AI can now 'forget' old information by mathematically rotating it out of phase, rather than deleting it.
April 15, 2026
Original Paper
Time is Not a Label: Continuous Phase Rotation for Temporal Knowledge Graphs and Agentic Memory
arXiv · 2604.11544
The Takeaway
This paper implements 'geometric shadowing,' where obsolete facts are 'rotated' in vector space so they no longer 'rank' during retrieval. This allows a model to prioritize new data without actually losing the old data—it's still there, just shifted 'out of sight.' This is how biological memory often works: we don't delete files, we just let them 'fade' or 'shadow.' For developers of agentic memory, this is a game-changer: you can maintain a perfect history of everything while ensuring the agent always 'thinks' with the most up-to-date facts. It's memory as a wave, not a list.
From the abstract
Structured memory representations such as knowledge graphs are central to autonomous agents and other long-lived systems. However, most existing approaches model time as discrete metadata, either sorting by recency (burying old-yet-permanent knowledge), simply overwriting outdated facts, or requiring an expensive LLM call at every ingestion step, leaving them unable to distinguish persistent facts from evolving ones. To address this, we introduce RoMem, a drop-in temporal knowledge graph module