Teaching
Our teaching covers various aspects of the neurosciences and is part of the MSc Neuroscience program at the University of Freiburg and of the curriculum of the Bachelor and Master of Science Biology offered by the Faculty of Biology.
Search for current lectures in HISinOne.
Current Master Thesis projects, Bachelor Thesis projects and Research projects
Learning hippocampal place code via landmark features (Master thesis project / Bachelor thesis project / research project)
The firing rates and spike timings of hippocampal place cells have been postulated to subserve spatial memories. How exactly locomotive activity shapes both the rate and time code is still unclear. We hypothesize that place fields can be learnt from the recognition of salient landmark features in a 2D environment. A feed-forward network is proposed to test the theory. In detail, the input layer of neurons receives spatial currents based on the field of view of the animal, and thus, are positionally and directionally tuned. Connectivity to the output layer could be learned via teaching signals upon recognition of landmarks. The resultant place fields in the output layer would be expected to align with their properties in experimental findings such as directional selectivity, asymmetric firing maps and heading-independent theta sequences in 2D. In this project, python programming will be used to simulate the dynamics of biological neural networks and analyze the simulation results, which will then be further compared to experimental findings.
Contact: hoi.yiu@biologie.uni-freiburg.de
Supervised autoencoder networks for classification (Practical student project)
Recent technological advancements, have increased the use of supervised classification algorithms, such as deep neural networks (DNNs), as decoders for behavioural tasks in neuroscience. Even thought, such algorithms exhibit low error rates, the use of cross-validation is required for reliable results, and to that end the need for large datasets (Big Data) is a recommended. Nonetheless, often the availability of Big Data in neuroscience is scarce. However, inherently unsupervised non-linear networks, namely autoencoders, have shown promising results when applied for predictive coding. Here a supervised version of such networks will be employed for classification purposes in an attempt to avoid the use of cross-validation schemes where one needs to separate data into train and test sets.
The MNIST database of handwritten digits will be used in this practical, in conjunction with state-of-the-art supervised autoencoders. The MNIST database is well studied for more than 20 years, with a variety of different classifier architectures, that provide model accuracies with minor deviations from the ground truth. Here the effect of network architecture, to the correct classification rates (CCRs) will be investigated. Specific parameters of interest are the depth of the network, the activation functions between layers as well as the size of the latent space. The use of permutation tests will provide valuable insight on the classification results and the ability to identify the ground truth within the data reliably. A comparison between the use of binary and multilabe classifiers will also be investigated. Additionally, the effect of size, dimensionality and noise will be explored with surrogate datasets resembling neuronal activity. Basic understanding of classification algorithms built in Python and interest in data analysis is advantageous.
Contact: markos.athanasiadis@biologie.uni-freiburg.de
Comparing Burst Sequences in Artificial Neural Network with Real Data from Mouse PFC (Master thesis project / Bachelor thesis project / research project)
Aim: Detecting burst sequences in artificial neural network and compare it with real data recorded from mouse prefrontal cortex (PFC).
The objective of this project is to verify the findings obtained through burst sequence analysis of calcium traces recorded from the prefrontal cortex (PFC) by comparing them with sequences generated by an artificial neural network (ANN). We conducted calcium trace recordings from the mouse PFC, capturing data before, during, and after the learning of a task. Our analysis using burst sequence analysis revealed significant alterations in sequence rates and patterns across different behavioral conditions throughout the learning process.
To further validate our approach, we aim to evaluate our methodology on data generated by an ANN. Consequently, the primary goal of this project is to develop a model, preferably a recurrent neural network, capable of emulating the neural activity in the PFC. Subsequently, we will implement the developed pipeline for sequence analysis to detect and quantify the pattern of bursts generated by the ANN model. Ultimately, the outcomes of this research will enhance our comprehension of the neural mechanisms underlying sequence generation in the PFC.
Contact: hamed.shabani@bcf.uni-freiburg.de
Closed-loop components of hippocampal place field activity
(Bachelor Thesis project / Master Thesis project / Research project)
We will test recent state-of-the-art dimensional reduction method on the product space of hippocampal firing rates and behavioral tracking data from freely behaving rats. This analysis approach will elucidate the behavioral correlates of hippocampal network activity beyond spatial tuning of single cells. Students are expected to have basic programming skills in python and an interest in modern data science approaches.