BrainComp2022 - Computational Challenges of ConnectivityIn-Person Event

19 Sept 2022, 09:55 → 23 Sept 2022, 12:00 Europe/Berlin

Cetraro, Italy

Together Neuroscience and Computing are driving forces for research and innovation. They enable new insights into the brain‘s complexity as well as biological information processing and lay ground for progress in Future Computing. Making use of this collaborative effort by bringing together relevant key players in the field of Neuroscience and Future Computing, the workshop on Brain-Inspired Computing (BrainComp) aims to shed a light on the digital transformation of Neuroscience by High Performance Computing (HPC).
The next BrainComp workshop will take place from 19-23 September 2022 in Cetraro (Italy) and is by invitation only. Special focus of this year's edition  are the Computational Challenges of Connectivity.

Sponsors and Co-Organizers of the event

Programme Committee

Katrin Amunts, Research Center Juelich, Germany
Lucio Grandinetti, University of Calabria, Italy
Thomas Lippert, Research Center Juelich, Germany
 

 

Monday 19 September

Welcome | Katrin Amunts and Lucio Grandinetti

Connectivity: The network perspective

1 Generalizability of connectome-based predictive models

The development of connectome-based predictive models of behavioral phenotype has more recently opened new perspectives for understanding brain-behavior relationships in basic neuroscience, but also for precision medicine. However, the insight provided by the machine learning models and the further deployment of these approaches are conditioned by their generalizability. In our recent work, we tackled this crucial question across several popular datasets in the field.

Speaker: Dr Sarah Genon (Research Center Juelich)

Presentation_Sarah Genon_BrainComp 2022.pdf

2 Influence of local neuronal and circuit features on the functional mapping of the cortical network

The brain forms a highly interconnected network or connectome that can nowadays be mapped in ever-increasing detail thanks to the improvement of the resolution of imaging techniques and reconstruction methods. However, the structural mapping does not always correspond to the functional mapping, estimated by means of functional or effective connectivity. Functional connectivity is dynamic and varies with the functional state of the network. Indeed, functional connectivity varies with brain states, thus functional network is different in sleep states, wakefulness, anesthesia, disorders of consciousness and so on. Thus, depending on the brain state, a network can change from being dominated by the balance between integration and segregation (wakefulness), to be highly integrated and fully synchronized (slow wave sleep or deep anesthesia). The state of the network and the functional consequences, such as the causal interactions across areas, can be investigated in resting states, task-evoked or following stimulation or perturbation. Our results and those of others suggest a deep change such that in states of unconsciousness the causal interactions across areas are lost with respect to wakefulness. But, what are the mechanisms underlying these flexible dynamics? Are they synaptic, cellular, subcellular changes? We will show the impact of individual ionic currents and synaptic receptors on functional connectivity and network complexity, bridging the ionic channels with the network’s spatiotemporal network dynamics.

Speaker: Prof. Mavi Sanchez-Vives (Institut d'Investigacions Biomèdiques August Pi i Sunyer (IDIBAPS))

11:15 Coffee Break

Connectivity: The network perspective

3 The Thermodynamics of Mind

Finding precise signatures of different brain states is a central, unsolved question in neuroscience. The difference in brain state can be described as differences in the detailed causal interactions found in the underlying intrinsic brain dynamics. We use a thermodynamics framework to quantify the breaking of the detailed balance captured by the level of asymmetry in temporal processing, i.e. the arrow of time. We also formulate a novel whole-brain model paradigm allowing us to derive the generative underlying mechanisms for changing the arrow of time between brain regions in different conditions. We found precise, distinguishing signatures in terms of the reversibility and hierarchy of large-scale dynamics in three radically different brain states (cognition, rest, deep sleep and anaesthesia) in fMRI and electrocorticography data from human and non-human primates. Overall, this provides signatures of the breaking of detailed balance in different brain states, perhaps reflecting levels of conscious awareness.

