wiki:2015/Workgroups

Work-Groups, Discussion-Groups, Recreational Groups

Use these pages to define or subscribe to work-groups, discussion-groups and recreational groups with corresponding mailing lists.

Work-groups focus on practical/hands-on projects while discussion-groups are meant as small-group brainstorming sessions on the topics proposed, that can take place in the evenings or after dinner.

The workgroup mailing lists are handy for organizing meetings, working sessions, discussion sessions, while at the workshop.

  • To subscribe to an existing work group click on the relevant Subscribe link (you need to be logged-in to see it).
  • To propose new workgroups and create your own mailing list, log-in and edit this page (the edit button is at the bottom of the page). Use one of the existing workgroups as a template, add a proper description, and write your name as the workgroup leader.

Workgroup leadersmust also subscribe as members to be included in the mailing list!

Work group / Discussion group Leaders

Work group / Discussion group leaders should coordinate the work group / discussion group activities, define and set the meeting times,

and most importantly, document the workgroup activities on the corresponding workgroup web-page. If you're a workgroup/discussiongroup leader and your group page does not exist yet, create it (by simply clicking on the workgroup title) ASAP and describe in more detail the planned workgroup activities (e.g. how long it will last, what is expected to happen, what material is required, etc.)

Workgroup Schedule

Workshop scheduleWorkgroup schedule key


Work Groups

Neuromorphic Sensors and Processing

Members: Bragi Lovetrue, Fabien Alibart, Amir Reza, Antonio Ríos, Aristeidis Tsitiridis, Christopher Bennett, Christoph Posch, Damir Vodenicarevic, Dan Neil, Nasim Farahini, Francesco Galluppi, Jorg Conradt, Julien Martel, Jordi-Ysard Puigbó Llobet, Leon Bonde Larsen, Lukas Everding, Marc Osswald, Marcel Stimberg, Matthew Tata, Orhan Celiker, Ole Richter, Paul B Isaac's, Michael Pfeiffer, Yulia Sandamirskaya, Scott Stone, Shih-Chii Liu, Sim Bamford, Siohoi Ieng, Subhrajit Roy, Tobi Delbruck, Gianvito Urgese, Viviane Ghaderi, Wenjia Meng

Coordinators: Tobi Delbruck, Shih-Chii Liu, Ryad Benosman, Marc Osswald, Bernabe Linares-Barranco, 'Alejandro Linares-Barranco' others TBD

This workgroup will coordinate activities in processing visual and auditory neuromorphic sensor outputs.

Real-time learning of behavior in a hybrid analog-digital system

Members: Mostafa Rahimi Azghadi, Borys Wrobel, Damir Vodenicarevic, Dan Neil, Dora Sumislawska, Fredrik Sandin, Gengting Liu, Giacomo Indiveri, Marc Osswald, Naous Rawan, Nathan Scott, Ning Qiao, Orhan Celiker, Ole Richter, Paolo Motto Ros, Paul B Isaac's, Michael Pfeiffer, Philipp Weidel, Qian Liu, Daniel Renz, Richard George, Yulia Sandamirskaya, Steve Nease

Leader: Jonathan Binas

Our latest neuromorphic hardware setup consists of an analog neuron array with 128k synapses that is directly interfaced with a number of digital general purpose processors. Combining the best of both worlds, this hybrid analog-digital configuration opens up a variety of new possibilities, from pre-processing of sensory signals on the neural hardware to the implementation of custom learning algorithms in the digital hardware. In this workgroup, we will attach our system to a robotic platform and implement networks and algorithms that allow it to learn basic control and/or navigation operations in an unsupervised or reinforcement framework.

Flying neurons

Members: Bragi Lovetrue, Christoph Posch, Dan Neil, Fabio Stefanini, Nasim Farahini, Federico Corradi, Francesco Galluppi, Fredrik Sandin, Gabriela Michel, Gengting Liu, Jonathan Binas, Jorg Conradt, Julien Martel, Lukas Everding, Marc Osswald, Nicolai Waniek, Orhan Celiker, Ole Richter, Paolo Motto Ros, Paul B Isaac's, Patrick Camilleri, Michael Pfeiffer, Daniel Renz, Ryad Benosman, Yulia Sandamirskaya, Shih-Chii Liu, Steve Nease, Sergio Solinas, Tobi Delbruck, Wenjia Meng

Leader: Marc Osswald

In recent years quadrotors evolved to a popular UAV platform that more and more became reliable and affordable such that nowadays they are used everywhere. While flight physics and motor control are well understood, vision-based control and autonomy are still open challenges. Algorithms such as SLAM and visual odometry have recently been well studied but their implementation on mobile platforms where computational resources and power are limited remains difficult. These machine vision algorithms are based on frame-based vision sensors producing redundant data which upon processing leads to a waste of power or sacrifice in latency. In contrast, event-based systems are data-driven, power efficient and they have low latency which makes them highly suitable for mobile platforms. In this workgroup, we will build a quadrotor from scratch and equip it with the latest event-based sensor (DAVIS) and our neuromorphic online-learning multi-neuron array (ROLLS). The ultimate goal is to develop a network that learns to perform attitude control and flight stabilization and implement it on the neuromorphic processor. The quadrotor will be euquipped with a variety of sensors such as IMU and optical flow which can be either used as teaching signals or further sensory inputs to the network. Within the scope of this workgroup it will also be possible to record some data from indoor flights and to develop visual altitude control, visual odometry and SLAM (directly from events or from event-based low-level features such as edges or corners) and sense & avoid algorithms that then can be tested on the quadrotor.

