The LifeHD system is designed for continuous unsupervised learning from real-world data streams without labels on resource-constrained devices, based on Hyperdimensional Computing (HDC) paradigm with novel components for optimal performance.
LifeHD shows significant improvements in unsupervised clustering accuracy and energy efficiency compared to NN-based baselines, with up to 74.8% accuracy improvement and 34.3x energy efficiency increase.
LifeHD includes two extensions, LifeHDsemi and LifeHDa, to handle scarce labeled samples and power constraints, enhancing its adaptability to practical edge platforms and IoT scenarios.
The work demonstrates the importance of continuous learning in sensor systems without supervision and the potential of Hyperdimensional Computing for lightweight edge computing applications.
#lifehd-system #hyperdimensional-computing #unsupervised-learning #resource-constrained-devices #edge-computing
Collection
[
|
...
]