Estimation in High-dimensions

(3 publications)
Here we look at the fused lasso and trend filtering, which are locally adaptive non-linear filters that can outperform wavelet denoising. We also explore heteroscedasticity in high-dimensional models, for variance function estimation and locating master regulators in genomics data.
Higher-Order Total Variation Classes on Grids: Minimax Theory and Trend Filtering Methods (NIPS)
Veeranjaneyulu Sadhanala, Yu-Xiang Wang, James Sharpnack, Ryan J. Tibshirani
Neural Information Processing Systems (NIPS), 2017.
Approximate Recovery in Changepoint Problems, from L2 Estimation Error Rates (arXiv, NIPS)
Kevin Lin, James Sharpnack, Alessandro Rinaldo, Ryan J. Tibshirani
Neural Information Processing Systems (NIPS), 2017.
Mean and Variance Estimation in High-dimensional Heteroscedastic Models with Non-convex Penalties (arXiv)
J. Sharpnack and M. Kolar
In Submission.
Variance Funtion Estimation in High-Dimensions (pdf)
M. Kolar, and J. Sharpnack
With Oral Presentation
International Conference of Machine Learning, ICML 2012