DepthBench: Measuring How Residual Connections Enable More Computational Depth
Preprint, 2026
Keyu Wang
I'm an M.Sc. student in Machine Learning at the University of Tübingen and a research assistant at the Max Planck Institute for Intelligent Systems, advised by Dr. Shiwei Liu. Before that, I obtained my B.Eng. in Artificial Intelligence from Southeast University, where I was advised by Prof. Guilin Qi.
Research
My research interests center on the architectures, data and optimization of foundation models, and how they shape training dynamics, generalization, and scaling behaviours.
Experience
Max Planck Institute for Intelligent Systems
2025.12 – 2027.01
Tübingen, Germany
Publications
·
DepthBench: Measuring How Residual Connections Enable More Computational Depth
Preprint, 2026
One LR Doesn't Fit All: Heavy-Tail Guided Layerwise Learning Rates for LLMs
ICML 2026
When Fewer Layers Break More Chains: Layer Pruning Harms Test-Time Scaling in LLMs
COLM 2026
Demystifying the Roles of LLM Layers in Retrieval, Knowledge, and Reasoning
ICASSP 2026
Can Large Language Models Understand DL-Lite Ontologies? An Empirical Study
Findings of EMNLP 2024
An Embedding-based Approach to Inconsistency-tolerant Reasoning with Inconsistent Ontologies
IJCKG 2023
Resources
ScalingDepth: A collection of residual connection designs.
Service
Conference Reviewer: COLM 2026, CPAL 2026/2027
Workshop Reviewer: On-device Intelligence @ NeurIPS 2026
Co-organizing: SEU-131AIClub
Misc.
My Chinese name is 汪可予. I come from Tongling, China, a small city by the Yangtze River.
I'm a big fan of animation, sports, chess, and board and card games.