Researchers have proposed a new benchmark called MemFuseBench for evaluating memory systems that can integrate information from multiple sources. The benchmark includes a pipeline that generates scenarios with tagged observations and questions, allowing for the evaluation of temporal reasoning and cross-source evidence fusion. A structured memory system called MemFuse has also been introduced, which preserves source-level evidence in event-layer atomic memory and organizes related events into cluster-layer fused memory. Experiments on MemFuseBench showed that MemFuse outperformed other memory systems under various settings. --- Why it matters: This matters to AI researchers because it addresses a common challenge in realistic settings where relevant information is often fragmented across multiple sources, requiring agents to integrate dispersed observations into coherent episodic memories. Source: https://arxiv.org/abs/2608.18704