyper-surfaces. Phase-field modeling provides a versatile tool to capture the motion of such elastic hyper-surfaces in fluid flow under bending and surface tension forces. However, the in-plane stretching of the surface has been widely neglected or approximated by inextensibility of the material.
In this talk, we develop a novel phase-field model for elastic hyper-surfaces in Navier–Stokes fluids, which includes bending, tension, and in-plane stretching. The model is based on a coupling of a phase-field model for two-phase flow to a fully Eulerian description of the surface deformation tensor. We apply the method to microfluidic experiments on lipid vesicles and illustrate how the results can be used to better understand the mechanical properties of cell membranes.
Get to know what your fellow Ph.D students are working on!
Time: 27.09.2023 at 4:00 pm
Place: 2.09.2.22. Details are announced via mailing list "SIAM-chapter-list", or ask Franziska or Hans.
Speaker: Paula Pirker Diaz
Topic: Self-regulation of a network of Kuramoto oscillators (Complexity Science and Complex Systems Theory)
Abstract: Persistent global synchronization of a neuronal network is considered a pathological, undesired state, and it is often caused by the loss of neurons or their regulatory agents, e.g. glial cells.
Considering a brain-like network characterized by a modular organization combined with a dynamic description of the nodes as Kuramoto oscillators, we have applied a self-regulation mechanism to keep local synchronization while avoiding global synchronization at the same time. In this talk, I am going to explain our model and the results obtained, which show that the desired state is achievable despite the simplicity of the approximation.
Get to know what your fellow Ph.D students are working on!
Time: 26.07.2023 at 4:00 pm
Place: 2.09.0.17. Details are announced via mailing list "SIAM-chapter-list", or ask Franziska or Hans.
Speaker: Johannes Tillil
Topic: Random Forests with Grouped Variables in Application to Disease Module Detection
Abstract: Random forests use single variables for splitting nodes in decision trees. By using multiple important correlated variables simultaneously for node splitting, one might increase the predictive power of random forests. This can be done with linear combinations of specific variable groups, e.g. modules in networks. We performed a simulation study evaluating the approach for detecting modules in gene expression networks that are causal for some phenotype and found slight improvements over standard random forests.
Get to know what your fellow Ph.D students are working on!
Time: 31.05.23 at 4:00 pm
Place: 2.09.2.22. Details are announced via mailing list "SIAM-chapter-list", or ask Franziska or Hans.
Speaker: Vira Raichenko
Topic: Applications of Persistent Homology to silkworm cocoons
Abstract: A silkworm weaves the incredibly interesting and intricate structure of the cocoon which can be a great inspiration for bio-materials. Therefore, we are interested in developing a framework for understanding the cocoon fiber network through geometrical and topological methods and developing it further for fibrous and entangled materials. In this talk, I am going to give an overview of methods from topological data analysis that can be used for applications to 3D image data, and, in particular, to the scans of silk-worm cocoons.
We invite all Bachelor, Master and PhD Students to join us!
Get to know what your fellow Ph.D students are working on!
Time: 26.04.23 at 4:00 pm
Place: 2.09.1.22. Details are announced via mailing list "SIAM-chapter-list", or ask Franziska or Hans.
Speaker: Maximilian Graf
Topic: An Introduction to Statistical Estimation Theory
Abstract: The aim of this talk is the introduction of basic concepts of estimation theory illustrated by some simple examples. We focus on the notion of risk bounds and minimax risk. In this context, the estimation problem of expert ranking will be shortly explained.
We invite all Bachelor, Master and PhD Students to join us!
Dear Members of the SIAM Chapter at Uni Potsdam,
on March 29th, we will elect a new set of SIAM-officers (President, Vice-President, Treasurer, Secretary). If you are intereset in joining the organizational side of the chapter or would like some more information on what duties each role entails, please write an email to one of the officers and we will answer your questions. There will also be an introduction of the SIAM chapter and roles on the day of the election.
Please note that only members of the chapter will be able to vote - join us by subscribing to our mailing list!
Get to know what your fellow Ph.D students are working on!