Gambling neurons

Members: Fabio Stefanini, Nasim Farahini, Gabriela Michel, Giacomo Indiveri, Jonathan Binas, Marc Osswald, Michael Pfeiffer, Daniel Renz, Ryad Benosman

Leader: Marc Osswald

Can you predict roulette spins and beat the house edge? Various videos on Youtube claim that their methods produce predicted earnings instead of the expected loss of about 2.7 percent. The most successful among a variety of approaches involve cameras that observe and record the initial conditions. It has been shown that this perfectly works in a laboratory with high-tech equipment but it is much harder in a real casino environment where light conditions are bad and the system has to make predictions in real-time. While normal cameras fail in such a scenario the DVS seems as if to the manner born. Let's develop an event-based algorithm that predicts roulette spins and beats the house edge. If we are successful we can test it on Saturday night in one of Sardinia’s casinos.

For developing and testing, it would be very useful if somebody could bring a casino wheel (preferably an Italian one, similar to the ones they use in Italian casinos).

Analog synaptic circuit design competition

Members: Mostafa Rahimi Azghadi, Shuzhan Bi, Damir Vodenicarevic, Dora Sumislawska, Nasim Farahini, Federico Corradi, Gengting Liu, Giacomo Indiveri, Julien Martel, Jordi-Ysard Puigbó Llobet, Naous Rawan, Ning Qiao, Ole Richter, Paolo Motto Ros, Paul B Isaac's, Richard George, Yulia Sandamirskaya, Sim Bamford, Steve Nease, Suraj Honnuraiah, Sergio Solinas, Viviane Ghaderi

Leader: Dora Sumislawska

We are going to learn to design analog weight-update synaptic circuits for implementing spike-based learning in neural processing systems. One of our goals will be to design synaptic update circuits independent of the absolute value of the weight voltage. We will then investigate the implementation of different types of learning rules that have different synaptic weight update strategies (including weight-dependent ones). I will present preliminary attempts with current-mode analog VLSI designs, discussing over their advantages and drawbacks, and then open the discussion for new exciting ideas. The "best" synapse circuit will win a prize (to be determined) at the end of the workshop. Looking forward to brainstorming over novel ideas!

Finding spatiotemporal correlations in spiketrains with FPGA and neuromorphic hardware

Members: Bragi Lovetrue, Mostafa Rahimi Azghadi, Adrian Whatley, Shuzhan Bi, Borys Wrobel, Dan Neil, Julien Martel, Leon Bonde Larsen, Lukas Everding, Matthew Tata, Naous Rawan, Nathan Scott, Nicolai Waniek, Orhan Celiker, Ole Richter, Paolo Motto Ros, Paul B Isaac's, Richard George, Scott Stone, Steve Nease, Taras Iakymchuk, Victor Minces, Wenjia Meng

Leader: Richard George, Julien Martel

As a toy-problem, we will make use of two dynamic vision sensors as our event-sources and compute both the spatial and temporal correlation of events between the two sensors. Other event-sources that follow some AER-scheme are welcome as well, to further add sources to this project. Our processing substrate will be INIs backbone system for interfacing neuromorphic hardware, featuring an embedded processor-on-FPGA.

Besides its use in machine vision, having a spatio-temporal correlator that can handle large numbers of event-sources has another use that may be of particular interest to neurobiologists. Analysing connectivity in a sample of neural tissue for reconstructing network topology is a challenging problem. The state of the art approach here relies on optical methods like viral transfection-studies and calcium imaging. Although providing great amounts of information, a human operator is needed for classifying synaptic connections off-line. However, recent advances in microelectrode-technologies provide extracellular recordings in high spacial resolution. Like the DVS, some of these Sensor-Arrays encode recorded spikes following the AER protocol which makes them ideal inputs for neuromorphic systems. Lets develop an architecture that is capable of inferring causal relationships between observed events and constructs a connectivity matrix from the gained information.

Universal AER Communications over Ethernet for Multi-System Applications

Members: Amir Reza, Adrian Whatley, Shuzhan Bi, Claudio Luck, Dan Neil, Fabio Stefanini, Jonathan Binas, Nathan Scott, Nicolai Waniek, Orhan Celiker, Paolo Motto Ros, Johannes Partzsch, Paul B Isaac's, Alexander Rast, Richard George, Taras Iakymchuk, Gianvito Urgese

Leader: Alexander Rast

In past workshops, we have explored the idea of developing standards for AER communication over Ethernet, so that multiple hardware systems, as well as ordinary computers (with a network connection) can intercommunicate in real-time to create heterogeneous simulations. In this workgroup we will explore linking systems in practice with a working draft version of a protocol. The protocol we propose can be thought of as a "trial balloon" for a formal standard. We will have the SpiNNaker neuromorphic platform available with protocol support as well as a "device simulator" that can either emulate an external device supporting a standard or act as a host-based intermediary for non-configurable external hardware. We will have a development library also available for those interested in developing protocol support for their own external hardware devices (with an Ethernet port) By the end of the workshop, we hope to have at least one multi-system simulation running between SpiNNaker, some other neuromorphic device (which could be either an I/O device like a retina or another general-purpose neuromorphic chip), and one (or more) laptop PC(s). Interest permitting, we may also hold a discussion group on standards for AER communications, feeding in any insights from ongoing work in this workgroup

Hardware Multiscale Heterogeneous Neural Model Simulation

Members: Bragi Lovetrue, Borys Wrobel, Dan Neil, Florian Schuler, Gengting Liu, Michael Hopkins, Marcel Stimberg, Naous Rawan, Nathan Scott, Orhan Celiker, Johannes Partzsch, Paul B Isaac's, Patrick Camilleri, Qian Liu, Alexander Rast, Subhrajit Roy, Suraj Honnuraiah, Gianvito Urgese

Leader: Alexander Rast

What neuron model is best for reasonably large-scale simulations? In this workgroup we argue: "depends upon the problem!" Both considerations of what the hardware platform can support and what neural dynamics are required for the network model being simulated need to be considered. In fact, there's no reason to believe even within the same simulation, the neural model should be uniform. We propose the following: bring your (favourite) neural model, and we'll have a go at implementing it and integrating it in a hardware simulation. In the group, we'll compare and benchmark neural models in various real behavioural tasks and perhaps? even come up with some conclusions as to what models are good where. We will have the SpiNNaker hardware platform available to run simulations on, but if you've got your own hardware, bring it too! We'll examine ways to integrate it into multi-system simulations as well as independent benchmarking. So if there's a neuron model you've long wanted to see run on hardware, or if you just want to see how your model or hardware might fare against others, this is your chance.