Speaker: Prof. Gustavo Deco (Institució Catalana de Recerca i Estudis Avançats (ICREA), Pompeu Fabra University (UPF))

Presentation_Gustavo Deco_BrainComp 2022.pdf

12:45 Lunch Break

Connectivity: The network perspective

4 MEDUSA: an HPC-based simulation environment to create decoders of white matter microstructure

The Ginkgo team of NeuroSpin's BAOBAB/GAIA laboratory is developing within the framework of the Human Brain Project an environment called MEDUSA (Microstructure Environment Designer Using Sphere Atoms) that allows the creation of realistic virtual tissues representative of cellular environments encountered in the human brain, to simulate the scattering process and thus predict the MRI signal that you would get for each virtual tissue. The major contribution of artificial intelligence techniques opens up new perspectives for the development of in vivo imaging methods of the cytoarchitecture of the cortex since it becomes possible to exploit the microscopic information embedded in the water diffusion process present in the brain and whose trajectories embody an imprint of the local cytoarchitecture of the tissue. Initial results have shown that this environment allows the development of models of the local cytoarchitecture, such as computational models which are far more robust than the analytical models commonly used to decode the local cytoarchitecture of brain tissue. This approach based on the use of artificial intelligence techniques provides an appropriate framework for the decoding of the cellular disorders induced in the white matter in case of stroke.

Speaker: Dr Ivy Uszynski

Presentation_Ivy Uszynski_BrainComp 2022.pdf

Final questions and wrapping up

20:30 Dinner at the hotel's restaurant

21:30 Welcome party at the hotel's main terrace

Tuesday 20 September

Networks and brain segregation

5 Biophysics and data science approaches towards Central Nervous System translational medicine

Presenter: Giulia Rossetti1,2,3
Team: Rui Pedro Ribeiro1, Jonas Goßen1, and Alejandro Giorgetti1,4
1 IAS-5 / INM-9, Forschungszentrum Jülich, Germany; 2 Jülich Supercomputing Center, Forschungszentrum Jülich, Germany; 3 University Hospital Aachen, RWTH Aachen, Germany; 4 Department of Biotechnology, University of Verona, Italy

By merging structural macromolecular data with systems biology simulations, we developed a framework to simulate the signal-transduction kinetics induced by ligand-neuronal GPCR interactions, as well as, the consequent change of concentration of signaling molecular species, as a function of time and ligand concentration. Therefore, this tool brings to the light the possibility to investigate the subneuronal effects of ligand binding upon receptor activation, deepening the understanding of the relationship between the molecular level of ligand-target interactions and higher-level cellular and physiologic or pathological response mechanisms. We show here the application to the adenosine receptor A2a, where we combined structural data from drug-target interactions filtered by a random forest classifier with the above-mentioned signal-transduction kinetic model to identify new antagonists of adenosine receptor A2a. The latter has emerged as an attractive approach for treating Parkinson’s disease and other pathologies. Among the tested compounds in a radioligand binding assay, we found a promising drug candidate, which differs significantly from previously discovered ligands.

Speaker: Prof. Giulia Rossetti (University Hospital Aachen (RWTH University), Research Center Juelich)

Presentation_Giulia Roasetti_BrainComp 2022.pdf

6 Brain reconstruction from histology: Quantitative multimodal mapping of cell types in the (full) human brain

Realistic models of human brain function and disease require high detail, human-specific knowledge of brain architecture. The BigBrain data set has established a new standard for uniform, high resolution, rigorous maps of the human brain. We set out to expand this development, by integrating multimodal mapping of fibers and molecularly defined cell types. For this, we use updated histological, imaging, as well as analysis techniques based on high performance computing. We developed a protocol that allows for combining polarized light imaging with multiplex chromogenic immunohistochemistry, to image both fibers and molecularly defined cells and processes, followed by cytoarchitecture, sequentially, in the same full human brain section.

Speaker: Dr Roxana Koijmans (Netherlands Institute for Neuroscience)

7 What cortical folding patterns could tell us about individual brains?

Stacking memory and transistors in 3D is a kind of Graal for computer chip design, which could speed up transit times much beyond what is possible with 2D design. Cortical folding is the trick found by nature to achieve this goal. Each brain however has a specific cortical folding pattern, as unique as a fingerprint. This talk will describe a research program aiming at deciphering this variability, to understand whether each specific folding pattern corresponds to a specific cortical architecture.

