University of Münster  [ 🔗 website ]

Decoding brain states with rhythmic and non-rhythmic features of brain signals
Rhythms are a prominent feature of brain activity and they are known to change across brain states. However, traditional approaches typically rely on an assumption that is probably universally violated in real data – namely that underlying signals are stationary sinusoidal waveforms. In addition, the importance of aperiodic activity and waveform shape is becoming increasingly clear. Therefore, it seems advantageous to characterise brain activity along many dimensions. Here, I report first results of a new project where we employ large-scale time-series phenotyping of MEG data. We use a large set of features to characterise source-localised MEG data and identify features that are informative for predicting age, changes in brain state or that follow cortical gradients. In all cases, these non-standard features outperformed traditional spectral measures.

Radboud University, Nijmegen  [ 🔗 website ]

Signal and noise in neural variability
The brain is a dynamic system that is characterized by remarkably variable activity. Increasing evidence from neuroimaging experiments indicates that such variability is a key feature of human computational and behavioral flexibility. However, discriminating between parts of complex time series that are an informative sign of a latent neural process, an instrumental signal within the brain, or uninformative “noise” remains a long-standing challenge in neuroscience. Historically, this has inspired algorithms that (over-)emphasize specific features such as periodic rhythms in EEG/MEG research. In contrast, emerging perspectives envision neuro-computational noise as a signal of interest in its own right, calling for a more comprehensive appreciation of recorded fluctuations. In this talk, I will use research examples to highlight current perspectives with a focus on EEG/MEG research, illustrate principled approaches to practically decompose human brain variability, and point out ongoing theoretical and methodological challenges.

University of Düsseldorf  [ 🔗 website ]

Doing everything your MEG manufacturer advises you not to do – Cleaning MEG/EEG data from stimulation artifacts
Deep Brain Stimulation (DBS) is an invasive treatment option for neurological and psychiatric disorders which can improve the patient’s quality of life to a great extent. Despite its clinically established use for e.g. Parkinson’s disease, its underlying mechanism is still unclear. In order to better understand the mechanisms of DBS, its influence on whole brain activity needs to be

investigated. However, DBS produces large artifacts due to the applied current and the electrode itself in MEG and EEG data. To complicate things, these artifacts seem to differ between MEG and DBS systems. In this talk, I will present approaches to mitigate these artifacts to achieve insights into the mechanisms of deep brain stimulation.

Maastricht University [ 🔗 website ]

Decoding Speech from Intracranial EEG
Speech is our most natural way of communication and the loss of the ability to speak is therefore devastating to patients. A speech neuroprostheses that directly reconstructs speech processes from neural activity could provide a new means of communications to these severely affected patients.

In this presentation, I will present some approaches to reconstruct different representations of speech from intracranial recordings and highlight how they can be used to build a speech neuroprosthesis. The decoding of speech processes is particularly challenging, as not only the neural, but also the target signal has complex, nonlinear dynamics. I will stress the use of interpretable machine learning models for this task to ensure that meaningful activity is decoded and scientific insights might be generated as a side product.

University of Tübingen [ 🔗 website ]

Encoding models and how to use them – deciphering the cortical representation of continuous speech.
As speech evolves in time, auditory cortices track and respond to salient acoustic and linguistic features in the speech signal, a phenomenon known as speech tracking. Speech tracking is a prerequisite for successful mapping between the acoustic signal and linguistic categories. In my talk I will first discuss recent work using encoding models to characterize local cortical encoding of the spectro-temporal, phonemic and syllabic content of naturalistic speech throughout the entire human speech cortex. The results of this work show that speech representation in primary areas is dominated by spectro-temporal representations, whereas representations in associative speech cortex are geared towards the phonemic and syllabic structure of speech. I will then present a recent study where we used encoding models to dissociate between tracking of syllabic content and oscillatory entrainment to it, with no evidence of the latter.

University of Leipzig  [ 🔗 website ]
Max Planck Institute for Human Cognitive and Brain Sciences Leipzig, Germany 

Inferring causality from non-invasive brain stimulation in Cognitive Neuroscience
Across the last decades, there is increasing interest in the application of non-invasive brain stimulation (NIBS) to probe causality in cognitive neuroscience. NIBS techniques provide powerful tools to experimentally manipulate brain function and study the consequences at the behavioral and neural network level, thereby complementing correlative neuroimaging.

Many NIBS studies follow the simple logic of asking “is region X causally involved in task Y?” While this approach has made significant contributions to cognitive neuroscience, response modulation by NIBS appears to be more complex. I will provide examples for common approaches and pitfalls in the study of cognition with different NIBS protocols. Example studies suggest that many influencing factors which are usually ignored should be considered when explaining NIBS effects. Considering the interaction of internal and external factors will increase the current understanding of NIBS-induced modulation of brain function. Multimethod approaches provide insight into distributed network effects and the modulation beyond single brain areas.

University of Tübingen  [ 🔗 website ]

Inferring circuit mechanisms from neural power spectra: parameterization, simulation, and probabilistic machine learning
Brain signals are measurements from a messy bio-electro-chemical dynamical system and are often useful as correlates of behavior and cognition. But to further take advantage of them, we can leverage other experimental data and computational tools to decode physiological mechanisms underlying their fluctuations. This requires 3 ingredients—data analysis methods (parameterization), mechanistic models (simulation), and a way to link models to data (inference)—combined in a process known as inverse modeling or data assimilation.

