Similar Segment Learning Multitask Optimization Based on Locally Sensitive Hashing

Evolutionary multitask optimization (EMTO) has attracted considerable interest attention recently because it enables knowledge transfer (KT) between tasks, allowing them to be solved simultaneously. However, a critical limitation in this field is that KT has mainly relied on index-aligned dimensions, which may not always be appropriate. Instead, the KT should focus on similar dimensions rather than strictly index-aligned dimensions. To address this limitation, this paper proposes a novel similar segment learning multitask optimization (SSLMTO), which incorporates a cyclic time-window segmentation strategy (CTSS) and a clustering strategy based on locally sensitive hashing (LSH-CS). Specifically, In CTSS, individuals are segmented using a cyclic time window to ensure that each dimension can learn from other index-unaligned dimensions, enhancing the population diversity and facilitating more flexible KT. Furthermore, LSH is employed to cluster the segments and identify similar segments, achieving the similar segment learning and promoting the exploitation capability of the algorithm. In addition, a parameter adaptive strategy (PAS) is used to adjust the parameters, enabling the SSLMTO to adapt to a wider range of problems. The experimental results demonstrate that SSLMTO performs significantly better than other state-of-the-art EMTO algorithms on two widely used multitask benchmarks, CEC2017 and CEC2022, highlighting the effectiveness of the SSLMTO.