Xin Li李鑫

02 / 03

Train

How can models learn from verifiers, rewards, and other models?

What the training signal should be — which reward, and how much each teacher's feedback should count.

Work on this thread7 papers

TMLR 2026

Re:Form: Reducing Human Annotations in Scalable Formal Software Verification with RL in LLMs: A Preliminary Study on Dafny

Chuanhao Yan*, Fengdi Che*, Xuhan Huang*, Xu Xu*, Xin Li*, Yizhi Li*, Xingwei Qu*, …, Jie Fu

Trains 0.5B–14B models to write Dafny specifications with minimal human annotation — no natural-language chain-of-thought, auto-generated training data, rewards from the Dafny verifier; on 300 out-of-domain DafnyComp programs, the 14B RL model passes the verifier on 14.0% at pass@1, against 8.3% after SFT alone and 2.7% for Claude, the model that generated its training data.

npj Wireless Technology 2026

Reconstructing wireless signals for low altitude networks using small language models

Xin Li, Ran Liu, Chau Yuen

Small language models can learn physical reasoning from rewards. Fine-tuned with GRPO and physics-informed rewards, a 0.5B model reconstructs RSSI along UAV flight paths at 2.85 dB mean absolute error, 49% below its untrained baseline and on par with GPT-4o, transfers zero-shot to spatial interpolation, and runs at 3 ms latency.

EMNLP 2026 Industry

GraphReduce: Coverage-Preserving LLM Aggregation for E-commerce Review Insights

Hao Jiang, Xin Li, Yichi Zhang, Weisi Lin

Turns thousands of review tuples into a ranked list of product insights while every tuple stays attached to the insight it supports: membership is computed outside the LLM, so coverage is 100% with no duplicates. An anti-copy reinforcement objective lifts the upstream extractor from 59.3% to 93.1% production quality.

Preprint 2026

AesCode: Aesthetic Code Generation with Decoupled Cross-Modal Rewards

Xin Gao, Yifan Yang, Xin Li, Pan Wang, Yan Li, Muzhao Tian, Mingxi Cheng, Bei Liu, Kai Qiu, Qi Dai, Dong Chen, Yue Dong, Leandros Tassiulas, Rex Ying, Ji Li, Chong Luo

Code models cannot see how layout and color come together on the canvas. AesCode describes the target as a design graph and rewards it on two channels: deterministic verifiers score what can be parsed from code and rendering, and a VLM judge scores the rest against a per-sample Visual Graph Rubric. Trained with RL, AesCode-8B and AesCode-32B improve visual quality by 31.1 and 22.4 points over same-scale Qwen3-VL baselines given the same references.