Ensemble methods for classification and regression: making machine learning more neuromorphic

Members: Fabien Alibart, Christopher Bennett, Shuzhan Bi, Damir Vodenicarevic, Dan Neil, Fabio Stefanini, Federico Corradi, Giacomo Indiveri, Jordi-Ysard Puigbó Llobet, Leon Bonde Larsen, Lukas Everding, Naous Rawan, Nathan Scott, Nicolai Waniek, Orhan Celiker, Ole Richter, Paolo Motto Ros, Paul B Isaac's, Qian Liu, Daniel Renz, Selina La Barbera, Shih-Chii Liu, Steve Nease, Subhrajit Roy, Gianvito Urgese, Victor Minces, Wenjia Meng

Leader: Fabio Stefanini

Sophisticated machine learning algorithms are successfully employed to solve complex tasks such as image and speech recognition. However, the implementation of such algorithms requires powerful machines. Can we even think to achieve interesting performance by using much simpler algorithms, perhaps old-fashioned, but certainly more "neuromorphic"? Under which figure of merit could we say that these algorithms perform significantly better? Building upon last year's successes, the aim of this workgroup is to test realistic models of synaptic plasticity on typical machine learning benchmarks using software simulations. In particular, the work will focus on ensemble methods (also referred to as committees of weak classifiers) as an elegant tool to exploit variability on the single elements, an issue which we know is at the heart of electronic hardware implementations. We provide a newly developed software framework that allows the integration of these type of models on a classification pipeline in order to apply the models to freely available datasets. This project will require only basic Python coding skills, promised. A dataset classification competition will be considered, with prices to be assigned. Possible side projects: developing a simple chess-playing engine (see the chess recreational group), collecting data for a new dataset (see the neuromorphic benchmarks discussion group), beating Marc in predicting the outcome of the roulette given the initial position of the ball.

Learning to decode behaving rat's cortex with neuromorphic chips

Members: Bragi Lovetrue, Alessandro Vato, Shuzhan Bi, Borys Wrobel, Fabio Boi, Fabio Stefanini, Federico Corradi, Florian Schuler, Gabriela Michel, Giacomo Indiveri, Houman Safaai, Jordi-Ysard Puigbó Llobet, Matthew Tata, Naous Rawan, Nicolai Waniek, Orhan Celiker, Paolo Motto Ros, Paul B Isaac's, Qian Liu, Daniel Renz, Richard George, Yulia Sandamirskaya, Timoleon Moraitis, Gianvito Urgese, Victor Minces, Wenjia Meng, Mehmet Fatih Yanik

Leader: Vito De Feo

Neuromorphic hardware has been proposed as an ideal candidate for decoding cortical activity in real-time. However, this task is very different from a typical pattern classification task as often seen in machine learning. For example, the data is often a low-dimensional representation of what's really going on in the brain (i.e., only a few neurons are recorded), the signal-to-noise ratio is often very high (i.e., high inter-trial variability) and signals are not continuous in time rather they are spike-encoded in a rather unclear way. During the last few years, we at the IIT in Geneva (IT) also collected evidence that the decoding depends on the "state" of the brain, a sort of slow component of the dynamics of the network's underlying dynamic system. We tackled the challenge of decoding rat's cortex with success and we are now interested into take advantage of the properties of neuromorphic chips to achieve the task. This workgroup will thus try to decode rat's motor cortex in real-time using neuromorphic chips with online learning capabilities. To this purpose, we will bring together people from our lab Geneva's IIT and Zurich's INI and two hardware setups ready to fire!

Memristive nanodevices synapses: let’s build an intelligent memory!

Members: Bragi Lovetrue, Mostafa Rahimi Azghadi, Fabien Alibart, Manuel, Christopher Bennett, Shuzhan Bi, Damir Vodenicarevic, Fabio Stefanini, Julien Martel, Naous Rawan, Orhan Celiker, Paolo Motto Ros, Paul B Isaac's, Michael Pfeiffer, Qian Liu, Daniel Renz, Yulia Sandamirskaya, Selina La Barbera, Shih-Chii Liu, Sim Bamford, Sergio Solinas

Leader: Selina La Barbera

Memristive nanodevices represent a powerful candidate for future neuromorphic computing systems and a potential tool to investigate the human brain memory in terms of plasticity and learning. We have recently demonstrated that fundamental processes observed in biological synapses can be successfully reproduced and controlled through emerging memristive devices operation. If various plasticity mechanisms have been already demonstrated in memristive systems, we showed that more synaptic features can be embedded in a single memory component by exploiting the basic physics of filamentary resistive switching. Currently we are interested in exploiting the intrinsic, non-linear and dynamic device volatility in order to explore several learning strategies to model different circuit topologies and level of processing by large scale memristive devices circuit modeling.

Thus, the aim of this workgroup is devoted to exploit different memristive nano-devices technologies to “benchmark” them on a particular neuromorphic task by highlighting a rich panel of features in terms of functionalities and performances required for the material implementation of bio-inspired circuits.

