Lectures | Work Groups | Discussion | Recreation | Social | Create New Session



The participants introduce themselves to the Workshop - using only 6 words!

General introduction to the way neuromorphic life is at CC.

Then Sebastianian introduces the main conecpts of Neuromorphic Engineering, which leads on to general discussion on what we think we are doing here.

Mercifully, there will be a …

Go to group page

Fast and Slow learning

One-shot concept acquisition, vs incremental calibration and tuning.

Florian directs the process...

Go to group page

Information Processing and Plasticity: the role of the dendrites
The Role of Time in Computation: computing with dynamics
Exploration and Active Sensing
Embodiment and Morphological Computation
Balanced Activity & Working Memory
Lessons from Machine Learning
Control and Error Prediction

Work Groups

CLUSTER: In-memory computing and learning in hardware

#### PUT A LINK HERE TO YOUR WORKGROUP IF IT RELATES TO THIS THEME (your workgroup can be linked to multiple clusters) ######

Go to group page

CLUSTER: Few(one)-shot online learning amenable to hardware implementation

PUT A LINK HERE TO YOUR WORKGROUP IF IT RELATES TO THIS THEME (your workgroup can be linked to multiple clusters) ######

Announcement of a discussion group around the issues of single-shot and single-pass learning mechanisms:  https://capocaccia.cc/en/event/ccnw23/workgroup/options-for-single-pass-and-single-shot-learning-mechanisms/


Go to group page

workgroup test
CLUSTER: Robotics, control, and reinforcement learning

Real-time event-based laser tracking: https://capocaccia.cc/en/event/ccnw23/workgroup/real-time-event-based-laser-tracking/

#### PUT A LINK HERE TO YOUR WORKGROUP IF IT RELATES TO THIS THEME (your workgroup can be linked to multiple clusters) ######

Go to group page

ALIVE for transfer/few-shot learning

I will bring a neuromorphic prototype chip with on-chip learning. The chip is called ALIVE, which stands for Always on Learning wIth Various synapsEs.

The ALIVE chip consists of two neural cores that implement two different dendritic error-based learning rules using multi-compartment neurons with several active dendritic branches. Both cores …

Go to group page

CLUSTER: Sensory processing

#### PUT A LINK HERE TO YOUR WORKGROUP IF IT RELATES TO THIS THEME (your workgroup can be linked to multiple clusters) ######

Speech denoising group: https://capocaccia.cc/en/event/ccnw23/eventsession/275/detail/

Go to group page

CLUSTER: Biological realism in modelling network activity

#### PUT A LINK HERE TO YOUR WORKGROUP IF IT RELATES TO THIS THEME (your workgroup can be linked to multiple clusters) ######

Plastic balanced random networks in NEST: https://capocaccia.cc/en/event/ccnw23/eventsession/269/detail/


Go to group page

Example workgroup

dummy text

Go to group page

Few short learning NeuroBench challenges

One of the proposed tasks in the new NeuroBench suite (http://arxiv.org/abs/2304.04640) involves "Adaptive Learning of Keywords and Gestures".

In this workgroup we'll have a go at applying various existing approaches to this task (and, obviously, any amazing new ideas you come up with at Capocaccia)

Go to group page

Event-based animal tracking

Tracking animals, moving through natural environments is a big challenge in biology. Historically, this has been semi-automated with students-in-the-loop but this approach doesn't really scale to long experiments/lots of individuals.

Event-cameras are really interesting in this context as they 'automatically' remove the background and extract motion. Last summer I took …

Go to group page

CLUSTER: Biological realism in modelling computational primitives

#### PUT A LINK HERE TO YOUR WORKGROUP IF IT RELATES TO THIS THEME (your workgroup can be linked to multiple clusters) ######

Go to group page

Accelerated analog computation with BrainScaleS-2

We bring along multiple BrainScales-2 systems and invite everyone to try some of our demos or implement their own neuromorphic experiments.

