DPPM: Dual-Path Parametric Memory for Personalized Language Models
cs.CL
Submitted: 2026-10-08
Updated: 2026-10-08
Comments: 12 pages, 5 figures
Code: https://github.com/isyuhaochen/DPPM
License: http://creativecommons.org/licenses/by/4.0/
The gist: Long-term personalization requires language models to use interaction history to track users' preferences across sessions.
Terminology
Abstract
Long-term personalization requires language models to use interaction history to track users' preferences across sessions. Parametric memory encodes this interaction history into model parameters or adapters, reducing the need to include it in the inference context. However, independent context compilation leaves cross-session integration unspecified, while recurrent updates can attenuate earlier evidence. To address these challenges, we propose Dual-Path Parametric Memory (DPPM). Its Evidence path directly pools representations of the interaction history to preserve earlier evidence, while its Delta path sequentially updates an associative state to capture changes. Fusing both outputs produces history-conditioned LoRA adapters that combine evidence accumulation with ordered revision. Across multiple backbones, DPPM outperforms the evaluated baselines, achieving 54.22% on PersonaMem-v2 and 86.79% on PrefEval. These results suggest that DPPM provides a simple and effective design choice for cross-session personalized parametric memory.
Sources
- On-Policy Distillation of Language Models: Learning from Self-Generated Mistakes
- Titans: Learning to Memorize at Test Time
- Doc-to-LoRA: Learning to Instantly Internalize Contexts
- M3-Embedding: Multi-Linguality, Multi-Functionality, Multi-Granularity Text Embeddings Through Self-Knowledge Distillation
- VehicleMemBench: An Executable Benchmark for Multi-User Long-Term Memory in In-Vehicle Agents
- Mem0: Building Production-Ready AI Agents with Scalable Long-Term Memory
- LightMem: Lightweight and Efficient Memory-Augmented Generation
- Learning What to Share and What to Personalize: Hierarchical Strategy Co-Evolution for Agent Memory
- PersonaMem-v2: Towards Personalized Intelligence via Learning Implicit User Personas and Agentic Memory
- delta-mem: Efficient Online Memory for Large Language Models
- PERMA: Benchmarking Personalized Memory Agents via Event-Driven Preference and Realistic Task Environments
- SHINE: A Scalable In-Context Hypernetwork for Mapping Context to LoRA in a Single Pass
- Evaluating Very Long-Term Conversational Memory of LLM Agents
- Locating and Editing Factual Associations in GPT
- Mass-Editing Memory in a Transformer
- MemGPT: Towards LLMs as Operating Systems
- Linear Transformers Are Secretly Fast Weight Programmers
- DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models
- Towards Evolving Context Parameterization for Large Language Models
- Learning to (Learn at Test Time): RNNs with Expressive Hidden States
Related papers
- Exploring Solution Divergence and Its Effect on Large Language Model Problem Solving
- Ishigaki-IDS-Bench: A Benchmark for Generating Information Delivery Specification from BIM Information Requirements
- Subliminal Steering: Stronger Encoding of Hidden Signals
- MedStruct-S: A Benchmark for Key Discovery, Key-Conditioned QA and Semi-Structured Extraction from OCR Clinical Reports
- The End of Transformers? On Challenging Attention and the Rise of Sub-Quadratic Architectures
- Untangling the Mechanisms of Misleading Context in Medical Question Answering