This emerging research direction requires an interdisciplinary approach, from biology, neurosciences, device physics and computer science thus, let's come to play and to learn how artificial synapses learn!

Memristive synapses for spike timing- and rate-based synaptic plasticity!

Members: Bragi Lovetrue, Mostafa Rahimi Azghadi, Fabien Alibart, Manuel, Christopher Bennett, Jean-Pascal Pfister, Naous Rawan, Paolo Motto Ros, Johannes Partzsch, Paul B Isaac's, Yulia Sandamirskaya, Selina La Barbera, Steve Nease, Subhrajit Roy, Suraj Honnuraiah

Leader: Mostafa Rahimi Azghadi

In this workgroup we will discuss and review previous implementations of memristive synapses. We will also talk about the applications and challenges in the design of neuromorphic memristive systems. As a project, we will collaborate on the design of a new memristive synapse, which aims at improving the synaptic plasticity (learning) capability of previously proposed devices. In particular, we will work toward implementing a memristive synapse, which can replicate a number of interesting features of biological synapses, including the spike-rate based plasticity. In addition, we test the synapse against implementing higher order STDP rules, such as triplet-based STDP.

Designing spiking recurrent neural networks performing enhanced signal restoration

Members: Bragi Lovetrue, Christopher Bennett, Shuzhan Bi, Borys Wrobel, Dan Neil, Federico Corradi, Florian Schuler, Giacomo Indiveri, Jordi-Ysard Puigbó Llobet, Leon Bonde Larsen, Marcel Stimberg, Nathan Scott, Orhan Celiker, Paul B Isaac's, Daniel Renz, Saray Soldado, Steve Nease, Subhrajit Roy, Suraj Honnuraiah, Taras Iakymchuk

Leader: Florian Schuler

In this workgroup we will design spiking recurrent neural networks in software that perform a form of signal restoration that respects the shape of the Tuning curve. In biological experiments, the Tuning curve of orientation-selective neurons in visual cortex has been found to be Gaussian-shaped. This shape of the Tuning curve is thought to be a product of the recurrent connections rather than of the noisy feedforward input. Hence, the networks we will design converge to a Gaussian-shaped activity profile from a noisy input.
We will hold this workgroup as a competition. Participants can work with the neuronal model that they are familiar with. They can use all sorts of learning procedures to increase performance of their network. Performance of the activity profile is measured in terms of least squares to a Gaussian fit of the non-Gaussian input. Click on the title of the workgroup so see more information.

Self-construction of the brain using the simulation platform Cx3D (Cortex 3D)

Members: Bragi Lovetrue, Adrian Whatley, Christopher Bennett, Damir Vodenicarevic, Dan Neil, Florian Schuler, Gabriela Michel, James Knight, Julien Martel, Jordi-Ysard Puigbó Llobet, Leon Bonde Larsen, Orhan Celiker, Paul B Isaac's, Michael Pfeiffer, Daniel Renz, Rodney Douglas, Saray Soldado, Suraj Honnuraiah, Sergio Solinas

Leader: Gabriela Michel

One of the most fascinating aspects in neuroscience is how individual neuronal elements interact to assemble the brain. A few precursor cells with the same encoded genetic information replicate and generate specialized cell types yet combine to form a complex but coherent processing system. This stands in stark contrast with conventional technology, designs orchestrated by engineers that have a clear, complete, and detailed map of the system. We have studied self-construction using the java-based simulation software Cx3D, which allows an accurate physical representation of cortical development in a 3D environment.
The main purpose of this workgroup is to give a general introduction to the Cx3D simulation platform. We will first start with an explanation of the basic functions and features of Cx3D. Afterwards we will work with some basic tutorials. Finally, the workgroup attendants will have to solve a basic task based on axonal growth. This is very relevant for cortical self-assembly since it is unknown how a cell finds its synaptic target across areas and we can start studying inter-areal connectivity with this software. It is required that the attendants have Eclipse and Cx3D installed.
Important links: https://eclipse.org/downloads/ https://www.ini.uzh.ch/~amw/seco/cx3d/
If you have any question, do not hesitate to contact me.
Successful participation in this workgroup will be rewarded with Margaritas ;).

Locust-Inspired Collision Detection on SpiNNaker

Members: Bragi Lovetrue, Amir Reza, Antonio Ríos, Shuzhan Bi, Borys Wrobel, Dan Neil, Gengting Liu, James Knight, Marcel Stimberg, Matthew Tata, Qian Liu

Leader: Patrick Camilleri

In this workgroup, we are going to implement collision avoidance for a pushbot robot (http://www.inilabs.com/products/robots/pushbot) using the model of the locust LGMD (Lobula Giant Movement Detector) neuron. A previous implementation using a frame based camera (http://doi.org/10.1016/S0921-8890%2899%2900063-9), IR sensors and several desktop PCs, but we think this is an ideal model for a DVS sensor (silicon retina) and neuromorphic hardware - specifically SpiNNaker.

Important links:
Collision avoidance using a model of the locust LGMD neuron - http://www.ini.uzh.ch/admin/extras/doc_get.php?id=41818
A Collision Detection System for a Mobile Robot Inspired by the Locust Visual System - http://www2.imse-cnm.csic.es/locust/publications/conferences/Yue-Rind-CRA05-BCN.pdf

Real-time Maze Playing and Learning with Spiking Hardware

Members: Bragi Lovetrue, Shuzhan Bi, Dan Neil, Jordi-Ysard Puigbó Llobet, Lukas Everding, Orhan Celiker, Michael Pfeiffer, Qian Liu, Daniel Renz, Scott Stone, Shih-Chii Liu, Subhrajit Roy, Mehmet Fatih Yanik

Leader: Dan Neil

In the Labyrinth maze game you navigate a rolling marble across a maze with pan and tilt. We are going to build a learning agent that can do this. This workgroup will look into real-time control of the physical marble to learn to effectively solve the maze, avoiding the holes and walls. Deep Q-Networks, introduce by Google DeepMind? and capable of learning to play a large number of Atari games often to a level exceeding that of a human player, have demonstrated that it is possible to effectively learn a control policy from sparse, infrequent rewards using only visual inputs.

