Topic: convolutional networks

A two-day live course introduces fully connected neural networks, convolutional networks for computer vision, recurrent networks for text and speech, transfer learning and large language models. Participants use PyTorch to build and train models on high-end GPUs through an HPC cluster, connecting from their own computers to the Austrian Scientific Computing JupyterHub. Temporary cluster accounts support the exercises, and a short pre-assignment prepares participants before the course. The English-language course is intended for software developers, data scientists and researchers moving from

Build and evaluate deep-learning models for biological prediction using Python and Jupyter notebooks. Lectures and collaborative exercises develop convolutional networks for classification, regression and image segmentation, with attention to preprocessing, augmentation, class imbalance, cross-validation and overfitting. Later sessions explore learning-rate and training-duration choices, recurrent networks and applications to omics data. Participants interpret model performance and discuss their own research problems. 2026-11-23–2026-11-27, 14:00–20:00 daily, Europe/Berlin (CET, UTC+01:00). O

The brain must recognize objects in the face of identity-preserving transformations such as changes in lighting, position, and orientation. However, to achieve general purpose geometric reasoning these view transformations should also be represented in neural codes. A natural strategy to encode both the identity of an object and its view is to utilize an equivariant code, where the neural representation transforms in a manner consistent with the transformations to the inputs, and consistent across objects. A classic example of such a code is given by an intermediate layer of a convolutional ne

Weight sharing among neurons is widely used in deep learning: convolutional networks are an obvious example, but in addition transformers need it for matrix-matrix multiplications. Without weight sharing, deep networks perform badly on hard tasks. This is potentially problematic for the brain, since weight sharing is biologically implausible - a fact that is ignored in deep learning models of brain activity, especially of visual processing, which are becoming widely used. Recently it was shown that partial weight sharing can be implemented with a "sleep phase", in which plasticity is anti-Hebb

Training deep neural networks typically relies on backpropagating high-dimensional error signals---a computationally intensive and biologically implausible process. However, since most tasks involve low-dimensional outputs, we propose that low-dimensional error signals may suffice for effective learning. To test this hypothesis, we introduce a novel local learning rule based on Feedback Alignment that leverages indirect, low-dimensional error feedback to train large networks. Our method decouples the backward pass from the forward pass, enabling precise control over error signal dimensionality