Speaker: Prof. Jean-Francois Mangin (Atomic Energy and Alternative Energies Commission, NeuroSpin Research Centre )

Presentation_Jeff Mangin_BrainComp 2022.pdf

11:00 Coffee Break

Networks and brain segregation

8 Human brain segregation and networks

Brain segregation has been in the focus of research for more than 120 years. It has early been understood that the microstructural organization is closely linked to brain function and behavior, as formulated in early cytoarchitectonic (Brodmann, 1909), or myeloarchitectonic studies (Vogt and Vogt, 1919). Although Brodmann and the Vogts were working together in one lab, their cyto- and myeloarchitectonic maps follow different nomenclatures and ontologies. Even today, there are very different concepts for brain segregation as well as brain maps. We propose a multi-level human brain atlas that is based on coherent cyto-, fiber- and receptor architectonic mapping. Key elements are the three-dimensional cytoarchitectonic probabilistic maps of cortical areas and nuclei of the Julich Brain Atlas and the high-resolution BigBrain model, available at EBRAINS. The siibra software tool suite enables a programmatic access to the different maps and data that are anchored to the brain areas. Considering the large amount of data and resulting requirements of their processing and analysis, methods of High-Performance Computing are mandatory. The “Big Three”, cyto-, fiber-, and receptorarchitecture of the Julich Brain Atlas, represent a spatially and semantically organized reference to link structure and function across the scales.

Acknolwedgements: The Julich Brain Atlas project has received funding from the European Union’s Horizon 2020 Framework Programme for Research and Innovation under the Specific Grant Agreement No. 945539 (Human Brain Project SGA3) and was supported by the Joint Lab Supercomputing and Modeling for the Human Brain (SMHB) and the Helmholtz International BigBrain Analytics & Learning Laboratory (HIBALL).

Speaker: Prof. Katrin Amunts (Research Center Juelich, C.&O. Vogt Institute for Brain Research)

9 Dynamics of cellular environments underlying aging and Alzheimer’s Disease progression

The brain has a complex cellular environment, with a tight crosstalk between multiple cell types that is critical for its correct function. Despite extensive research much remains unknown regarding the effect and dynamics of the cellular environment of the brain on the progression of disease, especially in the context of progressive neurodegenerative diseases such as Alzheimer’s Disease (AD). I will discuss our insights from profiling the transcriptomes of millions of single cells from mouse models and human brain samples, which enabled us to expose the cellular cascade underlying the progression of AD and highlight specific cells. Specifically, by applying machine learning algorithms to single nucleus and bulk RNA-sequencing data, we built detailed cellular maps of the aging human brain, exposing the vast diversity of cells in the aging brain. We developed new computational approaches to capture cellular environments and follow their dynamics along disease progression. Our analysis associated unique subpopulations of glial cells to early and late stages of AD, and uncovered selective vulnerability of inhibitory neuronal subtypes. Moreover, constructing a manifold of cellular environments across individuals, revealed two trajectories that captured distinct cellular cascades associated with different disease outcomes, one of which captured the progression of AD. These new insights are shaping our understanding of the unique cellular environment of the healthy and Alzheimer’s disease brains and its dynamics along the progression of the disease.

Speaker: Dr Naomi Habib (The Edmond and Lily Safra Center for Brain Sciences at the Hebrew University of Jerusalem)

Abstract Habib_ Braincomp.pdf

13:00 Lunch Break

Networks and brain segregation

10 Mapping the basal ganglia microstructural changes in normal aging and Parkinson’s disease

It is essential to map structural changes in subcortical brain regions to gain a better understanding of their function in both health and disease. We developed a method for quantifying microstructure profiles in vivo in a single human brain. Our results demonstrate that spatial profiles in the putamen, caudate globus pallidus, and midbrain are robustly reproduced across individuals, clinical conditions, and datasets. By exploiting multiparametric quantitative MRI, we identify distinct, spatially dependent, aging-related alterations in water content and iron concentration. In Parkinson’s disease (PD) patients, we find abnormal profiles in the putamen, revealing changes in the posterior putamen that explain patients’ dopaminergic loss and motor dysfunction.

