MSc thesis projects - Cognitive sensor nodes and systems

Here is a list of possible MSc thesis projects related to the theme Cognitive sensor nodes and systems. This is intended just to give an idea, actual projects are usually defined after discussion with the advisor.

[2023] A factor-graph representation for bio-plausible local learning rules

In this project, we will investigate a factor-graph representation to derive local learning rules that will allow for low-cost adaptive AI hardware.

[2023] Meta-learning for low-cost adaptive neuromorphic hardware

In this project, we will investigate meta-learning, an emerging machine learning technique that allows tuning the training process in order to learn faster and more efficiently (i.e. learning-to-learn), to compensate for the non-idealities of synaptic plasticity in neuromorphic hardware.

[2023] Mixed-signal neuromorphic hardware for sub-10µW on-chip online learning at the edge

In this project, we will investigate how a few well-chosen analog blocks can boost the efficiency of a neuromorphic architecture for on-chip online learning at the edge.

[2023] A hardware architecture based on the least-control principle for bio-plausible local online learning

In this project, we will investigate a neuromorphic hardware implementation of the least-control principle (LCP), which offers a bio-plausible way of training artificial neural networks.

[2023] Monostable multivibrator networks for low-cost spike-based computing

In this project, we will explore monostable multivibrator (MMV) networks for low-cost spike-based computing and deploy them in a real-world event-based processing use case.

[2023] Dendritic circuits in neuromorphic hardware

In this project, we will investigate a key element of biological neurons: dendrites. This will include a survey of the key dendritic functions, of which a subset will then be implemented in custom neuromorphic hardware.

[2023] Sparse spiking neural networks for radar-based human activity recognition

In this project, we will minimize the computational footprint of always-on human activity recognition by taking inspiration from the brain, where neurons transiently exhibit high spike-based activity patterns only when unexpected input is received.