TESLA

TESLA estimates time-varying networks from time-series nodal attribute data using temporally smoothed ℓ1-regularized logistic regression to recover sparse dynamic structures.

Key Features:

• Dynamic Network Estimation: Captures temporal evolution of network topology including topological rewiring and semantic changes across time.
• Temporally Smoothed ℓ1-Regularized Logistic Regression: Implements logistic regression with temporal smoothing and ℓ1 regularization to model time-varying interactions.
• Lasso-Style Sparse Structure Recovery: Uses ℓ1-regularization to produce sparse and interpretable network structures suitable for high-dimensional data.
• Convex Optimization Framework: Formulates the estimation as a standard convex optimization problem solvable with generic convex solvers for scalability to large networks.

Scientific Applications:

• Biological Systems: Reverse engineering of gene regulatory networks in Drosophila melanogaster, applied to datasets of over 4,000 genes across the organism's life cycle.
• Social Networks: Reconstruction of latent sequences in temporally rewiring political and academic social networks from longitudinal data.

Methodology:

Temporally smoothed ℓ1-regularized logistic regression (lasso-style sparse structure recovery) formulated as a standard convex optimization problem and solved with generic convex solvers, with temporal smoothing across time points constrained by data sampling frequency.