Speaker: Dr Aviv Mezer (The Edmond and Lily Safra Center for Brain Sciences at the Hebrew University of Jerusalem)

11 The Brain’s Linguistic Homunculus

What is the best way to understand the talking brain? Optimally, it is a computational model that has all the known neural properties, from which all known language phenomena are derivable. How close are we to this desideratum? Some argue that we are rather close, and present successes in the field of Natural Language Processing (NLP), as well as in neurolinguistics. I will review some large NLP systems, and show how they display systematic failures; I will then review two recent works that have attempted to fit NLP models to neural data, taken from experiments in sentence processing. Finally, I will present an alternative picture, and show the evidence that supports it: the talking brain is supported by a specialized language homunculus, whose organs align with the brain’s cytoarchitectonic parcellation, and their functional structure is best described by current linguistic theory. This homunculus, I will argue, is the object that needs computational modeling.

Speaker: Prof. Yosef Grodzinsky (The Edmond and Lily Safra Center for Brain Sciences at the Hebrew University of Jerusalem)

Presentation_Yosef Grodzinsky_BrainComp 2022.pdf

18:00 Coffee Break

Networks and brain segregation

12 Machine Learning in Bioinformatics: Efficient mining of omics data and clinical documents

Omics sciences (e.g. genomics, proteomics, and interactomics) are gaining an increasing interest in the scientific community due to the availability of novel, high throughput platforms for the investigation of the cell machinery, and have a central role in the so called P4 (predictive, preventive, personalized and participatory) medicine.
High-throughput experimental platforms and clinical diagnostic tools, such as next generation sequencing, microarray, mass spectrometry, and medical imaging, are producing overwhelming volumes of molecular and clinical data and the storage, integration, and analysis of such data is today the main bottleneck of bioinformatics pipelines.
On the other hand, textual documents, such as clinical records, Electronic Health Records, and patient's texts describing their healthcare experiences (Narrative Medicine), are more and more used in the biomedical research to extract patient's opinions and sentiments about their healthcare experience, by using NLP, Text Mining, and Sentiment Analysis methods.
Finally, networks are more and more used to model molecular interactions (e.g. biological pathways and protein-protein interaction networks), as well as brain structural and functional relationships (e.g. brain connectome), and the exploration (e.g. motif discovery) and comparison (e.g. network alignment) of such networks is gaining an increasing interest.
The talk introduces main omics data and presents some efficient bioinformatics and sentiment analysis tools and their application in biomedical research.

Speaker: Prof. Mario Cannataro (University of Catanzaro)

Evening Talk

13 Are we alone with our brains - thoughts on extraterrestrial life and intelligence

Speaker: Dr Frank Baetke

Final questions and wrapping up

Gala Dinner at the hotel's pool terrace

Wednesday 21 September

Networks and computing

14 Perspectives of the Federated HPC Infrastructure FENIX

Speaker: Prof. Thomas Lippert (Research Center Juelich, Frankfurt Institute for Advanced Studies)

15 Large-scale Deep Learning for Cytoarchitecture Classification in the Human Brain

The human brain can be subdivided into cytoarchitectonic areas. They are defined by the spatial organization of neuronal cells, including their distribution, size, type, orientation, as well as their arrangement into cortical layers and columns. Cytoarchitectonic areas are indicators for connectivity and function, making them a central component of multi-modal human brain atlases. Scaling the identification of cytoarchitectonic areas in histological brain sections to many brains and sections is critical to account for the high variability between brains, motivating the development of automated methods for cytoarchitecture classification. We provide an overview of current deep learning-based classification methods, describe associated methodological and technical challenges, and preview future developments.

Speaker: Christian Schiffer (Research Center Juelich)

Presentation_Christian Schiffer_BrainComp 2022.pdf

11:00 Coffee Break

Young Researchers Session

16 Cytoarchitectonic mapping, 3D-reconstruction, and texture analysis of the human bed nucleus of the stria terminalis

The bed nucleus of the stria terminalis (BST) is a basal forebrain structure mainly involved in anxiety disorders and stress response. Its small size and having cell densities similar to its surrounding structures lead to difficulties in precise delineations using common MRI techniques. Histology-based maps help to overcome these challenges and can serve as a spatial and structural reference.
This talk gives an overview about the neuroanatomy and microstructure of the BST and its subdivisions. Cytoarchitectonic mapping and textural analysis as a quantitative validation method will be explained. Finally, two sets of maps will be presented: First, probabilistic maps, showing the localization and interindividual variability in MNI standard reference spaces. And second, a surface-based high-resolution 3D-reconstruction of the BST in the BigBrain will be shown to visualize the complex shape of the BST and its subdivisions in high anatomical detail.

