Center-of-Mass Interpretation of Evolution Strategies
arXiv cs.AI · 27 Oct 2026, 03:59 AEST
A new geometrical interpretation of evolution strategies (ES) updates is presented. This approach visualizes the update as a movement towards the reward-weighted center of mass of a population of perturbed models.
Key facts
- Evolution strategies (ES) update a model by sampling a population of perturbed copies, scoring each one, and moving in the reward-weighted direction of the perturbations.
- We present a geometrical interpretation of this update instead.
- The mean-subtracted ES update is exactly this displacement, scaled by a scalar factor that depends on the mean reward, for any finite population.
- In the limit of a large population or under antithetic sampling, the centroid sits on the model itself, so ES has an even simpler geometric interpretation.
- Our contribution is an intuitive, geometric way to visualize an ES update, exact for any finite population.
Read aloud
A new geometrical interpretation of evolution strategies ES updates is presented. This approach visualizes the update as a movement towards the reward-weighted center of mass of a population of perturbed models. Evolution strategies ES update a model by sampling a population of perturbed copies, scoring each one, and moving in the reward-weighted direction of the perturbations. We present a geometrical interpretation of this update instead. The mean-subtracted ES update is exactly this displacement, scaled by a scalar factor that depends on the mean reward, for any finite population. In the limit of a large population or under antithetic sampling, the centroid sits on the model itself, so ES has an even simpler geometric interpretation. Our contribution is an intuitive, geometric way to visualize an ES update, exact for any finite population.
Sources in this cluster
- Center-of-Mass Interpretation of Evolution Strategies
arXiv cs.AI · 27 Oct 2026, 03:59 AEST