M. Sc. Alexandru Vasilache
- 20.11.2025
- Software-Level Sparsity Optimization for Low-Power Spiking Neural Networks
- Group: Prof. Becker
- vasilache ∂does-not-exist.fzi de
Corrector: Prof. Dr. Yulia Sandamirskaya (ZHAW Zürcher Hochschule für Angewandte Wissenschaften)
Summary of the dissertation
Modern artificial intelligence is very powerful, but also consumes a lot of energy, which limits its use in mobile devices such as robots or biomedical implants.
Alexandru Vasilache's dissertation addresses this problem by developing software methods for optimizing so-called spiking neural networks (SNNs), which work in a particularly energy-efficient manner based on the model of the brain.
The methods developed specifically reduce unnecessary neuronal activity and connections and thus enable significant energy savings, which allows the use of complex AI directly on the end device.
We congratulate Alexandru Vasilache on this success!
Neuromorphic Computing
Neuromorphic computing mimics the structure of the human brain by replicating the spike activity of individual neurons, providing the flexibility to simulate only the active neurons to achieve extreme computational efficiency. This is further enhanced by the binary information transfer paradigm represented by a spike, resulting in an impressive reduction in memory, energy and computational overhead. The simulation of single neurons also allows the implementation of biologically plausible learning methods that enable local learning.
Embodied intelligence
Embodied intelligence focuses on the fusion of physical robotic bodies with cognitive processes. This approach integrates sensory perception, motor control and cognition, mimicking how living organisms interact and adapt to their environment. Embodied intelligence systems use real-world experience as a critical component of their cognitive processes, enabling adaptive behavior and learning through direct interaction with the physical environment.
Generalist AI
In contrast to specialized AI, which only excels in certain areas, generalist AI, or Artificial General Intelligence (AGI), aims to replicate human-like cognitive abilities to learn and accomplish multiple tasks. At the heart of AGI is the ability to generalize, i.e. learning from previous experiences and extrapolating this information to new, unfamiliar situations. This ability to constantly learn from experience without relearning previous information is also known as continuous learning.