FLock.io's papers on synthetic data generation accepted at ACL 2026

AIMPACT Update: On May 5 (UTC+8), FLock.io announced that two papers on synthetic data generation have been accepted by ACL 2026. The paper "CircuitSynth: Reliable Synthetic Data Generation" has been selected for the Findings, while "GraphSynth: Resolving the Diversity-Reliability Trade-off with Probabilistic Factor Graphs" has been accepted for the Main Conference. This research addresses the long-standing challenge of the diversity-reliability trade-off in synthetic data generation. Traditional methods often suffer from hallucinations and logical inconsistencies when generating data required for high-risk applications. CircuitSynth is a neuro-symbolic framework that employs Probabilistic Sentential Decision Diagrams (PSDDs) as traceable semantic priors to decouple semantic reasoning from language generation. GraphSynth is a knowledge-graph-guided probabilistic subspace generation framework that models attribute overlap properties using factor graphs. This breakthrough holds significant implications for AI training in domains with scarce data, such as medical diagnosis and weather prediction, enabling the development of more reliable and professional AI models.