Reproduce “Synthetic User Behavior Sequence Generation with Large Language Models for Smart Homes” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zhiyao Xu, Dan Zhao, Qingsong Zou, Jingyu Xiao, Yong Jiang, Zhenhui Yuan, Qing Li
IoTGen, an LLM-based framework, generates synthetic user behavior sequences for smart homes, enabling adaptive training of intelligent models. It incorporates a Structure Pattern Perception Compression (SPPC) module, which reduces LLM input token length while preserving critical sequence information, demonstrating comparable performance to full dataset training in anomaly detection evaluations.
Key insights
- Traditional similarity-based data compression methods are inadequate for smart home behavior sequences, as they risk losing distinct yet crucial pattern information.
- Autoencoder-based reconstruction error can effectively quantify the 'importance' of structural patterns in IoT data, enabling efficient compression without compromising critical information for downstream tasks.
- Large Language Models, when provided with structured prompts and compressed contextual data, can perform complex reasoning to adapt and generate realistic user behavior sequences for changing smart home environments.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.