LLM Predicts Double Perovskite Space Groups with Few-Shot Learning

Jongwon Park, Inhyo Lee, Junhyeong Lee, Seunghwa Ryu· August 12, 2026 View original

Key takeaways

  • DyRIS is an LLM-agent framework for predicting space groups of double perovskites.
  • It uses dynamic, diversity-enhanced few-shot prompting to handle imbalanced datasets.
  • Rule-guided inference incorporates crystallographic domain knowledge for refinement.
  • DyRIS significantly improves prediction accuracy for underrepresented "minor" space groups.

Who benefits

Materials ScienceChemical EngineeringPharmaceuticalsRenewable Energy

Summary

DyRIS, an LLM-agent-based framework, predicts space groups of double perovskites using dynamic and diversity-enhanced few-shot prompting combined with rule-guided inference. It achieves competitive overall accuracy and significantly improves performance on underrepresented "minor" space group classes in imbalanced materials datasets.

This research introduces DyRIS (Dynamic and Diversity-enhanced Few-shot Retrieval and Rule-Guided Inference for Space-Group Prediction), an LLM-agent-based framework designed to predict the space groups (SGs) of stable double perovskite (DP) structures. Predicting SGs is challenging due to the broad compositional tunability of DPs and the strong imbalance in available datasets, where some SG classes are dominant ("major") and others are underrepresented ("minor"). DyRIS tackles this by employing diversity-enhanced dynamic few-shot prompting, which retrieves relevant in-context examples while actively mitigating the over-representation of major SGs. This retrieval mechanism is complemented by a rule-guided inference step that incorporates crystallographic domain knowledge, such as B/B' cation ordering, quantitative indicators, and major-SG bias control, to refine and rank the final Top-3 SG candidates. Evaluated on 3,528 thermodynamically filtered DP entries, DyRIS achieves competitive overall accuracy. Crucially, it significantly outperforms baselines in predicting minor SG classes, improving Minor-SG Top-1 accuracy by 3.26 percentage points relative to CrabNet. Ablation studies confirm the individual contributions of each component, demonstrating the power of combining retrieval-based LLM reasoning with expert domain knowledge for imbalanced materials datasets.

Why it matters

For materials scientists and chemists, DyRIS offers a powerful new AI tool to accelerate the discovery and design of novel double perovskite materials by accurately predicting their crucial structural properties, especially for rare or understudied compositions.

How to implement this in your domain

  1. 1Explore using LLM-agent frameworks like DyRIS for materials discovery tasks, especially with imbalanced datasets.
  2. 2Integrate domain-specific rules and expert knowledge into LLM-based prediction systems to enhance accuracy and reliability.
  3. 3Implement diversity-enhanced few-shot prompting strategies to improve model performance on minority classes.
  4. 4Apply this methodology to other scientific domains facing similar challenges with imbalanced data and complex predictions.

Original post by Jongwon Park, Inhyo Lee, Junhyeong Lee, Seunghwa Ryu

"arXiv:2608.10483v1 Announce Type: new Abstract: Double perovskites (DPs) offer broad compositional tunability, but predicting the space groups (SGs) of stable structures remains difficult because available datasets are often strongly imbalanced toward dominant SG classes. We refe…"

View on X

Originally posted by Jongwon Park, Inhyo Lee, Junhyeong Lee, Seunghwa Ryu on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses

More in AI Research

AI Engineering & DevToolsAI Research

TACTICL Compresses Tabular ICL Models, Retaining Adaptability.

TACTICL is an automated framework for compressing tabular in-context learning (ICL) models by jointly pruning transformer layers and replacing them with lightweight adapters. This method significantly reduces model size and computational demands while preserving robustness to data shifts and in-context adaptability.

Mykhailo Koshil, Matthias Feurer, Katharina EggenspergerAug 12, 2026
AI Engineering & DevToolsAI Research

MoE Proxy Models Cut LLM RL Debugging Costs.

This paper introduces Mixture-of-Experts (MoE) proxy models designed for low-cost reproduction and diagnosis of failures during Large Language Model (LLM) Reinforcement Learning (RL) post-training. These proxy models significantly reduce computational resources and time needed for debugging, while accurately preserving training dynamics and fault responses.

Yikai Wang, Chuansai Zhou, Yuhang Zhou, Weiqiang Wu, Cong Wu, Yue Deng, Ben Feng, Mingming Zhu, Beirong Zhou, Zhibin Wang, Sheng Zhong, Chen Tian, Wangze ZhangAug 12, 2026
AI Engineering & DevToolsAI Research

New Algorithm Boosts Stochastic Optimal Control Efficiency.

This paper introduces Path Integral Value Matching (PI-VM), a novel value-based algorithm for Linear Quadratic Stochastic Optimal Control (LQ-SOC) that significantly improves computational efficiency and stability. By deriving a temporal recursive form of the value function and integrating Girsanov theorem with experience replay, PI-VM matches state-of-the-art precision with order-of-magnitude efficiency gains.

Bangyan Liao, Chenglei Yu, Yuchen Yang, Chuanrui Wang, Zhisheng Song, Peidong Liu, Tailin WuAug 12, 2026