This group will use deep networks in conjunction with the extremely rapid sensing provided by the DAVIS sensor to learn to solve the labyrinth game, panning and tilting the board according to the network to navigate the marble to the goal.

Workshop web applications and IT infrastructure

Members: Adrian Whatley, Claudio Luck, Giacomo Indiveri, Kathrin Aguilar Ruiz-Hofacker, Rodney Douglas

Leader: Claudio Luck

All about the current and in-development web applications, and the IT infrastructure at the workshop.

Discussion Groups

Benchmarking Neuromorphic Systems

Members: Bragi Lovetrue, Mostafa Rahimi Azghadi, Fabien Alibart, Adrian Whatley, Christopher Bennett, Dan Neil, Fabio Stefanini, Nasim Farahini, Florian Schuler, Gabriela Michel, Giacomo Indiveri, Jorg Conradt, Julien Martel, Jordi-Ysard Puigbó Llobet, Marc Osswald, Marcel Stimberg, Nathan Scott, Nicolai Waniek, Orhan Celiker, Paolo Motto Ros, Johannes Partzsch, Paul B Isaac's, Michael Pfeiffer, Alexander Rast, Yulia Sandamirskaya, Selina La Barbera, Shih-Chii Liu, Sim Bamford, Siohoi Ieng, Steve Nease, Subhrajit Roy, Taras Iakymchuk, Gianvito Urgese, Wenjia Meng

Leader: Alexander Rast

Quantitative benchmarking of neuromorphic systems is becoming an increasingly important matter, especially since systems are now reaching the sorts of speeds and scales where understanding a platform's strengths and weaknesses may be critical to committing to running a simulation on it. How should neuromorphic systems be benchmarked? On the one hand, there is a desire to demonstrate what up until now have been the claimed advantages of hardware: lower power, higher speed, larger model size, and (in some cases) more task-appropriate computational capabilities. On the other hand, it is perhaps trying to compare apples and oranges to benchmark systems against each other, much less against conventional compute platforms. Most neuromorphic systems have been devised for very specific purposes and any benchmark must at least address this issue. We will discuss what potential benchmark tests or suites could be devised, relevant performance metrics, and possibly also look into comparisons against biology - how do we benchmark neuromorphic systems against the brain? Difficult questions yes, but now is the time to start framing the answers.

The changes of firing behavior of single neurons during postnatal development

Members: Bragi Lovetrue, Borys Wrobel, Dan Neil, Eleni Vasilaki, Florian Schuler, Gabriela Michel, Jonathan Binas, Jordi-Ysard Puigbó Llobet, Nicolai Waniek, Paul B Isaac's, Michael Pfeiffer, Yulia Sandamirskaya, Saray Soldado, Suraj Honnuraiah, Sergio Solinas, Gianvito Urgese

Leader: Florian Schuler

Early activity patterns observed shortly after birth are different from the activity present in the mature mammal. A lot of changes happen during postnatal development such as the transition of GABAergic signaling from excitation to inhibition. Another example is the Calcium plateau in the apical dendrite of neocortical pyramidal cells. Furthermore, the firing properties of cerebellar Purkinje neurons change during postnatal mouse development. In the discussion group we want to exchange our opinions about this.

Vision Benchmarks

Members: Bragi Lovetrue, Amir Reza, Dan Neil, Fabio Stefanini, Nasim Farahini, Jonathan Binas, Julien Martel, Lukas Everding, Marc Osswald, Nicolai Waniek, Orhan Celiker, Michael Pfeiffer, Qian Liu, Viviane Ghaderi, Mehmet Fatih Yanik

Leader: Qian Liu

Spiking Neural Networks inspired by our understanding of observed biological structure and functions have been successfully applied to visual recognition/classification tasks. A new series of vision benchmarks for spike-based neural processing are now needed to quantitatively measure progress within this rapidly advancing field. A large dataset of spike-based visual stimuli is needed to provide a baseline for comparisons. Furthermore a complementary evaluation methodology is also crucial to assess the accuracy and efficiency of an algorithm. We can summarise existing models, propose the dataset and discuss about the evaluation methodology in order to promote meaningful comparison in the field of neural computation. Likely, we will put up a working group to classify digits with spiking MNIST database.

PyNCS development

Members: Adrian Whatley, Christopher Bennett, Dan Neil, Fabio Stefanini, Nasim Farahini, Federico Corradi, Fredrik Sandin, Giacomo Indiveri, Leon Bonde Larsen, Marcel Stimberg, Matthew Tata, Nathan Scott, Ning Qiao, Orhan Celiker, Ole Richter, Paolo Motto Ros, Sergio Solinas, Gianvito Urgese

Leader: Fabio Stefanini

PyNCS is a set of Python modules that can be used to control spike-based neuromorphic systems. The workshop is a good opportunity for the main developers and contributors to the project to sit together and finalize development of old issues, plan future enhancements, discuss roadmaps. The discussions will hopefully get very technical but newcomers are encouraged to step-in, especially Python experts and hardware designers who wish to see if and how they can benefit from PyNCS. Depending on how things evolve we might turn this group into a workgroup and get messy with coding.

