MemTrial: Reducing Market Noise in LLM Portfolio Agents
October 9, 2026
MemTrial improves LLM portfolio agents by using a fractional factorial design to isolate the impact of specific memories from market moves. It assigns credit to individual experiences using Banzhaf values to prevent agents from incorrectly attributing market trends to specific decisions.
HOW THIS AFFECTS YOU
●
researcherThis method provides a way to perform more accurate credit assignment in reinforcement learning agents operating in noisy environments.