Tutorials & E-Library

The HBP Neurorobotics Platform (NRP) lets you simulate robots controlled by neural networks in dynamic environments over the internet. It offers a variety of tools to design and customise virtual robotics experiments.

Many users have asked how to deal with simulations that generate a lot of data that must be saved. The simplest strategy for saving simulation results is to capture them to Vectors during the simulation, and write them to one or more files after the simulation is complete. This works as long as everything fits into available memory.

If memory limits are exceeded, the first thing to do is go back and decide whether you really need to keep all of that stuff. For example, when simulating spiking network models, the only data that must be saved are the spike-gid pairs. Given these it is possible to go back after the end of a simulation and reconstruct the details of activity in a subnet--but that's another story. In this document we presume that you have already selected just those variables that are absolutely essential.

The answer to your problem is to break the simulation into shorter segments, each of which is brief enough to avoid memory overrun, and save the results from each segment before advancing to the next. After the last segment has been executed and its results saved, you can assemble the segmented data into a single record, if you like. Here are a couple of simple examples to help you get started.

The availability of a large number of models in ModelDB (Migliore et al 2003), helps investigators test their intuition of model behavior and provides building blocks for future modeling applications to the interpretation of experimental findings. However the NEURON (Hines and Carnevale 2001) model legacy code entered by publication authors was generally not developed with presentation as a high priority. The original code can be difficult to analyze and it sometimes happens that variables are reset so that the values at run time are different than the first values indicated in the top of the code. ModelView overcomes these problems by providing a (run-time state) preview of the properties of a model (anatomy and biophysical attributes). Having this information available for viewing in ModelDB lets investigators quickly develop a conceptual picture of the model structure and compare parameter differences between runs. It makes it possible to ask detailed questions about the model that would have been time-consuming to answer without ModelView.

NEST Desktop is a web-based GUI application for NEST Simulator, an advanced simulation tool for the computational neuroscience.

The app enables the rapid construction, parametrization, and instrumentation of neuronal network models.

This notebook will explain how to set up an optimisation of simple single compartmental cell with two free parameters that need to be optimised. As this optimisation is for example purpose only, no real experimental data is used in this notebook.

The EBRAINS Live Papers are structured and interactive supplementary documents to complement journal publications, that allow users to readily access, explore, and utilize the various kinds of data underlying scientific studies. Interactivity is a prominent feature with several integrated tools and services that will allow users to download, visualize, or simulate data, models, and results presented in the corresponding publications.

In this tutorial-cum-demonstration, you will learn how to view published Live Papers and how to access and explore the resources provided in these Live Papers.

While the Pixel Classification module is useful for segmenting objects with discernible brightness, color or textural differences in comparison to their surroundings, the carving module’s purpose is to aid in the extraction of objects from images that are only separable by their boundary - i.e. objects that do not differ from the rest of the image by their internal appearance.

The purpose of this workflow is to enable counting of objects in crowded scenes such as cells in microscopy images. Counting is performed by directly estimating the density of objects in the image without performing segmentation or object detection.

This tutorial will guide you through this workflow and teach you how to use it for your research.

ilastik object classification workflow tutorial. As the name suggests, the object classification workflow aims to classify full objects, based on object-level features and user annotations. An object in this context is a set of pixels that belong to the same instance. Object classification requires a second input besides the usual raw image data: an image that indicates for pixels whether they belong to an object or not, i.e. pixel predictions, a segmentation or a label image. This can be obtained e.g. using the Pixel Classification Workflow. The workflow exists in two variants to handle different types for the second input:

Video demonstrating how to export ilastik tracking results for visualization/proofreading to MaMuT.

The Pixel Classification workflow assigns labels to pixels based on pixel features and user annotations. The workflow offers a choice of generic pixel features, such as smoothed pixel intensity, edge filters and texture descriptors. Once the features are selected, a Random Forest classifier is trained from user annotations interactively. The Random Forest is known for its excellent generalization properties, the overall workflow is applicable to a wide range of segmentation problems. Note that this workflow performs semantic, rather than instance, segmentation and returns a probability map of each class, not individual objects.

This video illustrates the ilastik Tracking workflow, starting from raw data and an existing segmentation.