The most recent generation of BrainScaleS-2 ASICs features 512 analog multi-compartment neuron circuits with 256 plastic synapses each.
The analog circuits are tightly coupled to on-chip digital event routing …

Go to group page

Macroscopic Neuron Challenge with Lu.i

The Lu.i neuron PCB

Lu.i (phonetic notation of »Louis«, an homage to Louis Lapicque, who first formulated the leaky integrate-and-fire model) is an electronic neuron circuit mimicking and illustrating the basic dynamics of real, biological neurons. The printed circuit board (PCB) features a configurable, fully analog implementation of the leaky …

Go to group page

SNNs for Human Pose Estimation (or something like that)

Human Pose Estimation task is extremely challenging with event-driven cameras. We have been able to do it with an ANN approach but we are curious how would we go about approaching with an SNN. It's a 13 point skeleton regression problem with 2D pose from a single camara, in the …

Go to group page

GPU enhanced neural networks

Fancy running your SNNs 10x faster? Our GPU enhanced Neuronal Networks (GeNN) library is freely available from https://genn-team.github.io/ and provides an environment for GPU accelerated spiking neural network simulations. GeNN is capable of simulating large spiking neural network (SNN) models at competitive speeds, even on single, commodity GPUs. In GeNN, …

Go to group page

Balancing real or simulation cart-pole system with RL

Equipment: Cart-pole simulator and real cart-pole used by Tobi Delbruck for MPC tests


Possible directions:

  • Online Spiking RL (R-Eprop) for non-ideal cartpole stabilization solving in simulation after transfer learning of weights for ideal cartpole trained with DQN 

  • Deep Q-Learning for cartpole swing-up in simulation

  • Implementation of inference …

Go to group page

Argo robot sailboat

The Argo Joyways Drangonforce 65 boat with Rasberry Pi 3, 9-DOF IMU, GPS, and 3 axis wind sensor. Aim is to get it sailing and collect some data towards imitation-learning to sail with simple MLP or RNN. See slack #hardware-argo channel.

Go to group page

Plastic balanced random networks in NEST

In this workgroup, we will train spiking balanced random networks via supervised eligibility propagation (Bellec et al., 2020) in the NEST simulator (Sinha et al., 2023). One concrete idea would be to endow a network including spiking central pattern generators (CPGs) for locomotion to drive a robotic arm. Traditionally known …

Go to group page

Real-time event-based laser tracking

Goal: Track a laser pointer (controlled by a human) with a second, programmable laser pointer based on the input from an event camera in real-time. One additional twist is that the event camera can move! So, the mapping should, ideally, be adaptive. The controllable laser should also be governed …

Go to group page

Using Lu.i Dynamics for CartPole Stabilization

This workgroup is now CLOSED!

  1. We aim to design an LQR controller for the linearized CartPole about its unstable equilibrium. This is very standard fro linear control systems.

  2. Next we want to map the feedback coefficients to an Lu.i neuron.

  3. Bringing the LIF neuron close to its linearity limits, we …

Go to group page

Speech denoising with RNN-based SNNs

Hearing impairment is a large societal problem, and speech enhancement is one type of help for this problem. Speech enhancement aims to filter clean speech from speech mixed with background noise. Such enhancement would ideally be integrated with hearing aides. The power, latency and energy constraints of hearing aides make …

Go to group page

Sandcastle Beachmarking

Expanding silica to the cm scale

Go to group page

Discussion Groups

Accessible event-address processing: lowering the barrier for neuromorphic tech

Neuromorphic sensors typically emit address-event representations (AER), but integrating such sensors with our infrastructure is difficult and time consuming. We all feel this pain, and we can do much better as a community.

In this discussion session, we'll present our take on how we can lower the barrier-to-entry for working …

Go to group page

Options for single-pass and single-shot learning mechanisms


As well as being a notable feature of biological learning mechanisms and therefore worthy of study in its own right, single-shot and single-pass learning offers many advantages including (potentially huge) gains in speed and energy use, as well as continuous and adaptive learning.  I would like to explore the …

Go to group page

Large language models on neuromorphic hardware

We will discuss large language models from various neuromorphic perspectives:

  • implementation in hardware (inference)

  • few-shot learning / in-context learning

  • alternative model design suited for neuromorphic devices

What we will not discuss:

  • training (online/offline)

  • ethics

  • Is it really intelligent?


Go to group page

Thinking beyond transformers: Can the brain do better?

Is attention all you need?

How does the success of transformers affects our approch to designing intelligent systems?

Does the brain follow similar principles, and can brain-inspired models do better?

What we do not talk about in this session:

  • hardware implementation

  • ethics

  • AGI

Go to group page

Computational Zones of the Pyramidal Cell

I will review results in the literature that suggest that neocortical pyramidal cells should have three separate computational zones:

1) soma and proximal dendrites
2) distal dendrites
3) apical tuft

The reason why this level of detail is needed is that each of these zones differ significantly in several key …

Go to group page

Neuromorphic chips: why so many?

Brief intro on (our understanding of) why there are so many different flavors of neuromorphic chips and discussion on what's the point of doing this.

