Hi there! I am an aspiring researcher curious about the theory of Intelligence. My work focuses on the mathematical foundations of Deep Learning, specifically applying spectral theory and differential geometry to understand and improve Artifical Intellegence architectures.
Currently, I am investigating the spectral geometry of Attention mechanisms to address rank collapse in Transformers.
- Geometric Deep Learning: Analyzing the manifold topology of token embeddings and the curvature introduced by non-linear activation functions.
- Spectral Theory: Applying Random Matrix Theory (RMT) to analyze the singular value spectrum of Transformer weights.
- Distributed Systems: Building decentralized orchestration tools for high-performance computing.
LinkedIn: linkedin.com/in/ruasnv