Work Groups
Brief introduction to using SpiNNaker
Due to a bit of interest there will be an informal introduction on how create and run PyNN simulations on SpiNNaker
Neuromorphic Cognitive Robots
In this group, we will put together neuromorphic hardware (ROLLS, DYNAPxxx, DAVIS, and Loihi), robots (miniature vehicles Pushbots and Omnibot and a stereo pan-tilt unit), and neuronal architectures to work on a number of high-level "cognitive" problems: neuromorphic SLAM, sequence learning, learning sensorimotor mappings, active stereo vision, and long term ...
Introduction to using the open-source HDL verilog code of ODIN
ODIN is an online-learning digital spiking neuromorphic processor designed and prototyped in 28-nm FDSOI CMOS at Université catholique de Louvain (UCLouvain).
It consists of a single 256-neuron 64k-synapse crossbar neurosynaptic core with the following key features:
- synapses embed spike-dependent synaptic plasticity (SDSP)-based online learning,
- neurons can phenomenologically reproduce the ...
Hands-on tutorial: how to design and execute plasticity experiments on the BrainScaleS-2 system
With an examplary reinforcement learning task at hand, you will
- understand the analog, accelerated BrainScaleS-2 architecture,
- learn how to set up experiments on the prototype systems,
- and write code for the embedded plasticity processing unit.
During the course of this tutorial you will work on a system provided by us ...
'Insect people' united
We will never be divided (by cortical chauvinism)
Current sub-projects are:
* Event-based stereopsis (James)
* Sub €100 panoramic UV camera hacking (Jamie)
* Fusing visual navigation with olfactory cues (Jamie)
* Modelling of Drosophila Larvae on DyNapse (Afshin)
The aim is to have an initial meeting fairly early on and figure out hardware ...
DVS/DAVIS event camera crash course
I brought along about 3-4 dynamic vision sensors (DVS) and dynamic and active pixel vision sensors (DAVIS) that can be used for different projects.
Depending on interest, we can decide the number of afternoon sessions.
In the workshop I will present:
1. software in Java and C++ for recording, viewing ...
Hands-on GeNN tutorial
GeNN is a GPU enhanced Spiking Neuronal Network simulation library based on NVIDIA CUDA technology. It can be used directly from C++ or via Brian 2 and PyNN interfaces.
We have a couple of robots with onboard GPUs with us at Capocaccia and GeNN can be used to control these ...
Event Driven Perception for Robotics
Event Driven Perception for Robotics wishes to induce a paradigm shift in robotics, based on the biologically inspired emerging concept of event-driven sensing and processing.
This workgroup aims at merging event-driven vision with classical robotics problems. Particularly, we are currently implementing an attentive system based on a neuromorphic algorithm by ...
Sea Eagle - Localization of floating object at sea
Safe navigation of vessels has become a major challenge. Large commercial vessels can afford sophisticated and expensive devices able to alert for the proximity of floating objects. Small craft must relay on human vision. Although most commercial cameras cannot compete with the human eye, an optical device could still provide ...
Robotic Gas-Based Navigation
We have created a small mobile robotic platform which is setup with a three Raspberry Pi 3's, a Raspberry Pi Screen, a SpiNN3 and two Enoses created by one of our lab postdocs (Damien Drix). Each Enose is made of four metal oxide (MOX) sensors and a temperature and humidity ...
Closing the motor control loop with neuromorphic hardware
We focus on solving an engineering problem of interfacing (i) sensory input (DAVIS camera), (ii) a spiking network on a neuromorphic processor and (iii) a motor output (robotic arm) to build an event-based framework to test some learning schemes (unsupervised STDP and reinforcement learning) that close the perception-action loop.
As ...
Neuromorphic tactile/vibration processing
In this workgroup, we would like to explore/apply time-based computation for/to tactile processing.
We intend to develop a prototype of event-based vibration measurements units and use it to transform any surface into a tactile surface.
