Representation learning lies at the core of modern artificial intelligence, enabling neural networks to uncover meaningful, compact, and transferable latent features from high-dimensional data. Despite remarkable empirical successes, existing representation learning methods lack interpretability, theoretical guarantees, and principled design, leading to the gap between theoretical foundation and practical applications.

Signal processing offers a rich set of mathematical tools - such as spectral analysis, filtering, sampling, frequency transform and stochastic modeling – which provides theoretical foundations for representation learning. On the other hand, the development of representation learning can stimulate the development of advanced signal processing in large-scale, geometric, and data-driven settings, especially in graph signal processing, manifold analysis and high-dimensional signal processing.

This research topic aims to foster the joint evolution of signal processing and machine learning, focusing on developing novel theoretical concepts and enabling analytical tools, which could further bridge the gap between signal processing and practical applications.

This research topic aims to advance the frontiers and integration of signal processing (SP) and modern artificial intelligence, with a particular emphasis on representation learning. It focuses on three key directions: 1) establishing theoretical principles for representation learning from a signal processing perspective; 2) developing novel signal processing concepts and methodologies inspired by interpretable and efficient representation learning; and 3) addressing real-world challenges and practical applications through SP-inspired representation learning approaches.

The collected contributions are expected to provide analytical, principled, and interpretable foundations for representation learning grounded in signal processing theory, while extending classical SP frameworks to non-Euclidean, large-scale, and data-driven settings. Submissions that demonstrate the impact of signal processing–based representation learning in practical applications are also encouraged, with the goal of bridging theoretical foundations and real-world deployment challenges.

Topics include, but are not limited to

1. Foundations and Theory

Spectral and frequency-domain analysis for representation learning

Stability, robustness, and generalization in representation learning

Information-theoretic and stochastic signal models for representation learning

AI-enabled model analysis and interpretability in representation learning

2. Graph and Non-Euclidean Representation Learning

Graph signal processing for graph representation learning

Frequency-aware graph neural networks and spectral graph learning

Learning on manifolds, simplicial complexes, multilayer networks, and hypergraphs

Dynamic and temporal graph representation learning

3. Data Privacy and Security for Representation Learning

Privacy-preserving signal processing for representation learning

Signal processing for resilient and robust representation learning

Federated representation learning and unlearning

4. Multimodal and Large-Scale Learning

Cross-modal signal alignment and representation learning

Spatio-temporal signal processing for representation learning

Scalable and lightweight representation learning

Foundation models and pre-training for multimodal signals

5. Practical Applications

Wireless communications and computer networks

Biomedical and brain signal analysis

Geoscience and remote sensing

Image, point cloud, and multispectral imaging processing

Autonomous systems and robotics perception

Acoustics, speech, and natural language audio processing.

Are you interested in contributing a manuscript to this Research Topic?

Signal Processing for Representation Learning: Theoretical Foundations and Emerging Applications