Temporal Binary Compression in Livia
- TBC is a hierarchical, time-based memory summarization mechanism that condenses interaction history into layered, compact representations.
- It compresses detailed recent interactions and summarizes older conversations through pairwise merging over exponential time intervals.
- Designed to emulate human-like memory decay, TBC maintains context and emotional coherence while keeping storage requirements low.
Searching arXiv for the cited paper and closely related acronym-collision papers to ground the article. arXiv search: "(Xi et al., 12 Aug 2025) Livia emotion-aware AR companion progressive memory compression Temporal Binary Compression" Temporal Binary Compression (TBC) is Livia’s time-based progressive memory compression scheme for long-term interaction history. Introduced in the emotion-aware AR companion system Livia, TBC is one of the two core memory-management algorithms in the Memory Compression Agent, alongside Dynamic Importance Memory Filter (DIMF). Its stated purpose is to keep the companion’s memory scalable while preserving the emotionally and contextually important parts of past conversations, thereby supporting continuity in dialogue and relationship personalization without requiring expensive full-history storage (Xi et al., 12 Aug 2025).
1. Definition and architectural placement
Within Livia, TBC is defined as a hierarchical, time-based memory summarization mechanism for long-term interaction history. It belongs to a modular AI architecture in which specialized agents handle emotion analysis, dialogue generation, memory management, and behavioral orchestration. TBC operates specifically inside the Memory Compression Agent, where it is paired with DIMF as part of Livia’s progressive memory compression strategy (Xi et al., 12 Aug 2025).
The scheme is described as the backbone of long-horizon memory scaling. Its function is not to preserve every utterance indefinitely, but to reorganize interaction history into a hierarchy of temporal representations. In this design, recent interactions remain accessible in fine detail, while older interactions are transformed into increasingly compact summaries. The intended result is a bounded memory structure that still supports emotionally coherent and contextually continuous interaction.
The role of TBC is therefore infrastructural rather than merely archival. It is meant to preserve what happened recently in detail, what happened earlier as compact summaries, and overarching user history as even more abstract summaries. This temporal stratification is presented as central to Livia’s ability to sustain believable long-term memory with a manageable storage footprint.
2. Motivation and design rationale
TBC was introduced because Livia is designed to be an emotionally aware companion that should remember prior interactions without letting memory storage grow uncontrollably. The paper motivates the method with four requirements: recent interactions should remain detailed; older interactions should be condensed; memory should decay in a human-like way, where fine detail fades but high-level meaning remains; and the system must stay practical for a lightweight backend using SQLite rather than expensive full-history storage (Xi et al., 12 Aug 2025).
These constraints define the design logic of TBC. The compression policy is age-driven: the further back in time an interaction lies, the coarser its representation becomes. The method is thus not a generic summarization pass over the entire memory store, but a progressive temporal policy aligned with recency, bounded storage, and persistence of high-level semantic continuity.
The paper explicitly characterizes TBC as a way to “systematically compress memory based on temporal intervals,” to “emulat[e] human-like memory decay,” and to “merg[e] memory entries pairwise within defined exponential time intervals (e.g., daily, weekly, monthly).” This framing places TBC in a lineage of temporal abstraction methods rather than bit-level binary coding. A plausible implication is that the term “