Sound propagation, and more specifically time-of-arrival difference of a pulse will allow us ...
iCub Tutorial
We've brought our iCub humanoid robot from the Italian Institute of Tecnology. This iCub comes equipped with two ATIS neuromorphic sensors, tactile skin over the arms and torso, and IMU sensor in the head and encoder information for each joint positions. We can control the robot in 6-DOF for the ...
Sensory fusion with EMG and DVS/DAVIS sensors
As the traditional single sensor information acquisition system is becoming more and more difficult to meet the complex practical applications, the fusion application of multisensory have been made a great deal of research. In this workgroup, we propose to fuse EMG and DAVIS signals to classify the human upper limb ...
CortexControl crash course - an easier way to use Dynapse
In this workgroup we will present CortexControl, a software that offers a python console and a GUI to use the Dynapse.
We will show how to:
* Connect to a Dynap-SE
* Tune biases
* Use RPyC
* Receive events
* Set up a network like a reservoir
* Use the Poisson spike gen
* Stimulate a ...
Sensory Fusion with AER-based Sensors for Decision Making in Neuromorphic Robots
Single neuromorphic sensors, such as event based retinas or auditory sensors, have been used in various projects. Many spiking neural network models have been designed to compute data from these sensors. For example networks for written digit classification, spoken digit classification, object classification and detection, sound characterization and localization and ...
Transfer Deep RNN on Neuromorphic Hardware
It is a truth universally acknowledged (well, at least by another workgroup), that the neuromorphic community is in need of efficient algorithms for spatiotemporal processing, which can guide the behaviour of autonomous robots in complex environments, AND spiking recurrent neural networks (sRNN) are suitable candidates for such efficient processing, however ...
SCAMP-5 Vision Sensor
SCAMP is a vision sensor chip that includes a programmable processor in each pixel. The computations are implemented using mixed-mode circuits, with much of the datapath implemented in analog. The chip can implement low-level vision algorithms directly on the sensor plane, before outputting the computation results as gray-scale or binary ...
Context-dependent integration of information in rate-based recurrent neural network
The goal of this workgroup is to study rate-based recurrent neural network models of context-dependent integration of sensory evidence. Such models successfully describe some aspects of cortical computation (Mante et al., 2013). How context-dependent input selection and integration work can be understood from the linear system behaviour around fixed points ...
Learning intrinsic neural adaptation to enhance sRNN computational power
The neuromorphic community is in need of efficient algorithms for spatiotemporal processing, which can guide the behavior of autonomous robots in complex environments. Spiking recurrent neural networks (sRNN) are suitable candidates for such efficient processing, however, it remains a challenge to stabilize their dynamics. Recent theoretical work suggests that having ...
OpenNAS tutorial: a Neuromorphic Auditory Sensor HDL code generator
OpenNAS is an open source VHDL-based Neuromorphic Auditory Sensor (NAS) code generator capable of automatically generate the necessary files to create a NAS-VHDL project for FPGA. OpenNAS guides designers with a friendly interface that allows configuring the NAS using a five-steps wizard for later code generation. It includes several audio ...
Discussion Groups
Reduced-precision computation, Stochastic Rounding
There is increasing interest in saving on energy usage and memory bandwidth/footprint in more conventional machine learning implementations by reducing the size and complexity of the arithmetic types used, with a number of successful implementations. We aim to discuss how far this can go in terms of the precision ...
Frameworks for computing with time in spiking neural networks
Spikes are events with an associated time. This time can be used as a variable to encode an process information, and many neuroscientific studies have found such temporal codes in living organisms (barn owl, sand scorpion...) . However, this type of computation has not yet transitioned to artificial systems, with only ...
Unsupervised learning in neuromorphic computing: self-organizing maps vs. spiking neural networks
During the last years, Deep Neural Networks (DNNs) have reached the highest performances in image classification. Nevertheless, such a success is mostly based on supervised and off-line learning: they require thus huge labeled datasets for learning, and once it is done, they cannot adapt to any change in the data ...
