Difference between revisions of "Neuromorphic Intelligence In An Embedded System in Collaboration with AiCTX"
|Line 108:||Line 108:|
Latest revision as of 14:45, 10 November 2020
aiCTX is a leading-edge neuromorphic computing company. It provides dedicated mixed-signal neuromorphic processors which overcome the limitations of legacy von Neumann computers to provide an unprecedented combination of ultra-low power consumption and low-latency performance. aiCTX has a unique technological edge and IP portfolio that comes from over 20 years of experience in mixed-signal neural processor design, advanced neural routing architectures, and neural algorithms. Starting from 2019 IIS and aiCTX are providing several semester and master projects in the field of neuromorphic intelligence using their processor to build a whole working embedded system. The student will deal with both hardware and software building a prototype to cover a specifc application. In particular we identify 3 possibile projects but more can be found:
More Algorithms project topic
Benchmark how varying percentage of mismatch, for difference neuron / synapse parameters, degrades reservoir performance, for a fixed network between chips. Training output layer for reservoir networks, in the presence of mismatch between chips. Guarantee that the resulting network has equivalent / baseline performanc between chips Various forms of structured reservoirs, applied to specific tasks Wilson/Cowan oscillator reservoirs Excitatory subnetwork reservoirs Correlation-based recurrence reservoirs Implement one in-reservoir training approach, using our pipeline FORCE full-FORCE Devene / Machens / etc Implement one reservoir transfer approach, using our pipeline non-spiking → spiking spiking → DynapSE in loop non-spiking → DynapSE in loop
Object tracking with SCNNs and benchmarking Dataset generation for surveillance with DAVIS Data collection and labeling Model development Implementation of standard CNN models with SNNs (Like Lenet 5 example in our documentation) Inception Resnet Investigate implementing recurrent CNN networks using sinabs SW package
Hardware-software projects suggestions
Mutual-information based optimisation of signal processing front end Apply to wake-phrase or audio application Or vibration-based fault detection True anomaly detection with on-line learning Vibration-based anomaly detection Applying algorithms from Robert Legenstein’s group Spoken phoneme recognition network Voice-activity detector use case demo Explore weight-agnostic networks for mixed-signal reservoirs
Depending on the applicant's profile and project type, his tasks may involve some of the following:
(not all need to be met by the single candidate)
- Experience using the laboratory instrumentation - signal generators, oscilloscopes, DAQ cards, Matlab etc..
- knowledge of microcontroller programming and PC programming (C/C++, preferably embedded C)
- basic knowledge or interests on power converters, wireless communication, and circuit design at a components level (IC design is NOT involved)
- Motivation to build and test a real system
- PCB Desing or willing to learn it
- Machine learning and deep learning on PC and microcontroller (or the motivation/interess to learn it)
- Basic or strong motivation to learn Neuromorphic Ingtelligence
Figure's source: https://aictx.ai/ Detailed Task Description A detailed task description will be worked out right before the project, taking the student's interests and capabilities into account.
- Looking for Semester and Master Project Students
- Supervisors: Michele Magno
- 35% Theory
- 45% Implementation
- 20% Testing