Tiago Fernandes Tavares
Assistant Professor, Insper
Researcher in Representation Learning and ML Systems
About
I am an Assistant Professor at Insper in São Paulo, where I teach Linear Algebra and Information Theory, as well as Natural Language Processing.
My research is in machine learning, with a focus on representation learning. I study how neural networks recover, or fail to recover, meaningful structure from real data, and when, and within what limits, these models become useful tools. I have worked with audio, natural language, and other time-domain signals, including multimodal data.
I am always interested in connecting with researchers and engineers working on representation learning, ML systems, and privacy-preserving AI.
Selected publications
Diagnosing Neural Convergence with Topological Alignment Spectra
IJCNN 2026
A topological measure of whether independently trained networks converge to the same representation.
LogiPart: Local Large Language Models for Data Exploration at Scale with Logical Partitioning
IJCNN 2026
Local LLMs that partition large text collections into logically defined groups, keeping the data on-premises.
On Class Separability Pitfalls In Audio-Text Contrastive Zero-Shot Learning
ICASSP 2025
Shows how zero-shot results in audio-text contrastive models can be inflated by class overlap with pre-training data.
Unsupervised Improvement of Audio-Text Cross-Modal Representations
WASPAA 2023
Improves audio-text cross-modal embeddings without requiring additional labeled audio-text pairs.