Anton van den Hengel

Beyond DAGs: A Latent Partial Causal Model for Multimodal Learning

Procedural Pretraining: Warming Up Language Models with Abstract Data

Math Blind: Failures in Diagram Understanding Undermine Reasoning in MLLMs

Concept Component Analysis: A Principled Approach for Concept Extraction in LLMs

I Predict Therefore I Am: Is Next Token Prediction Enough to Learn Human-Interpretable Concepts from Data?

Towards Identifiable Latent Additive Noise Models

The Devil is in the Tokens: Token-Level Structural Analysis Uncovers Hallucinations in LVLMs

On the Value of Cross-Modal Misalignment in Multimodal Representation Learning

Analytic DAG Constraints for Differentiable DAG Learning

Synergy and Diversity in CLIP: Enhancing Performance Through Adaptive Backbone Ensembling

Primitive Vision: Improving Diagram Understanding in MLLMs

RandLoRA: Full rank parameter-efficient fine-tuning of large models

Beyond DAGs: A Latent Partial Causal Model for Multimodal Learning

Turning Challenges into Opportunities: How Distribution Shifts Enhance Identifiability in Causal Representation Learning

Open Eyes, Then Reason: Fine-grained Visual Mathematical Understanding in MLLMs

How Do We Select Right LLM for Each Query?

Towards Effective Updating of Pretrained Symbolic Music Models for Fine-Grained Bar-Level Control

Rethinking end-to-end prediction of adsorption energies from a causal perspective