Spiking-based Radar Processing for Automotive Applications
A self-driving car relies on the data fusion from three different sensors (Radar, camera and LiDAR) to explore the environment. Traditionally, the LiDAR is employed to map the surroundings, the camera to establish the route and classify the objects, and the radar is in charge of collision avoidance and automatic ...
Analog vs Digital debate
We will (yet again) discuss the benefits of analog vs digital circuit design. This year there will be a new side to this discussion coming from the memristor research for "in-memory computing" vs "von Neumann" architectures
Growing a Boltzmann machine
Instead of using a fixed network structure, how could a Boltzmann machine grow by adding latent variables in response to the difference between "wake" statistics (where the observed variables are clamped by observation) and "sleep" statistics (where even the observed variables are sampled according to the network's energy levels)?
This ...
A Fourier approach to Recurrent Neural Networks
Some participants have expressed interest in my recent theoretical result that relates the structure of a network to the computations that it can perform. The focus of this group/lecture will be to explain how to write down these equations that link the network structure to machine learning performance. The ...
Noise, variability and memory formation in spiking neural networks
In order to understand the mechanism driving spiking neural networks (SNNs) to recognize signals on time series; we evolve very small spiking neural networks for recognizing signals in a particular order in presence of noise on membrane potential and variation on silent interval between signals. For recognition of 3 signals ...
Common ground for memristive devices and neuromorphic hardware
Meristive devices are promising elements for the realisation of neuromorphic hardware. However, they are still under investigation and mostly unknown outside the "memristor community". It is very important for research to be able to think both at device level and at system level. Indeed, there is an awareness gap between ...
Neuromorphic Computing Technology: Reflections and Roadmaping
In the Coordination and Support Action project NEUROTECH (https://neurotechai.eu) we aim to support and develop the neuromorphic-computing-technology community in Europe. As part of this support and development, first, we would like to help the community to create an “image”, determining what NCT is, what research fields it embraces ...
Message Passing
Maneesh will present some more details in message passing.
For those who are not familiar with it, message passing is a framework for understanding how local interactions can be used for many learning tasks.
Area vs. Decision & behaviour
Barry will provide an overview how lesions affect behavioral decision making
Recreational Groups
Yoga, Acroyoga, slackline & Meditation
Namaste,
Before or after intensive neuromorph sessions, let's try to find some balance (both internal and external).
Yoga in front of the sea, slackline in the pinewood next to the hotel, acroyoga on the beach.
Schedule to be announced !
Football
“What a great day for football, all we need is some green grass and a ball.” - Bill Shankly.
Sometimes, happiness is simple.
Whether you're a pro footballer or never played before, join us, as the more we are the more FUN we share! ^^
Climbing
The Capo Caccia rock does not only offer a beautiful sight, but also an amazing spot to do rope climbing.
So please bring your gear (shoes, harness, rope & other material), so we can enjoy some hours climbing right next to the ocean.
We'll be there on Sunday the whole ...
Game of Thrones cinema!
A bunch of us have not watched yet this week's episode of GoT (S0802), and we are very anxious about it so let's do it here!
We will set up a "cinema" Friday at 18.30 at the big lecture room, grab some happy hour beers and watch it :) Feel ...
Bicycle Tour to Capocaccia
Please register if you need a bicycle on Sunday, so we can tell the hotel in advance.
Details follow.
PIZZA
We will go to a nice pizza place this Thursday 2nd of May
Champions League: Barça - Liverpool
A beautiful game of football!
Swing&Jazz dance lesson
Basics of Lindy-Hop dance for couples (man-woman, man-man, woman-woman, animal-man, ...whatever).
Jazz & Swing music. Super basic lesson where everybody can join.
Given that the football match is going to finish at 23:00 , it might be better if we anticipate the lesson at 19:00 (40/45 minutes before ...