Yuhang Liu's Homepage

About me:

I am a Research Fellow at Australian Institute for Machine Learning (AIML), part of The University of Adelaide, working with Prof. Qinfeng (Javen) Shi.

Research Mission:

Long-Term Objective: The C2C Landscape, Charting Principled Routes from Correlation to Causality through Data Diversity.

At the heart of my research lies the following foundational questions:

(i) What are the principled intermediate states between correlation and causality?
(ii) How can learning systems progressively traverse these states?
(iii) What concrete benefits do such states already afford?

Short-term Objectives:

Empower Interpretability, Controllability, Reasoning, and Planning in LLMs.
Explore Text Diversity to Guide the Discovery of Latent Patterns within Unstructured Data.
Develop Fundamental Theories for Representation Learning.

Email: yuhang.liu01@adelaide.edu.au

News:

(Jan 2026) Work in progress, "Concept Component Analysis: A Principled Approach for Concept Extraction in LLMs", https://arxiv.org/abs/2601.20420.
(Jan 2026) "Identifying Weight-Variant Latent Causal Models", accepted to appear at JMLR 2026.
(Jan 2026) "I Predict Therefore I Am: Is Next Token Prediction Enough to Learn Human-Interpretable Concepts from Data?", accepted to appear at ICLR 2026.
(Jan 2026) "Beyond DAGs: A Latent Partial Causal Model for Multimodal Learning", accepted to appear at ICLR 2026.
(Sep 2025) "Model Inversion with Layer-Specific Modeling and Alignment for Data-Free Continual Learning", accepted to appear at NeurIPS 2025.
(Sep 2025) "On the Value of Cross-Modal Misalignment in Multimodal Representation Learning", accepted to appear at NeurIPS 2025.
(Aug 2025) I will serve as an Area Chair for ICLR 2026.
(Jun 2025) "Causal Disentanglement and Cross-Modal Alignment for Enhanced Few-Shot Learning", accepted to appear at ICCV 2025.
(Jun 2025) "Seeing Beyond Labels: Source-Free Domain Adaptation via Hypothesis Consolidation of Prediction Rationale", accepted by TMLR 2025.
(Apr 2025) Guest Editor of Entropy Special Issue on “Rethinking Representation Learning in the Age of Large Models", 2025
(Apr 2025) "I Predict Therefore I Am: Is Next Token Prediction Enough to Learn Human-Interpretable Concepts from Data?." https://arxiv.org/abs/2503.08980, 2025
(Apr 2025) "Latent Covariate Shift: Unlocking Partial Identifiability for Multi-Source Domain Adaptation", accepted to appear at TMLR 2025.
(Jan 2025) "Coreset Selection via Reducible Loss in Continual Learning", accepted to appear at ICLR 2025.
(Jan 2025) "Analytic DAG Constraints for Differentiable DAG Learning", accepted to appear at ICLR 2025.
(Aug 2024) "Rethinking State Disentanglement in Causal Reinforcement Learning", Check out Arxiv, https://arxiv.org/abs/2408.13498
(Aug 2024) "InvariantStock: Learning Invariant Features for Mastering the Shifting Market". accepted to appear at TMLR 2024.
(Aug 2024) "Uncertainty estimation in HDR imaging with Bayesian neural networks". accepted to appear at PR 2024.
(Jul 2024) "CLAP: Isolating Content from Style through Contrastive Learning with Augmented Prompts". accepted to appear at ECCV 2024.
(Mar 2024) "Identifiable Latent Neural Causal Models". Check out Arxiv, https://arxiv.org/abs/2403.15711.
(Feb 2024) "Beyond DAGs: A Latent Partial Causal Model for Multimodal Learning". Check out Arxiv, https://arxiv.org/abs/2402.06223.
(Jan 2024) "Identifiable Latent Polynomial Causal Models Through the Lens of Change". Finally, accepted to appear at ICLR 2024.
(Oct 2023) Work in progress "Identifiable Latent Polynomial Causal Models Through the Lens of Change". Check out Arxiv, https://arxiv.org/abs/2310.15580.
(Sep 2022) Work in progress "identifiable latent causal content for domain adaptation under latent covariate shift". Check out Arxiv, https://arxiv.org/abs/2208.14161.
(Sep 2022) Work in progress "Identifying Weight-Variant Latent Causal Models". Check out Arxiv, https://arxiv.org/abs/2208.14153.
(Sep 2022) "Truncated Matrix Power Iteration for Differentiable DAG Learning", accepted to appear at NeurIPS 2022.
(Mar 2022) "Learning Bayesian Sparse Networks with Full Experience Replay for Continual Learning", accepted to appear at CVPR 2022.
(Jan 2022) Check the article summarizing part of our works using ML to help agriculture innovation, 2022.
(Mar 2019) One paper "Bayesian Nonnegative Matrix Factorization With a Truncated Spike-and-slab Prior" is accepted by ICME2019. Code coming soon.
(Feb 2019) One paper "Variational Bayesian Dropout" is accepted by CVPR2019. Project page? (Coming soon.)
(Nov 2018) Work in progress "Variational Bayesian Dropout". Check out Arxiv, https://arxiv.org/pdf/1811.07533.pdf.
(Sep 2018) Code for paper "Deblurring Natural Image Using Super-Gaussian Fields" has been released! Check "Software".
(Sep 2018) Code for paper "Frame-Based Variational Bayesian Learning for Independent or Dependent Source Separation" has been released! Check "Software".
(July 2018) One paper “Deblurring Natural Image Using Super-Gaussian Fields” is accepted by ECCV2018.
(Dec 2017) One paper “Frame-Based Variational Bayesian Learning for Independent or Dependent Source Separation” is accepted by TNNLS .

Google Sites

Report abuse