Neuromorphic Homeostatic Plasticity

Members: Bragi Lovetrue, Borys Wrobel, Chiara Bartolozzi, Dan Neil, Nasim Farahini, Florian Schuler, Giacomo Indiveri, James Knight, Jordi-Ysard Puigbó Llobet, Nicolai Waniek, Ole Richter, Johannes Partzsch, Paul B Isaac's, Sim Bamford, Steve Nease, Suraj Honnuraiah, Wenjia Meng

Leader: Steve Nease

Homeostatic plasticity is a system used by the brain to stabilize its population dynamics. This stabilization is accomplished by using feedback control to set neurons' average firing rates over a long timescale. A number of different mechanisms can implement this control: synaptic scaling multiplies the strengths of a neuron's synapses by an activity-dependent factor; intrinsic plasticity modifies the channel properties of the neuron itself; and metaplasticity modifies parameters of the learning rule. Many theoretical and simulation-based studies have shown that unstable population dynamics can result in systems without homeostatic control.

Homeostatic plasticity yields interesting design challenges for the neuromorphic engineer. For instance, the time constants over which homeostasis works are typically quite long, necessitating the design of circuits with low leakage current. This group will review previous neuromorphic circuit designs for homeostatic plasticity and discuss potential new directions based on recent results in computational neuroscience.

Non-Turing Computational Model for Neuromorphic Engineering

Members: Bragi Lovetrue, Christopher Bennett, Borys Wrobel, Matthew Cook, Dan Neil, Nasim Farahini, Giacomo Indiveri, Jordi-Ysard Puigbó Llobet, Naous Rawan, Nicolai Waniek, Paolo Motto Ros, Paul B Isaac's, Saray Soldado, Sim Bamford

Leader: Bragi Lovetrue

A computer is a machine that represents and processes information. Ever since the advent of modern computers, the goal of brain science is to understand how the brain works as a computer and the goal of artificial intelligence is to build brain-like computers. Deep learning and neuromorphic engineering are prime examples of the cross-fertilization between the goals of brain science and artificial intelligence. The founding fathers of modern computers, however, differ in their views of whether brains are essentially modern computers: Alan Turing insisted that brains and modern computers share the same computational model that bears his name, whereas John von Neumann believed that brains are fundamentally different from the architecture of modern computers that also bears his name.

However, it can’t be the case that both Turing and von Neumann are correct for one simple reason: a computer architecture is merely a physical implementation scheme of a computational model which is actually a mathematical construction. As a matter of fact, the von Neumann architecture can’t be fundamentally changed without a fundamental change to the underlying Turing computational model, and vice versa. What would be possible fundamental changes to the Turing Machine? What are the implications of Non-Turing computational models for neuromorphic engineering?

The Brian simulator for spiking neural networks

Members: Bragi Lovetrue, Amir Reza, Antonio Ríos, Christopher Bennett, Borys Wrobel, Damir Vodenicarevic, Nasim Farahini, Florian Schuler, Jordi-Ysard Puigbó Llobet, Marcel Stimberg, Orhan Celiker, Ole Richter, Paolo Motto Ros, Paul B Isaac's, Richard George, Saray Soldado, Gianvito Urgese

Leader: Marcel Stimberg

The Brian simulator is a simulator for spiking neural networks, written in Python, that focuses on ease-of-use and flexibility (http://www.briansimulator.org). It allows to write new neuronal and synaptic models by writing down their high-level mathematical description (using physical units), without having to fiddle around with any low-level code. Behind the scenes, Brian will use a code generation framework to generate and compile executable code. We built the current development version of Brian (Brian 2) with the goal that the code generation framework is fully extensible, so users should be able to adapt it to their specific target device with reasonable effort. Let's have a first meeting where I'll give an overview over Brian 2 -- depending on the interests of the participants we could then either discuss how to simulate your favourite model with Brian or have a more detailed look at the code generation and how to extend it for specific use cases (in which case this would rather transform into a workgroup, I guess).

The ultimate serial AER

Members: Bragi Lovetrue, Amir Reza, Adrian Whatley, Chiara Bartolozzi, Matthew Tata, Ning Qiao, Paolo Motto Ros, Paul B Isaac's, Patrick Camilleri, Alexander Rast, Richard George, Sim Bamford, Taras Iakymchuk

Leader: Chiara Bartolozzi

Neuromorphic sensors and computational devices are now used to build complex systems, interconnecting a number of different AER devices. A necessary step for the miniaturisation, integration and deployment of this technology in artificial devices such as robots, prosthetics, etc. is the implementation of a fast AER serial protocol. Few implementations are now being used and tested. This group will continue the discussion about typical use-cases, specs and desiderata of past years, and summarise the pros and cons of current implementations, aiming to defining a standard AER protocol that possibly captures all the needs of the neuromorphic community.

Spiking implementations of Deep Neural Networks

Members: Bragi Lovetrue, Abu Sebastian, Fabien Alibart, Amir Reza, Manuel, Borys Wrobel, Dan Neil, Jordi-Ysard Puigbó Llobet, Lukas Everding, Marcel Stimberg, Naous Rawan, Nathan Scott, Nicolai Waniek, Ole Richter, Paolo Motto Ros, Michael Pfeiffer, Daniel Renz, Taras Iakymchuk, Sergio Solinas, Gianvito Urgese, Mehmet Fatih Yanik

Leader: Michael Pfeiffer

Deep Learning is the state-of-the-art for visual and auditory recognition in machine learning. They are quite power-hungy when they need to be trained, and inference needs to be performed on them, so we have recently looked into spiking implementations of various types of networks. This improves latency and power consumption, which is potentially very important for real-world applications, and of course should be a good model for hardware implementations. Here we will discuss the techniques needed for spiking deep networks, providing both a tutorial and hopefully a lot of discussions to answer all your questions.

