[ Vis hele beskrivelsen ]

With an additive synthesizer one can design a sound by deciding the amplitude of a number of sine waves with different frequencies which are then combined. In this project the resulting waveform will be calculated using the inverse fast fourier transformation.

The project aim for energy and efficient synthesize audio on multicore platforms. 

-Implementation of a platform for running additive synthesizer experiments.

-Definition of test suites to enable performance messurments

-Propose methods for messurment of performance/energy toward efficient design of  inverse fast fourier transformation targeting synthesize of audio on an embeded platform.

[ Skjul beskrivelse ]

[ Vis hele beskrivelsen ]

www.ntnu.edu/cyborg

www.ntnu.edu/web/ntnu-biotechnology/ntnu-morphogenetic-engineering

www.ntnu.edu/socrates

The goal is to be able to use a living growing neural culture as control system for a robot. The robot will be a Pepper (www.ald.softbankrobotics.com/en/cool-robots/pepper).

The project build on previous work interfacing the Pepper robot ri neurosystems. The project work will be focused on controlling pepper by a an external simulated neuro inspired dynamic system. Further, exploring possible robot task that can be reashable within the cyborg framwork. 

[ Skjul beskrivelse ]

[ Vis hele beskrivelsen ]

The basic methods in this project will be Reservoir Computing and Self-modifying network.

Reservoir computing is a relative new approach to exploit echo state networks and liquid state-machines for computation. However, any sparsely connected network that include feedforward and feedback loops may be exploited as a reservoir.

Random-Boolean-Networks (RBN) is such a sparsely connected network that may be suitable for reservoir computing. RBN is closely connected to complex systems and have been studied with in this framework.

In this project we want to create dynamic systems that change over time as a model of growing and changing biological neural networks. The dynamic network nodes may be based on RBN-like structures (a discrete dynamic system). To capture features of the dynamic network an feedforward Artificial Neural Network (ANN) is proposed (a reservoir computing approach).

The long term goal is to understand how to construct hybrid machines, and/or to use inspiration from such machines to design bio-inspired computers.

.

[ Skjul beskrivelse ]

[ Vis hele beskrivelsen ]

The main long term goal of this project aims toward computational paradigms beyond today's machines and technology; Rethinking the fundamentals of computation

Ongoing research on cellular architectures explores parallel machines consisting of a vast amount of simple processing elements with only local communication. A key feature of such machine is self-assembly/self-organization. Further, the combination of dynamic systems and Artificial Neural Networks (ANN) is an emerging field of research. As such, design and implementation of ANN modules for readout is part of the project.

In the CARD group custom FPGA based machines are an important research tool. .

The goal of this project is to continue the work on our current FPGA based cellular machine to a new FPGA platform. Key aspects are performance and scalability. The new platform is a server with 4 state-off-the-art FPGAs cards using shared memory for communication.

A preliminary plan:
-Evaluation of current platform HW and SW.

-Explore pro and cons of the system

-Propose architecture for the new FPGA platform.

-Continue the design and implementation of HW and SW.
 

Experiments and evaluation of ANN learning processes will be part of the project.

The hardware is currently in Chisel and VHDL. Software is a c library for communication and configuration.
More the one student is possible.

Links:

Conway FPGA cards

Reading:

Bio-inspired hardware (POE)

Development of non-uniform CAs

Book on non-uniform CAs and cellular machines

Reservoir Computing paper

[ Skjul beskrivelse ]

[ Vis hele beskrivelsen ]

The main long term goal of this project aims toward computational paradigms beyond today's machines and technology; Rethinking the fundamentals of computation

Nanomagnetic arrayes can be viewed as self-organizing systems. In the SOCRATES project such nanomagnetic arrayes are our targeted technology for a  for a new computing paradigm. 

Nanomagnets can be modelled using super computers and GPU clusters. However, when array size increase above approx 50 the simulation time explode. As such a more abstract modelling approach is needed to be able to simulate large nanosystems.

In SOCRATES we use 2D arrayes of nanomagnets, the architecture is close to a cellular automata (CA) architecture. Further the nanomagnets, simmilar to CAs, self-organization through local interactions.

In this project the goal is to extract key properties from the low level nanomagnetic simulations as to be able to create a higher level of abstraction model, whilst keeping the emergent properties of the system.

The project will include these main activities

-Initial nanomagnetic simulations using mumax (likely on the EPIC cluster) to extract key behaviour of the nanosytem.

-Propose simulation setups capturing different regimes (state space trajectories) and emergent behaviour from self-organiziation in the nanomagnetic simulation.

-Building prototype CA based models to emulate behaviour of the nanomagnetic simulations.

-User the CA based model in experimental work. The platform will be based on Reservoir Computing.

If you find the discription intresting, please drop by my office for a chatt.

[ Skjul beskrivelse ]