Mark C. W. van Rossum

This paper introduces mistake-gated learning, a method that reduces synaptic updates by focusing on classification errors, improving energy efficiency and memory usage in continual learning.

This paper proposes energy-saving plasticity algorithms for neural networks, showing they reduce synaptic updates while maintaining learning performance.

This study integrates energy considerations into a reinforcement learning framework to show that switching between memory pathways based on energy reserves and reward prediction error prolongs lifespan in Drosophila.

The paper evaluates biases in population codes when few neurons are active, showing how biases depend on noise and tuning curves and introduces a method to estimate them.

The paper estimates the energy required for memory formation as 10mJ/bit and compares it to biophysical estimates and computer hardware energy requirements.

The paper examines how synaptic plasticity affects energy efficiency in learning and proposes a mechanism called synaptic caching to reduce metabolic costs.

The paper identifies biases in estimating multiple stimuli from neural population codes and shows how Gaussian Processes can model these biases.

This review discusses the functional consequences of pre- and postsynaptic expression of homeostatic and Hebbian synaptic plasticity, highlighting predicted benefits like improved sensory perception and memory savings.