Speaker: Andrea Brandstetter (Research Center Juelich)

Abstract_Braincomp_Brandstetter.pdf

BrainCompTalk_Brandstetter.pdf

17 Cytoarchitectonic mapping of the human olfactory tubercle and terminal islands

The olfactory tubercle and terminal islands are part of the basal forebrain, a brain area
characterized by great structural heterogeneity. The connectivity and function of the
olfactory tubercle have been evaluated using diffusion imaging and fMRI, however the
precise boundaries within 3D space are still not clarified (Zelano et al., 2007; Echevarria-
Cooper et al., 2022). In addition, it is particularly challenging to study these aspects of
terminal islands due to their small size, complex shape and scattered arrangement in the
basal forebrain (Meyer et al., 1989). Therefore, we have generated cytoarchitectonic maps
of the olfactory tubercle and terminal islands in order to determine their localization in the
stereotaxic space and intersubject variability. Moreover, we provided a 3D reconstruction of
the Great Terminal Island (GTI) in the BigBrain, to define its anatomical features.

Speaker: Joko Poleksic (University of Belgrade, Institute of Anatomy Niko Miljanic)

Abstract_Braincomp_Poleksic.pdf

Presentation_Joko Poleksic_BrainComp 2022.pdf

18 Variations in the shape of the central sulcus in hominids

While the folding pattern is stable and simple in small mammalian brains, it becomes rich and complex in larger brains, reaching a peak of complexity and inter-individual variability in humans.
However, among this variability some major features are stable within species and even between species. We examine the morphological variability of the central sulcus in regard with the underlying connectivity in hominids, within and between species, with a focus on the motor hand region and its significant issue of human evolution.

Speaker: Dr Ophélie Foubet (Atomic Energy and Alternative Energies Commission)

19 Training physical models of deep spiking neural networks

The nervous system accomplishes its energy-efficiency through a combination of analog computation and a sparse event-based communication. Especially the latter can pose challenges to learning or training strategies. The recent adoption of surrogate gradients, however, paved the way for an effective gradient-based training of spiking neural networks (SNNs). This talk presents a flexible and robust surrogate-gradient-based training framework for the mixed-signal neuromorphic system BrainScaleS-2. It overcomes the challenges inherent to the analog nature of the neuromorphic SNN and allows the training of multilayer networks with both feedforward and recurrent topologies. It reaches a performance on a par with software simulations on both vision and speech benchmark datasets. The framework can cope with even deliberately emphasized device mismatch and self-corrects for the resulting inhomogeneities in the neural dynamics. Finally, the combination of efficient coding schemes as well as the accelerated nature of BrainScaleS-2 yields quick inference latencies and a high classification throughput.

Speaker: Sebastian Billaudelle (Heidelberg University, Heidelberg, Germany)

Presentation_Sebastian Billaudelle_BrainComp 2022.pdf

13:00 Lunch Break

Networks and computing

20 Representation learning with trainable COSFIRE filters

Speaker: Prof. Nikolai Petkov (University of Groningen)

20220921Petkov-Braincomp.pdf

21 Pushing Global Brain Tractography towards Micrometer Resolution with Exascale Computing

Diffusion MRI provides orientation information about fiber tracts in the brain. To quantify connectivity, tractography has been developed to reconstruct fiber trajectories based on orientation distributions. Conventional tractography methods relying on analytical ODF models and streamlining methods are afflicted with artifacts and do not consider anatomical information as a prior. In addition, tractography results vary depending on the selected seed points and generate individual trajectories that are not related to their neighborhod.

