Learning Cell-Aware Hierarchical Multi-Modal Representations for Robust Molecular Modeling 1. A new framework called CHMR is proposed to address limitations in current cell-aware molecular modeling approaches, focusing on modality incompleteness and insufficient hierarchical modeling across molecular, cellular, and genomic levels. 2. CHMR introduces a tree-structured vector quantization module to capture latent biological hierarchies, enabling the model to better understand cross-scale biological mechanisms from molecules to cells and genes. 3. The framework incorporates modality augmentation and semantic consistency alignment to handle missing biological data, significantly improving robustness and generalization in molecular property prediction tasks. 4. Evaluated on nine public benchmarks with 728 tasks, CHMR outperforms state-of-the-art methods, achieving average improvements of 3.6% in classification and 17.2% in regression tasks. 5. CHMR's hierarchical and multi-modal learning approach provides a generalizable framework for integrative biomedical modeling, offering reliable and biologically grounded molecular representations. 📜Paper: arxiv.org/abs/2511.21120v1 #MolecularModeling #HierarchicalLearning #MultiModalRepresentations #BiomedicalModeling

We are so proud to announce that @omiguelnavida successfully defended his thesis proposal on "Multimodal Representation Learning for Agent Perception and Agency". Congratulations Miguel! #MultimodalRepresentations #RepresentationLearning #DeepReinforcementLearning #DeepRL

The paper proposes a novel scalable #GenerativeModel to learn #MultimodalRepresentations of high-dimensional data in an unsupervised learning context and a novel framework that allows #RL agents to zero-shot transfer policies across different subsets of input modalities.

The discussion topic this week was Multi-Modal representations. #RepresentationLearning #ClassOct2 #MultimodalRepresentations

#HaveYouMet Miguel Vasco? @omiguelnavida is a PhD student working how #ArtificialAgents can learn #MultimodalRepresentations from perceptual information provided by their environment. Fun fact: he once rode Space Mountain in Disneyland 12 times in a row. miguelvasco.com