Enhanced optimization with composite objectives and novelty pulsation

摘要: An important benefit of multi-objective search is that it maintains a diverse population candidates, which helps in deceptive problems particular. Not all diversity useful, however: candidates optimize only one objective while ignoring others are rarely helpful. A recent solution to replace the original objectives by their linear combinations, thus focusing on most useful tradeoffs between objectives. To compensate for loss diversity, this transformation accompanied selection mechanism favors novelty. This paper improves approach further introducing novelty pulsation, i.e. systematic method alternate and local optimization. In highly problem discovering minimal sorting networks, finds state-of-the-art solutions significantly faster than before. fact, our so far has established new world record 20-lines network with 91comparators. real-world stock trading, discovers generalize better unseen data. Composite Novelty Pulsation therefore promising solving through