Quant Researcher @ JoinQuant · Research in Trustworthy AI 🤖
Explainability 🔍 · Robustness 🛡️ · Faithfulness ✅ · Safety 🔒
I am Songning Lai (you can call me Sony). I received my B.S. from the School of Information Science and Engineering, Chongxin College, Shandong University, supervised by Prof. Zhi Liu. I was a Research Assistant at HKUST(GZ) AI Thrust & INFO Hub, supervised by Prof. Yutao Yue. I also had an international research experience as a Visiting Student at KAUST. I am currently a Quant Researcher at JoinQuant.
My primary research interest lies in Trustworthy AI, especially explainability, robustness, faithfulness, and safety. A large portion of my work centers on Concept Bottleneck Models (CBMs), including faithful vision-language interpretation (FVLC, ICLR 2024), medical explanation (Med-MICN, NeurIPS 2024), continual learning (CONCIL, ACM MM 2025 BNI), robust medical diagnosis (SVCT, ECML-PKDD 2025), and semantic security / backdoor attacks (CAT, TMLR 2026).
Beyond CBMs, I also work on faithful time series forecasting (FTS, ACM MM 2025), LLM explainability & knowledge editing (ACE, ICLR 2026), and trustworthy applications in autonomous driving (DRIVE, ICRA 2025). More details, project pages, and a fuller publication list are available on my personal website.
| Project | Focus | Venue | Links |
|---|---|---|---|
| CAT | Concept-level backdoor attacks on Concept Bottleneck Models | TMLR 2026 | Web · Code |
| CONCIL | Continual learning for multimodal Concept Bottleneck Models | ACM MM 2025 BNI | Web · Code |
| FVLC | Faithful vision-language interpretation via Concept Bottleneck Models | ICLR 2024 | Web · Code |
| ACE | Attribution-controlled knowledge editing for multi-hop factual recall | ICLR 2026 | Web · Code |
| Area | Topics |
|---|---|
| Concept Bottleneck Models | FVLC for faithful vision-language interpretation (ICLR 2024); Med-MICN for multi-dimensional medical explanation (NeurIPS 2024); CONCIL for continual learning in multimodal CBMs (ACM MM 2025 BNI); SVCT for robust medical CBMs (ECML-PKDD 2025); CAT for concept-level backdoor attacks and CBM security (TMLR 2026) |
| Trustworthy ML | Faithfulness, robustness, interpretability, and safety in CBMs and time series; FTS for faithful multimedia web forecasting (ACM MM 2025); time-series reliability under future uncertainty (ICML 2025, ICASSP 2026) |
| Applications | DRIVE for interpretable autonomous driving (ICRA 2025); ACE for LLM knowledge editing and explainability (ICLR 2026); medical imaging, multimodal learning, and trustworthy deployment in real-world systems |
- 03.2026 — CAT: Multimodal Deception in Explainable AI: Concept-Level Backdoor Attacks on Concept Bottleneck Models accepted at TMLR 2026. Project Page · Code
- 01.2026 — ACE: Attribution-Controlled Knowledge Editing for Multi-hop Factual Recall accepted at ICLR 2026. Project Page · Code
- 01.2026 — Towards Reliable Time Series Forecasting under Future Uncertainty: Ambiguity and Novelty Rejection Mechanisms accepted at ICASSP 2026 (CCF B).
- 08.2025 — CONCIL: continual learning for multimodal concept bottleneck models accepted at ACM MM 2025 BNI Track. Project Page · Code
- 07.2025 — FTS: faithful multimedia web forecasting accepted at ACM MM 2025. Project Page · Code
- 05.2025 — SVCT: Stable Vision Concept Transformers for Medical Diagnosis accepted at ECML-PKDD 2025. Project Page · Code
- 01.2025 — DRIVE: interpretable autonomous driving accepted at ICRA 2025. Project Page · Code
- 09.2024 — Med-MICN: multi-dimensional explanation for medical classification accepted at NeurIPS 2024. Project Page · Code
- 01.2024 — FVLC: Faithful Vision-Language Interpretation via Concept Bottleneck Models accepted at ICLR 2024. Project Page · Code
More news on my website.
| Year | Venue | Paper | Project / Code |
|---|---|---|---|
| 2026 | TMLR | CAT: Concept-Level Backdoor Attacks on Concept Bottleneck Models | Web · Code |
| 2026 | ICLR | ACE: Attribution-Controlled Knowledge Editing for Multi-hop Factual Recall | Web · Code |
| 2026 | ICASSP | Towards Reliable Time Series Forecasting under Future Uncertainty: Ambiguity and Novelty Rejection Mechanisms | — |
| 2025 | ACM MM BNI | CONCIL: Continual Learning for Multimodal Concept Bottleneck Models | Web · Code |
| 2025 | ACM MM | FTS: Faithful TimeSieve for Multimedia Web Forecasting | Web · Code |
| 2025 | ICRA | DRIVE: Interpretable Ensemble Framework in Autonomous Driving | Web · Code |
| 2025 | ECML | SVCT: Stable Vision Concept Transformers for Medical Diagnosis | Web · Code |
| 2024 | NeurIPS | Med-MICN: Multi-dimensional Explanation for Medical Classification | Web · Code |
| 2024 | ICLR | FVLC: Faithful Vision-Language Interpretation via Concept Bottleneck Models | Web · Code |
Full list: xll0328.github.io — Publications
| Type | Award |
|---|---|
| 🏆 | NeurIPS 2024 Travel Award |
| 🏆 | ICRA 2025 Travel Award |
| ⭐ | IEEE/EI CISP-BMEI 2022 Best Paper Award |
| 🥇 | First Prize, Contemporary Undergraduate Mathematical Contest in Modeling (National, top 0.6%) |
| 🥇 | First Prize, MathorCup University Mathematical Modeling Challenge (National, top 3%) |
| 🥈 | Second Prize, National Undergraduate Electronic Design Contest (Shandong Province) |
| 🥈 | Second Prize, National Crypto-math Challenge (East China) |
| 🎓 | Outstanding graduate of Shandong Province & Shandong University |
| 📊 | 40+ university-level awards (competitions, social practice, innovation, sports, volunteer, scholarship) |
| Period | Role | Place |
|---|---|---|
| Now | Quant Researcher | JoinQuant |
| Apr 2024 – Sep 2025 | Research Assistant | HKUST(GZ) AI Thrust & INFO Hub |
| Apr 2023 – Mar 2024 | Visiting Student | KAUST |
| Sep 2020 – Jun 2024 | B.S. (Information Science and Engineering) | Shandong University, Chongxin College |
Conference & Journal Reviewer: ECAI, ICML, KDD, ICLR, CVPR, ICCV, NeurIPS, ACM MM, IJCAI, Expert Systems with Applications, etc.
Jiayu Yang · Wenshuo Chen · Jiemin Wu · Weilin Ruan · Jiakang Li · and more on my website
Interested in collaboration or chat? 📧 Email me · 🌐 Full site