Within the community there are different views/opinions, and even within the CapoCaccia participants there are different viewpoints. 

There will likely be lively discussions! Be sure …

Go to group page

Understanding Transformers: Intuition and mathematics of attention

We give a thorough introduction to the intuition and equations of self-attention and transformers. No prior experience needed. This discussion serves to get us on the same page for further discussions on large language models (hardware implementations, new algorithm design, few-shot learning, is it AGI?).

Go to group page

How to control an elephant: the time constants of the elephant brain

Florian will answer the question: "why bigger animals have bigger brains?"

Go to group page

Neuro to Silicon

This discussion group is meant to bring algorithmic ideas, be it inspired by neuroscience or machine learning to analog or digital CMOS hardware. One whiteboard is reserved for people with ideas that they think, are worth investigating and one is reserved for people that have an idea for an analog …

Go to group page

Neuromorphic touch

We will discuss which are the encoding principles of biological touch and how we can derive useful processing circuits from biology. 

Go to group page

Are microphysiological systems ready to integrate with neuromorphic computing?

Microphysiological systems (MPS), often labeled as Organ-on-chip technologies, are developing rapidly for medical and pharmaceutical testing, however there have been a questions about whether these technologies could be implemented as a computational system. MPS are primarily appealing with the advent of induced pluripotent stem cells, where differentiated neurons do not …

Go to group page

Computing with Dynamics: Why aren't we doing this in Hardware?

Together with the speakers of the morning session, we will continue discussing the approach of "computing with dynamics". We want to start a dialog with computational neuroscientists and engineers: why aren't we implementing theses systems in hardware? what are their limitations? what needs to be solved so that we can …

Go to group page

Practical PyTorch for event-based computing

This session consists of two parts: a whirlwind introduction to PyTorch and its use in neuromorphics.

The first part (20 minutes) will explain the fundamentals of PyTorch by

  • How PyTorch works with tensors and network layers

  • How PyTorch propagates gradients

  • How PyTorch can optimize networks

The second part (30 minutes) …

Go to group page

Memristors: why, what and how?

In this discussion group we will introduce different memristor types, their strengths and weaknesses, and applications. 

Go to group page

Control with mixed-feedback excitable neural systems: exploiting the biophysics of spiking neurons

One of the main obstacles in the development of analog neuromorphic hardware has proven
to be their increased fragility to noise and fabrication mismatch. Biological systems, however, address similar limitations with amazing effectiveness and efficiency through
their unique multiscale, mixed-feedback organizational principles. Rodolphe will introduce his ongoing efforts to develop …

Go to group page

How are Mind Patterns represented in the Brain?

Christoph, the great, will tell us about his brain's theories.. and remember he wants to understand the brain! ;)

Go to group page

Methods in neuroscience

A short discussion of how neuronal data is collected.

We will talk about the main methods of collecting neuronal data. Advantages and disadvantages of different methods.

The main methods we will discuss will be electrophysiology and imaging.

Time scales, spatial resolution, circuit analysis, and methods to perturb the activity.

If …

Go to group page

Learning as synapse-level active inference

David will continue with what he was telling us his morning discussion on monday!

Go to group page

Predicting the future and solving weight transport in under 60'
Sunny on how to talk to the public about silicon Brains!
Making neuromorphic benchmarks embodied

We will brainstorm new benchmarks with embodied robotics and neuromorphic hardware that can push forward both our communities :) 

Go to group page

Computational role of long-range connectivity in cortical circuits

Discussion with Henry Kennedy on the computational role of long-range connectivity in cortical circuits

Go to group page

Recreation Groups


Lets gather together every other day to play football at the hotel's court! Or as the italians call it "calcetto" :) 

Go to group page


A group to organize outdoor rock climbing during the workshop. Please add yourself if interested! More info to follow...

Go to group page


For anyone interested in playing some matches :)

Go to group page

Morning meditation

Let's say 30 to 60' of yoga and meditation on the beach. Meeting at 6:30am in the bar terrace.

Go to group page

Football Champion's League Semi-final: One-shot temporal dynamics of the ball

Come to watch one of the best games of the beautiful game this season, Real Madrid vs Manchester City.

Go to group page

Grappling fundamentals

Outside of research, there is but one other discipline that takes you to the limit of what your mind and body can sustain. Join us for a grappling seminar on the lawn next to the disco to find out more!

The art of gently submitting your opponents to your will …

Go to group page

A peek around the old Capo Caccia hotel

Social Events

Workshop Dinner
Welcome Apero