Time: 22.02.23 at 4:00 pm
Place: house 9, room 2.22. Details are announced via mailing list "SIAM-chapter-list", or ask Franziska or Florian.
Speaker: Matthias Himmelmann
Topic: Framework Materials and Cylinder Packings
Abstract: In a geometric context, framework materials are structures composed of many rigid bars (or edges), pinned together at vertices where the bars can freely rotate relative to each other. These structures can be rigid or deformable, depending on the way in which the bars are joined together. Possible deformation pathways of a framework structure can be described in an algebraic context, with each rigid rod represented by polynomial constraints. We use this theoretical concept to robustly model (curvilinear) cylinder packings and their dynamical properties by considering Riemannian Optimization and the mathematical theory of filaments in tight contact.
We invite all Bachelor, Master and PhD Students to join us!
Get to know what your fellow Ph.D students are working on!
Time: 25.01.23 at 4:30 pm
Place: house 9, room 2.22. Details are announced via mailing list "SIAM-chapter-list", or ask Franziska or Florian.
Speaker: Maia Tienstra
Topic: Frequentist Ensemble Kalman Filtering
Abstract: In this talk I will briefly discuss statistical inverse problems, and how we can solve them using the Ensemble Kalman filter. I will then explain how in this setting, we implement early stopping in the statistical sense, and give an introduction into the two toy models we use to develop the theory. I will hopefully end with an overview of the numerical simulations that empirically lead us into the theory.
We are happy to invite you to our next SIAM social event next Wednesday, 30th of November at 4pm. We are joining forces with the SFB1294 - Data Assimilation to make the best Christmas cookies this SIAM chapter has ever seen! We meet in the Lesecafé, ground floor of house 14.
We will provide Cookie dough and utensils as well as some decorations and drinks. You are welcome to also bring some cookie cutters, other utensils for decorating, sprinkles. More drinks can be purchased in the Lesecafé or you can bring your own.
A plan of Golm campus is provided here: https://www.uni-potsdam.de/en/adressen/overview
We are happy to see you on Wednesday!
Get to know what your fellow Ph.D students are working on!
Time: 26.10.22 at 4pm
Place: house 9, room 2.22. Details are announced via mailing list "SIAM-chapter-list", or ask Franziska or Florian.
Speaker: Philipp Bartmann
Topic: Higher Order Laplacians on Simplicial Complexes
Abstract: As higher dimensional analogues of graphs, simplicial complexes appear in many areas of mathematics such as topology, combinatorics, data analysis and number theory. However, it is still somewhat a mystery what information about the geometry is encoded in the spectra of its Laplace Operators such as the Hodge Laplacian. In this talk we will define Laplacians for infinite weighted simplicial complexes, discuss basic properties like boundedness and essential self-adjointness and draw parallels to magnetic Schrödinger operators on graphs.
We are happy to invite you to our next SIAM social event next Wednesday, 27th of June at 4pm. Like last time, we will meet at the ping-pong tables behind house 11. In case of rain, we will meet in the coffee room in house 9 (room 2.09.2.10).
A plan of Golm campus is provided here: https://www.uni-potsdam.de/en/adressen/overview
We hope to see you at the seminar!
Can you summarize the content of your research in 3 minutes? Can you do it in an entertaining way? Then our SIAM Slam might be for you!
As a joint initiative of our SIAM Chapter and the SIAM Chapters of Aachen, Heidelberg and Magdeburg, we would like to invite you to SIAM Slam. There, different participants will present their theses or other scientific findings in a very short amount of time. The audience will choose the most entertaining and informative presentation.
The event will take place via zoom on the 10th of June at 18:00. Feel free to ask Florian for the login data.
Each participant will have exactly 3 minutes for their presentation. A slideshow is allowed, however, we advise reducing the number of slides. There will be a prize for the best presentation.
If you just want to be a spectator, you are welcome to join as well. Then, no registration is needed. In the second part of the event, we will run a short quiz based on the presentations. Here, the winner will also get a small prize. We plan to close the event with an opportunity to together play games online.
If you always wanted to try your hand at being a science communicator, this is your time. Who knows, maybe you are the next Richard Feynman or Stephen Hawking? :)