Looking Forward: The Next Big Thing

Members: Bragi Lovetrue, Abu Sebastian, Adrian Whatley, Borys Wrobel, Dan Neil, Gabriela Michel, Jorg Conradt, Jordi-Ysard Puigbó Llobet, Marc Osswald, Naous Rawan, Nicolai Waniek, Johannes Partzsch, Paul B Isaac's, Michael Pfeiffer, Qian Liu, Rodney Douglas, Saray Soldado, Steve Furber, Sim Bamford, Steve Nease, Subhrajit Roy, Taras Iakymchuk, Sergio Solinas, Viviane Ghaderi

Leader: Dan Neil, Nicolai Waniek, Viviane Ghaderi

Many neuromorphic devices are now entering a level of maturity. Among the many advances, VLSI neurons are incredibly low-power and more consistent than ever, the DVS has become more feature-filled and reliable with every generation, the SpiNNaker chips enable more types of neural computation at a power and scale unattainable before, and we are standardizing interconnects among everything. So here's the question: what is next?

To prepare, think about the most interesting advances from the past couple of years in your specific subfield. We'll get together to discuss these advances, talk about why we are building things, and maybe influence our ideas about what we'd like to build and study together.

Memory devices technologies for neuromorphic computing

Members: Bragi Lovetrue, Mostafa Rahimi Azghadi, Abu Sebastian, Manuel, Jordi-Ysard Puigbó Llobet, Naous Rawan, Paolo Motto Ros, Paul B Isaac's, Selina La Barbera, Steve Nease

Leader: Fabien Alibart

It seems that we have in CapoCaccia? lots of devices experts with various memory technologies. This discussion group will be the opportunity to present how each technology is today exploited in neuromorphic computing (i.e. analog, stiochastic...). We could all give a short presentation and perspective for our own technology and discuss what are the pros and cons.

Five Neuromorphic Startups That Can't Fail

Members: Dan Neil, Marc Osswald, Ole Richter, Michael Pfeiffer

Leader: Matthew Tata

Neuromorphic systems are awesome, but what are they "awesomer" at than other stuff? This group will brainstorm ideas for applications of neuromorphic systems that would disrupt or outcompete incumbent technology.

ANNOUNCEMENT: We will meet in the main lecture room at 9pm Thursday evening.

Spatial Navigation in Rodents and Robots

Members: Bragi Lovetrue, Borys Wrobel, Dan Neil, Eleni Vasilaki, Fabio Stefanini, Gengting Liu, James Knight, Jorg Conradt, Jordi-Ysard Puigbó Llobet, Marcel Stimberg, Nicolai Waniek, Orhan Celiker, Ole Richter, Paul B Isaac's, Patrick Camilleri, Philipp Weidel, Daniel Renz, Yulia Sandamirskaya, Shih-Chii Liu, Sergio Solinas, Viviane Ghaderi

Leaders: Nicolai Waniek

How do rodents navigate? And can we apply these principles to robots? During this workgroup we will first look at the latest models for grid cells, place cells, and goal-directed navigation in rodents. We will have an in-depth discussion about the plausibility of these models, and about their stability with respect to real world inputs, for instance from a robotic platform. We will then proceed to frame the models in a neuromorphic context, especially with distributed processing. Finally, and given enough interest, we may use the models to drive small robots towards a goal location without prior knowledge of the environment.

Recreational Groups

T-Shirt Competition

Members: Giacomo Indiveri, Yulia Sandamirskaya, Selina La Barbera

Leader: Giacomo Indiveri

Following the tradition of the Telluride Neuromorphic Workshop, this year for the first time we are launching a "CapoCaccia? T-Shirt Competition"! The competition is open to all participants. The goal is to create a design for a t-shirt, using at most 4 colors that represents the CapoCaccia? workshop spirit. Newcomers can learn about the "workshop spirit" by studying the 2015 pages of this web-site, and (importantly) by browsing the past years pages (2014, 2013, 2012, etc.) T-shirt designs should be submitted by Sunday April 19. To submit your design, upload it as a *.png file to this discussion group page (click on the title of this discussion group to access the page).

The workshop organizing committee will evaluate all submissions and pick the best one. Winners of the competition will receive an extra t-shirt for free!

Soccer

Members: Antonio Ríos, Christoph Posch, Fabio Stefanini, Marc Osswald, Cristiano Ronaldo, Johannes Partzsch, Saray Soldado, Gianvito Urgese, Vito De Feo

Leader: Marc Osswald

Let's play soccer together on the Hotels soccer pitch just around the corner.

Running

Members: Mostafa Rahimi Azghadi, Christopher Bennett, Nasim Farahini, Giacomo Indiveri, James Knight, Jorg Conradt, Orhan Celiker, Paul B Isaac's, Philipp Weidel, Yulia Sandamirskaya, Selina La Barbera, Siohoi Ieng, Taras Iakymchuk, Viviane Ghaderi

Leader: Lorenz Muller

Let's go running in the morning to offset the buffet bias on our weight change during CCCNEW.

Tennis

Members: Christopher Bennett, Christoph Posch, Marc Osswald, Marcel Stimberg, Michael Pfeiffer, Shih-Chii Liu, Tobi Delbruck, Victor Minces

Leader: Tobi Delbruck

Sign up to be informed of tennis possibilities, and bring your shoes and perhaps racket along.

Chess

Members: Fabio Stefanini, Nasim Farahini, Federico Corradi, Jordi-Ysard Puigbó Llobet, Paolo Motto Ros, Sim Bamford

Leader: Fabio Stefanini

Whether you know how to play chess or not, just sit together and spend some time to exercise your decision making system with one of the most ancient and fascinating board games. Boards and tutorials for newbies are provided but if you have a board you may want to bring it by. We will also analyze the effects of (1) coca cola, (2) beer and (3) Sardinian mirto (4) water on the decision making process. A final tournament could be arranged depending on the number of participants with prices to be decided. Software chess solvers will be allowed and welcome. KungFu? chess is an option to be evaluated. (Though KungFu? software might be too tricky to implement.)

