Researchers introduce parade to improve demographic Survey simulations with LLMs
A team of researchers has proposed PARADE, a framework designed to enhance the accuracy of demographic response simulations in large language models (LLMs). The method addresses limitations in current approaches by using LoRA modules—small, trainable adjustments—to represent demographic attributes separately from survey tasks. This allows for flexible combination of demographic traits during…
Key points
- PARADE uses LoRA modules to separate demographic traits from survey tasks for flexible combination
- Reduces KL divergence by up to 52.8% in simulated survey responses on World Values Survey and SocioBench
- Supports on-demand construction of population models for unseen demographic combinations
PARADE splits the model into a shared Task LoRA (for general survey logic) and attribute-level Demographic LoRAs (for specific demographic shifts). Experiments on the World Values Survey and SocioBench with three backbone models show it reduces KL divergence—a measure of distribution mismatch—by up to 52.8% compared to existing methods. The approach aims to improve realism in simulated survey responses while keeping inference efficient.
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Research
All →- Anthropic reports AI agents spend 80% of session waiting for humans, remeasurement shows about 30% · 1 src
- CoDA enables rapid codebook development for domain-specific text annotation · 1 src
- Researchers prove Prefix-Tuning can exactly steer pretrained transformers · 1 src
- Study probes if LMs understand variable roles via SAEs · 1 src
- Google and Meta release TorchTPU for native PyTorch on TPU · 1 src
Comments
via GitHub Discussions