Personal tools

Real-Time Motor-Imagery Classification Using Neuromorphic Processor

From iis-projects

Jump to: navigation, search

Short Description

Brain–computer interfaces (BCIs) aim to provide a communication and control channel based on the recognition of the subject's intentions, e.g., when performing motor-imagery (MI) from neural activity. MI-BCI systems are designed to find patterns in the (electroencephalogram) EEG signals and match the signal to the motor motion that was imagined by the subject. Such information could enable communication for severely paralyzed users, control of a wheelchair, or assistance in stroke rehabilitation.

MI-BCIs are still susceptible to errors, mostly due to high inter- and intra-subject variance in EEG data, resulting in low classification accuracy. Moreover, BCI requires very low latency and, to be effective, low power processing that can be implemented in a wearable device. Compared to the "traditional" artificial neural network, the spiking neural network (SNN) can provide both improved latency and energy efficiency. Previous works have shown its potentiality for biomedical signals such as ECG and EMG, and it has demonstrated a better performance when the sample size is limited. Some previous work has been presented showing the potentiality of SNN on brain signals, however, the state of the art use still very and a couple of classes motor-imagery and often they are not implemented in a neuromorphic processor, and none of them are presenting a whole system from the data acquisition to the processing.

The goal of the present project is to investigate and develop a novel neuromorphic system for Brain–computer interfaces, trained for multi-class motor-imagery, that embeds Intel Loihi as the processing core. SNN algorithms will be implemented and evaluated on real hardware. Moreover, the project has the goal of acquiring data from real subjects to have a data set to train and evaluate the algorithms on the proposed application scenario.

Goal & Tasks

The project(s) will address the following challenges:

  • Investigate and develop techniques and methods to perform motor imagery brain-computer interface with energy-efficient SNN.
  • The algorithms will be evaluated and optimized for the capability of the Loihi Platform to both increase the energy efficiency and, at the same increase the response time of the detection, aiming to achieve an always-on system.
  • Propose novel low power mixed analog-digital systems for biomedical signal (in particular EEG but suitable also for ECG and EMG) analyses to have a real-world acquisition system designed for neuromorphic processing, including Loihi.
  • Acquire a large dataset for BCI and possibly other biomedical applications to have the possibility to test and train the SNN that will be implemented on the hardware.
  • A complete hardware and software prototype of a smart sensor system, which includes all the subsystems (sensor acquisition, preprocessing, and processing and radio communication), will be developed to demonstrate the benefits of the proposed approach and the capability to achieve perpetual low latency and energy efficiency on the challenging scenario of BCI.
  • The working prototype with the Loihi processors will be evaluated to carry out the benchmark with traditional approaches using digital processors.


(not all need to be met by the single candidate)

  • Knowdleg of high and low level programming languages (e.g. Python, embedded C)
  • Knowdleg of embedded systems
  • Knowledge of machine learning and signal processing
  • Motivation to learn spiking neural networks simulation packages (e.g. BRIAN, ANNarchy, NEST, or NEURON)
  • Motivation to build and test a real system and acquiring field data

Detailed Task Description

A detailed task description will be worked out right before the project, taking the student's interests and capabilities into account.

Status: Available

  • Looking for Semester and Master Project Students
Supervisors: Michele Magno, Xiaying Wang


35% Theory and Algorithms
35% Implementation
30% Data acquisition, Verification, and Testing

IIS Professor

Luca Benini

↑ top

Practical Details