Chocolate

Members: Bragi Lovetrue, Adrian Whatley, Christopher Bennett, Borys Wrobel, Dan Neil, Eleni Vasilaki, Fabio Stefanini, Nasim Farahini, Fredrik Sandin, Giacomo Indiveri, Michael Hopkins, Jordi-Ysard Puigbó Llobet, Leon Bonde Larsen, Lukas Everding, Johannes Partzsch, Michael Pfeiffer, Alexander Rast, Yulia Sandamirskaya, Saray Soldado, Shih-Chii Liu, Subhrajit Roy, Victor Minces, Wenjia Meng

Leader: Alexander Rast

Returning annually - the CapoCaccia? Chocolate tasting. New format this year too - hopefully with some new participants! The last few years have seen a "competition"-style approach with various rounds of tastings. This year we introduce a "chocolate tasting workgroup". So, in the first week (useful for new participants) we will introduce concepts in chocolate tasting with a lot of practical examples, cover different origins and bean types, different styles of chocolate, different manufacturers. There will probably be a "basic" and "advanced" day on bar chocolate, plus chocolate confections - i.e. boxed chocolates like pralines, ganaches, etc. We finish the first week with an "is this fine" session - put your skills to work in identifying whether you think several unmarked samples are fine or mass-market chocolate. There will be enough content to excite both the first-time "newbie" and the veteran of many CC tastings. The second week will be evaluating and identifying fine chocolate - using your own brought in samples (as in previous years, bring in chocolates from whereever you are that you think are worthy of being considered "great"). The old competition isn't going away entirely - we'll revive that (in more compressed format) for both bars and confections (so bring some good ones!) but in addition there's a new element: "Guess-that-bar". During competition rounds participants will independently be asked to try to guess the origin and manufacturer of bar chocolates (possibly confections as well), and culminating in a "Grand Final" guessing competition where the participants will attempt to identify as many bars as possible. The winner will recieve unlimited first choice of all chocolate remaining from samples! (Plus possibly a few "bonus" bars from me). Others will of course be able to divide the remaining spoils amongst them at the end...

Climbing

Members: Christoph Posch, Dora Sumislawska, Nasim Farahini, Florian Schuler, Jonathan Binas, Jean-Pascal Pfister, Philipp Weidel, Alexander Rast, Daniel Renz, Sergio Solinas

Leader: Alexander Rast

Climbing is back this year with a new place to go as well as the old. We'll visit the well-known and extensive "Casarotto" area as always, on Capo Caccia itself, but following a successful recce last year I will also be taking people to the "Placche di Peyer" area which is within walking distance of the hotel. Casarotto offers spectacular climbing on overhanging limestone with grotesque Baroque rock formations - think big jugs and thuggish moves - in an impossibly scenic seaside location. Climbs from 5b to 8a (one line is more probably 8b/c) Placche di Peyer features shorter, more balancey routes on more compact, vertical rock. Still limestone, but fewer huge holds, more strange body postures on rounded surfaces. Climbs from 5b to 7c. They're short enough that they could be "highball" boulder problems as well, so if you don't feel like getting on the rope, bring your bouldering mat. I'll bring rope, quickdraws, belay devices. Please bring harnesses, shoes, and chalkbag. A second rope would be *enormously* useful so if someone has one please bring it. Also additional quickdraws would be very useful so please bring them if you can. I climb to possibly 7c, will have a go at 8a (with little real chance of success but THAT line remains...)

Diving (or snorkeling)

Members: Bragi Lovetrue, Christopher Bennett, Christoph Posch, Damir Vodenicarevic, Nasim Farahini, Florian Schuler, Gabriela Michel, James Knight, Paul B Isaac's, Daniel Renz, Saray Soldado, Siohoi Ieng, Subhrajit Roy

Leader: Florian Schuler

Let us go diving. Bring along the gear you already have. The rest can be rented. Several diving spots are possible. At Capo Galera Diving Center we have a discount of 10% on the rates published on their website for dives (but 0% for snorkeling).

UK Election

Members: Adrian Whatley, Michael Hopkins, James Knight, Alexander Rast, Sim Bamford

Leader: Adrian Whatley

This group will meet once and once only on the evening of Thursday 7th May to follow the results of the UK general election, subject to the availability of suitable TV channels or internet bandwidth permitting. For homesick Brits and all those with an interest in quaint electoral systems.

Watch Champions League

Members: Bragi Lovetrue, Mostafa Rahimi Azghadi, Abu Sebastian, Antonio Ríos, Gabriela Michel, Marc Osswald, Marcel Stimberg, Orhan Celiker, Philipp Weidel, Saray Soldado

Leader: Vito De Feo

This group will meet on the evening of Tuesday 5th May and Wednesday 6th May to follow the Champions League matches Juventus-Real Madrid and Barcellona-Bayern.

Swimming

Members: Christopher Bennett, Michael Hopkins, Nicolai Waniek, Ole Richter, Selina La Barbera, Siohoi Ieng, Viviane Ghaderi

Leader: Nicolai Waniek

What's the best way to get rid of these yummy extra-calories that we gain during CC15? Swimming, of course. But freezing in the water just on your own is kind of dull. So if you're interested in aquatic sports, want to get faster, improve your swimming style, or just simply want to go swimming in a group, join in. The plan is to go swimming (almost) each day around 17:30. Let's meet at the pool, and in case it's already occupied, we'll slowly drift towards the sea.

Last modified 3 years ago Last modified on 05/08/15 09:41:04

Attachments (2)

Download all attachments as: .zip