CORTEXA
← Browse

Junfeng Zhao

2 papers indexed

arxivcs.DBcs.AI2026-06-26

Database Context Compression for Text-to-SQL on Real-World Large Databases

Jingwen Liu, Weibin Liao, Xin Gao, Junfeng Zhao, Yasha Wang

Recent progress in Text-to-SQL has been driven by stronger language models and prompting strategies, yet performance on real enterprise benchmarks such as Spider 2.0 and BIRD remains far below that on classical academic datasets. We argue that the main bottleneck is no longer rea…

View free PDFSource page
arxivcs.LG2026-06-26

The Weakest Link Tells It All: Outcome-Supervised Process Reward Modeling via Learnable Credit Assignment

Tianyu Jia, Yue Fang, Hongxin Ding, Rihong Qiu, Zhibang Yang, Zhijing Wu, et al.

Process reward models (PRMs) enhance the reasoning capabilities of large language models (LLMs) by providing fine-grained feedback, yet training PRMs typically requires expensive stepwise annotations. Outcome-supervised PRMs offer a scalable alternative by learning from final-ans…

View free PDFSource page