Global spin-glass approaches using Markov Random Fields (and consequently associated to a Bayesian framework) were introduced at the beginning of the 2000’s [1] and proposed a suitable general framework to incorporate any kind of anatomical prior. However, global spin-glass approaches are computationally intensive. The construction of the entire set of connections is computed simultaneously in a competitive way, to reach an optimal solution. This solution to the problem corresponds to the minimum of a target energy composed of various potentials, such as a data attachment potential, various energy potentials introduced to constrain the model with anatomical or microstructural priors, and further regularization potentials. Recently, efficient global spin-glass methods have been introduced that used simplified anatomical priors (e.g., low curvature of axonal fibers) [2, 3, 4]. These algorithms enable tractography on a standard desktop computer applied to an individual brain at a millimeter resolution and reconstructs a hundred thousands of fibers in half a day.

We are currently acquiring high-resolution microscopic diffusion MRI data with a resolution of 300 micrometers using a 7T MRI system at Neurospin, France. In addition, we apply 3D Polarized Light Imaging (3D-PLI) microscopy to the same brain tissue to determine fiber orientations at a resolution of a few micrometers [5]. These microscopic data sets will be a few petabytes in size, which requires the global tractography algorithm to be able to handle such massive data in terms of communication and distribution in an efficient way. We are currently developing an open-source software that is designed to work on single machines as well as on high performance computing architectures, including the GPU architecture.

1. Poupon, C., C. A. Clark, V. Frouin, J. Régis, I. Bloch, D. Le Bihan, and J.-F. Mangin. 2000. “Regularization of Diffusion-Based Direction Maps for the Tracking of Brain White Matter Fascicles.” NeuroImage 12 (2) : 184–95.

2. Fillard, Pierre, Cyril Poupon, and Jean-François Mangin. 2009. “A Novel Global Tractography Algorithm Based on an Adaptive Spin Glass Model.” In Medical Image Computing and Computer-Assisted Intervention – Miccai 2009, edited by Guang-Zhong Yang, David Hawkes, Daniel Rueckert, Alison Noble, and Chris Taylor, 927–34. Berlin, Heidelberg: Springer Berlin Heidelberg.

3, Mangin, J.-F., P. Fillard, Y. Cointepas, D. Le Bihan, V. Frouin, and C. Poupon. 2013. “Toward Global Tractography.” NeuroImage 80: 290–96.

1. Christiaens, Daan, Marco Reisert, Thijs Dhollander, Stefan Sunaert, Paul Suetens, and Frederik Maes. 2015. “Global Tractography of Multi-Shell Diffusion-Weighted Imaging Data Using a Multi-Tissue Model.” NeuroImage 123: 89–101.

2. Axer, Markus, Katrin Amunts, David Grässel, Christoph Palm, Jürgen Dammers, Hubertus Axer, Uwe Pietrzyk, and Karl Zilles. 2011. “A Novel Approach to the Human Connectome: Ultra-High Resolution Mapping of Fiber Tracts in the Brain.” NeuroImage 54 (2): 1091–1101.

Speaker: Mr Felix Matuschke (Research Center Juelich)

BrainComp22_Felix_Matuschke.pdf

18:00 Coffee Break

Networks and computing

22 Challenges and opportunities for large-scale federated science and computing in the EU landscape

This presentation gives a bird’s eye view on the science, software, computing, and funding landscape in the EU, identifying potential challenges and opportunities.
After an introduction of the constraints and requirements informing this work, a high-level map of the ecosystem will be introduced. This ecosystem spans from science end-users to systems and service provides, funding agencies and other stakeholders. The natural point-to-point interactions that occur in this ecosystem appear at odds with a ‘gate keeping’ platform-based approach to large-scale science infrastructures compounded by the shift to national funding in the context of the ESFRI mechanisms.
The second part of the presentation identifies a number of key activities and practices already performed by the stakeholders in this
landscape: FAIR practices, the peer review process, science & software & systems operations. A co-design process with the working title Living Science appears at the convergence of these key activities. Living Science re-contextualizes existing science and technology practices into a process with a clear value proposition to stake holders in this
landscape: owner ship of working science now and in the long term.

