MMA: Multimodal Memory Agent
This addresses the issue of overconfident errors in multimodal agents for applications requiring reliable memory-based decision-making, representing an incremental improvement with novel benchmarking.
The paper tackles the problem of unreliable memory retrieval in long-horizon multimodal agents by proposing MMA, which dynamically scores memory reliability and uses it to reweight evidence and abstain when support is insufficient, resulting in reduced variance by 35.2% on FEER and improved accuracy on benchmarks like MMA-Bench.
Long-horizon multimodal agents depend on external memory; however, similarity-based retrieval often surfaces stale, low-credibility, or conflicting items, which can trigger overconfident errors. We propose Multimodal Memory Agent (MMA), which assigns each retrieved memory item a dynamic reliability score by combining source credibility, temporal decay, and conflict-aware network consensus, and uses this signal to reweight evidence and abstain when support is insufficient. We also introduce MMA-Bench, a programmatically generated benchmark for belief dynamics with controlled speaker reliability and structured text-vision contradictions. Using this framework, we uncover the "Visual Placebo Effect", revealing how RAG-based agents inherit latent visual biases from foundation models. On FEVER, MMA matches baseline accuracy while reducing variance by 35.2% and improving selective utility; on LoCoMo, a safety-oriented configuration improves actionable accuracy and reduces wrong answers; on MMA-Bench, MMA reaches 41.18% Type-B accuracy in Vision mode, while the baseline collapses to 0.0% under the same protocol. Code: https://github.com/AIGeeksGroup/MMA.