Latent Representation
Latent representation learning focuses on creating compact, informative summaries (latent representations) of complex data, aiming to capture essential features while discarding irrelevant details. Current research emphasizes developing effective methods for generating these representations, particularly using architectures like autoencoders, variational autoencoders (VAEs), Joint Embedding Predictive Architectures (JEPAs), and diffusion models, often within self-supervised or semi-supervised learning frameworks. These advancements are improving performance in various downstream tasks, including image classification, natural language processing, and medical image analysis, by enabling more efficient and robust model training and enhancing interpretability. The ability to learn meaningful latent representations is crucial for advancing machine learning across numerous scientific disciplines and practical applications.
Papers
Sketching the Expression: Flexible Rendering of Expressive Piano Performance with Self-Supervised Learning
Seungyeon Rhyu, Sarah Kim, Kyogu Lee
A Fair Experimental Comparison of Neural Network Architectures for Latent Representations of Multi-Omics for Drug Response Prediction
Tony Hauptmann, Stefan Kramer
GAUDI: A Neural Architect for Immersive 3D Scene Generation
Miguel Angel Bautista, Pengsheng Guo, Samira Abnar, Walter Talbott, Alexander Toshev, Zhuoyuan Chen, Laurent Dinh, Shuangfei Zhai, Hanlin Goh, Daniel Ulbricht, Afshin Dehghan, Josh Susskind
Contrastive Masked Autoencoders are Stronger Vision Learners
Zhicheng Huang, Xiaojie Jin, Chengze Lu, Qibin Hou, Ming-Ming Cheng, Dongmei Fu, Xiaohui Shen, Jiashi Feng