AAAI
CAMA: Enhancing Mathematical Reasoning in Large Language Models with Causal Knowledge
Builds a causal graph of reusable problem-solving knowledge and retrieves relevant strategies to guide mathematical reasoning.
Lei Zan (昝磊)
I study how machines can identify structure in complex data, reason with it, and remain useful when the world changes.
My work connects causal discovery, time-series modelling, root-cause analysis, and foundation-model agents. I care about methods that are both scientifically grounded and ready to work on real systems.
Bio
I am an AI research engineer with a PhD in machine learning and experience spanning both academic and industrial research.
During my industrial PhD with EasyVista and Université Grenoble Alpes, I developed causal discovery and root-cause analysis methods for heterogeneous operational data. The resulting methods were integrated into EV Observe, and part of the work was contributed to the open-source Tigramite library.
At Huawei Noah's Ark AI Lab, I extended this research to foundation models: mathematical reasoning agents, causal diagnosis for telecommunications networks, and synthetic observational and interventional data for time-series pre-training. Across these projects, my recurring question is how structural knowledge can make an AI system more reliable, interpretable, and efficient.
Publications
2026
Preprint
EvoCause: Towards Causal Graph Evolution with Large Language Models for Root Cause Analysis
Uses language-model guidance to refine causal graphs and improve root-cause identification in complex event systems.
Preprint
Post-Training in Time Series Foundation Models
Studies a unified post-training framework for adapting time-series foundation models across forecasting and classification tasks.
2024
CIKM
On the Fly Detection of Root Causes from Observed Data with Application to IT Systems
Introduces an efficient root-cause method for threshold-based IT systems without requiring a known causal graph.
CIKM
Causal Discovery from Heterogeneous Multivariate Time Series
Presents the doctoral research programme on learning causal structure from heterogeneous operational time series.
2023
AISTATS
Root Cause Identification for Collective Anomalies in Time Series Given an Acyclic Summary Causal Graph with Loops
Develops EasyRCA to identify the causes of collective anomalies using causal structure and normal and abnormal observations.
Preprint
Case Studies of Causal Discovery from IT Monitoring Time Series
Examines the practical constraints and evaluation questions that arise when causal discovery meets operational monitoring data.
2022
Entropy
A Conditional Mutual Information Estimator for Mixed Data and an Associated Conditional Independence Test
Proposes an estimator and independence test for datasets that mix quantitative and qualitative variables.
No publications match this filter.
Awards
2026
Spark Award
Huawei Digital Communications Wireless Department
2025
Grand Prize
Huawei France Tech Arena Coding Competition
2024
Invited Speaker
Young Statisticians and Probabilists Day, French Statistical Society