Speaker: Wouter Klijn (Research Center Juelich)

23 Integrating Fenix into EBRAINS – What can we do to simplify the usage of large-scale systems for the users?

Digital brain research requires access to computational resources, such as cloud, HPC, neuromorphic and smaller systems. Fenix offers access to federated HPC and Cloud resources and provides services to interact with these resources. However, many users are not proficient in using these systems efficiently and are typically more interested in working on their scientific problems than finding the best computing configuration for their specific problem and adapting it to different computing architectures. HPC plays a special role here as every system requires its own optimizations and often a user first must learn how to use such a system. A future digital brain research infrastructure should therefore support users in these issues while allowing transparent use of the system and leaving many optimizations to the runtime system. In this talk, different aspects of such a runtime and support system will be discussed. While performance was often one of the main criteria for choosing and optimization of a system, in the future another aspect, for example, energy consumption or idle time will be more important. Here, a federated infrastructure can help to optimize scientific workflows across different sites.

Speaker: Prof. Lena Oden (FernUniversität Hagen)

Final questions and wrapping up

20:30 Dinner at the hotel's restaurant

Piano concert at the hotel lobby

Thursday 22 September

Perspectives of computing technologies to decode the human brain

24 Brain Models and Digital Twins: Towards a proactive ethical approach

Computational models of the brain provide a unique opportunity to study the brain and to contribute to progress in personalized and precision medicine for brain diseases. Despite their clear utility, however, they raise some ethical, philosophical, social, and cultural questions. Identifying and examining them is key.
One of the first challenges that the identification and examination of these issues encounters is that of undermining the mistaken assumption that science and scientific research are value neutral and separated from their many contexts and that they should carry on unencumbered by “non-scientific” considerations about ethics, culture, and society.
In this talk, I begin by highlighting the importance of ethics as an integral part of scientific research and of attending to and managing ethical and societal considerations from the beginning of the research process. Then I provide a snapshot of some of the current ethical and philosophical issues raised by computational modeling in general in the context of healthcare. I finish by briefly discussing whether anticipated scientific developments in digital twin technology might raise novel concerns.

Speaker: Dr Arleen Salles (Center for Research Ethics and Bioethics (CRB) at Uppsala University)

Presentation_Arleen Salles_BrainComp 2022.pdf

25 Spike-based analog computing

Spike-based neuromorphic computing realizes an in-memory, event-based computing paradigm. By transferring results and ideas from Neuroscience to technology, it allows us to overcome the power wall our CPU-centric CMOS technology is facing.
This talk will present an analog hardware realization of spike-based neuromorphic computing developed at Heidelberg University: The BrainScaleS system.
It will summarize how the Heidelberg BrainScaleS accelerated analog neuromorphic architecture implements neuroscientific principles of neural dynamics, spike communication and local learning. BrainScaleS combines power efficiency with the necessary flexibility and programmability to reduce the resource requirements of AI as well as modelling biology.
The talk will demonstrate how analog, spike-based brain inspired computing is playing an important role in the search for novel computing technologies.

Speaker: Dr Johannes Schemmel (Heidelberg University, Heidelberg, Germany)

Presentation_Johannes Schemmel_BrainComp 2022.pdf

26 Enabling Neuromorphic Learning Machines with Multilevel Learning

Modeling successful online learning in real-world environments is an aspirational goal of neuroscience and artificial intelligence technologies. Using deep learning for modeling synaptic plasticity in the brain has led to recent breakthroughs, but standard deep neural networks struggle to achieve real-world, online learning. Furthermore, the randomized, energy- and data-intensive process for training them is incompatible with the physical nature and online operation of the brain.
Thanks to their evolution and multiple development stages, brains can integrate new knowledge by leveraging prior knowledge and structure, and do so with unparalleled efficiency.
In this talk, I will discuss multistage learning methods that can “train-to-learn” brain-inspired, neuromorphic systems that capture the brain’s architecture and dynamics.
These methods enable shifting the complex, energy and data-intensive stages of learning to data centers and away from the edge where power and area are limited, thus paving the way towards fast and energy-efficient neuromorphic learning machines.

Speaker: Prof. Emre Neftci (RWTH Aachen University, Research Center Juelich)

Closing

13:00 Lunch Break

Optional Meetings

20:00 Dinner at the hotel’s restaurant

21:00 Wine tasting at the hotel’s main terrace

Friday 23 September

Optional Meetings