|
The visual system does not act as a mere camera. As early as in the retina, the brain starts to interpret the information it receives in order to build the perception of visual reality we experience. My goal is to shed more light on this process. Taking advantage of recent deep learning methods, I work on building retinal models that enable us to gain more insight into processing in the earliest stages of the visual system.
|
|
I am interested in combining bioacoustics and deep learning, with a strong motivation to develop methods that are applicable to real-world biological research. For my master’s thesis, I am working on the automatic segmentation and classification of ring-tailed lemur vocalizations from extended audio recordings with high background noise, leveraging Audio Foundation Models.
|
|
I want to understand visual perception in the brain by leveraging novel machine learning techniques to build predictive models of neural population responses to natural images. Especially, I think it is important to combine a model’s predictive accuracy with interpretability to gain insights into mechanisms of biological computation. I am excited by the idea to implement the biologically inspired building blocks we develop into artificial neural networks.
|
|
Coming from graph learning and learning theory, I am interested in ways to use machine learning techniques to solve actual scientific questions in physics and biology. Furthermore, I like to think about ways in which we can use deep learning to solve or approximate NP hard problems such as graph coloring.
Website
|
|
Biologists conducted social learning experiments with free-ranging red-fronted lemurs. However, analyzing the videos manually proved to be challenging and time-consuming. Consequently, I want to use deep learning techniques to detect the various interactions occurring among the lemurs during the experiment.
|
|
I am excited about developing machine learning methods that help us understand complicated and high-dimensional data sets. Our current goal is to formalise what it means for data points to belong either to clusters or to a continuum, and to develop methods for quantifying the degree of clustering. Another project I am working on is developing a representation learning method for molecular graphs, where the representation is invariant, but an equivalent object can be recovered with high probability.
|
|
My research goal is to understand how visual perception in the brain works and how to apply this knowledge to improve perception in artificial neural networks. I work at the intersection of computational neuroscience and machine learning. I’m especially interested in unsupervised representation learning and to understand how the notion of objects arises in visual perception in biological systems and how we can use this knowledge to improve scene understanding of artificial neural networks.
|
|
I believe the most noble application of computer science is the advancement of science, answering complex and unsolved problems. I have a background in Software Engineering and a love for Physics that goes back to my childhood, and I believe I am currently living my dream working on the marriage of both fields.
|
|
I am working on a modern approach of bee monitoring in agriculture, which is currently being done manually. My goal is to use cameras to record and neural networks to detect and classify bees and other insects to get a better understanding of their population.
|
|
I'm passionate about applied deep learning and its potential to solve real-world challenges. My current focus lies in the domain of bioacoustics, where I'm working on my master thesis project in cooperation with the German Primate Centre. Specifically, I'm studying deep learning techniques to detect, segment and classify vocalizations of ring-tailed lemurs from large-scale audio recordings.
|
|
My research is focused on developing models that can learn the interactions and dynamics of moving objects from videos. For this purpose, I have been working on attention-based (self) supervised, semi-supervised and object-centric learning methods. Such object-interaction learned models hold potential for applications in areas such as multi-object tracking, segmentation of moving objects, and video instance segmentation. In addition to my primary research interests, I am also involved in projects related to active learning, and action and interaction recognition.
|
|
I am working on data sorting techniques for videos of behavioral experiments on lemurs. I am trying to reduce the data labeling time by preselecting video frames from lemur videos where they are identifiable, instead of finding these frames by hand. The goal is to reach similar performance with the automatically selected frames on a subsequent identification task.
|
|
I am working with image and video labeling software to create annotations in monkey videos. These are used for supervised training.
|
|
It seems like computer vision will be integrated into more and more areas in the future. This also applies to the manufacturing industry for the automation of many different processes. That's why I'm writing my master thesis in the field of visual quality control of industrial products in cooperation with the company Maddox AI. In particular, I focus on the knowledge transfer from multiple data sources.
|
|
I'm working on a graph-based neural network approach to learn and predict tree characteristics from graph representations of their skeletons. A similar method has been used to classify brain cells in mice, and trees look the same as far as computers are concerned.
|
|
In cooperation with the Max Planck Institute for Dynamics and Self-Organization I am working on my physics master thesis. The aim of my work is to utilize deep learning approaches to evaluate measurements on physical systems. More precisely, the focus lies on the reconstruction of particle positions in clouds from recorded digital holograms. In the end, a significant reduction in the computational cost might be achieved, making digital holography a more affordable technique.
|
|
Artificial neural networks were originally inspired by biology. I find the reverse way of using this technology to gain a better understanding of the brain, a fascinating idea. Here in the Neural Data Science group, I am working on incorporating the correlation of neurons into a model that predicts neural responses in the visual cortex.
|
|
Computer Vision is a fascinating topic for research, as it is at the same time a tool for other researchers to perform and analyze their experiments. My project deals with tracking droplets within a cloud, in order to better understand the emergence of raindrops. Analyzing videos to estimate those trajectories is highly resource demanding, which is the main challenge I want to tackle. I'm particularly interested in finding efficient ways to track a high number of objects.
|
|
A major argument for self-supervised learning is that it does not require reliance on expensive labeled data. I investigate contrastive representation learning by generating a synthetic dataset with Blender and using a new data augmentation, namely perspective change, to gain deeper insight into the underlying functionalities.
|
|
I study visual processing in the brain by building predictive models of population responses from the macaque and rodent brains to image and video sequences. I leverage on advances in machine learning and computer vision to both improve predictive power and to gain insights into the nonlinear computations of visual neurons. My goal is to be able to use these insights to enhance current computer vision methods.
Website
|
|
At the Neural Data Science group, I am working on models that predict neural responses from different areas of a macaque's visual system simultaneously. My goal is to learn something about the signal flow between those areas. Beyond that, my main interest lies in probabilistic and Bayesian machine learning which I would like to use to make ML more robust. I am fascinated by how these approaches allow us to quantify the uncertainty of our models helping us to make their usage safer.
|
|
Broad interest in data science topics. Currently writing my bachelor thesis on self-supervised learning.
|
|
Humans do not only excel at acquiring novel concepts from a single demonstration but can also readily identify or reproduce them. When shown a new object humans have no problem pointing at similar objects or drawing their outlines. My goal is to bring similar capabilities to computer vision systems.
Website
|
|
I study how we as a society can leverage both physical understanding and data-driven models for tackling complex challenges. At the Neural Data Science group I build robust deep learning models for trustworthy automated building damage assessment after natural disasters.
|
|
During my time at the Neural Data Science Group I focused on extending a convolutional neural network for modelling neural data in mouse primary visual cortex to take into account noise correlations.
|