Here, we will first cover the basics of Fourier(-based) analysis. Then, we will discuss spectral parameterization, i.e., FOOOF, as a data analysis tool to infer circuit properties from periodic and aperiodic components of neural power spectra, such as cellular and network timescales and excitation-inhibition balance. Lastly, following the same philosophy, we will discuss mechanistic circuit models with many more parameters, and explore how to use simulation-based inference (sbi) and probabilistic deep learning to perform model inference using real experimental data. Hands-on coding sessions with the associated python toolboxes is a central focus to facilitate analysis of your own data using spectral parameterization (with guidance on hyperparameter choices), as well as fitting your own models with sbi.

University of Freiburg  [ 🔗 website ]

Deep Learning for Interpretable EEG Decoding with Braindecode
Deep neural networks are nowadays widely used for EEG decoding, reaching good decoding performance on a broad variety of decoding tasks. In this workshop, we will first show how to use the open-source EEG deep-learning toolbox Braindecode to train well-performing deep networks on EEG data. After such training, one challenge of gaining scientific knowledge from EEG-trained deep networks is to understand their internal computations: What EEG features do they learn and how do they relate to the decoding targets? Therefore, in this workshop, we will also use interpretability methods for peeking into the learned functions of the trained networks to see what insights we can gain about the EEG signal from them.

Netherlands Institute for Neuroscience  [ 🔗 website ]

Extracting interpretable patterns from multivariate time series data using GED
Electrophysiological signals recorded at any spatial scale represent a mix of activity from different neural sources. This tutorial will focus on a supervised, as opposed to blind, source separation using generalized eigenvalue decomposition (GED). GED method extracts statistical (rather than anatomical) sources, which are expressed as linear weighted combinations of electrode time series (spatial filters). Supervision step involves data selection for implementing GED, as the method relies on the covariance matrices derived from different features of the data (e.g. narrow-band vs. broad-band activity; pre-stimulus vs. post-stimulus period; two experimental conditions). To build an intuition regarding the strengths (increased SNR, dimensionality reduction) and limitations of the method, we will apply it to several empirical EEG datasets (visual and tactile steady-state evoked potentials, cognitive control task). We will focus on practical aspects: The effects of noise, the amount of data used, covariance matrices derived from single vs. concatenated trials, matrix regularization, overfitting, and interpretation of GED results.

Jülich Research Center  [ 🔗 website ]

Manage your Data with DataLad: From local version control to data publication
Many open source software tools assist with the complex task of data management. This workshop introduces one of them, the data management and data publication tool DataLad (https://www.datalad.org). If you have always wanted to get going with version control, are curious to find out how to use DataLad, or want to see simple workflows for data management routines, this workshop is for you. You will learn about core concepts of research data management and how to implement them – from local version control for all parts of your research to data publication. With a focus on technical and conceptual aspects alike, the workshop aims to equip you with skills you could transfer to real-world data, with a mix of theoretical input and interactive hands-on applications.

The workshop will provide a cloud-computing environment for interactive participation, but participants should bring their own computers in order to access it.

Goethe University Frankfurt [ 🔗 website ]

Net2Brain: Unraveling cognitive functionality as captured by EEG using pre-trained artificial neural networks
This workshop will introduce Net2Brain, an innovative toolbox that enables researchers to harness the power of deep learning for neuroimaging data analysis. Unlike conventional toolboxes that offer limited functionalities focusing predominantly on supervised image classification models, Net2Brain stands out with its comprehensive capability to extract activations from a remarkable variety of over 1000 ANNs trained for diverse vision-centric tasks across image and video datasets. Moreover, Net2Brain also includes multimodal, and language models. The toolbox also has a built in in-depth comparative analysis pipeline that employs advanced techniques such as representational similarity analysis (RSA), weighted RSA and encoding models. Facilitating in-depth brain-model comparison, both within specified regions of interest and across entire temporal and / or spatial spaces using searchlight searches and temporal neural pattern dynamics analysis. The user-friendly platform is open-source, complemented with brain data and end-to-end analysis examples for immediate exploration. It also ensures streamlined incorporation of datasets and new models for evaluation. During the workshop, we will elucidate the capabilities of Net2Brain by presenting its application in recent studies. One of such studies utilizes EEG to explore how the human brain processes visual information to facilitate navigation planning in indoor scenes. The workshop will include a hands-on section during which you will be provided with interactive notebooks that run in your browser. Join us as we unravel the potential of Net2Brain in advancing cognitive computational neuroscience.

University of Birmingham  [ 🔗 website ]

A different perspective on oscillations with Empirical Mode Decomposition
Neuronal oscillations (and any other kind…) can be split into separate components using the data-drive sifting algorithm to perform an Empirical Mode Decomposition. The resulting modes contain mono-component oscillations that retain much of the non-linear and dynamic properties of the original signal. We can then represent the spectral content of these modes using instantaneous frequency as an alternative to the Fourier transform. This alternative framework avoids some of the common pitfalls of Fourier analyses, such as time-frequency trade-off, and provides a window onto challenging features such as rapid dynamics and waveform shape. Like all analyses the EMD algorithms are not perfect, but I argue that they provide a powerful complement to traditional approaches that is better optimised for approaching non-linear and non-stationary signals. I will illustrate these methods with an interactive demo using resting state EEG recordings from the LEMON dataset, MNE-Python and the EMD toolbox (http://emd.rtfd.io/).