Picture for Matus Telgarsky

Matus Telgarsky

Feature selection with gradient descent on two-layer networks in low-rotation regimes

Aug 04, 2022
Matus Telgarsky

Figure 1 for Feature selection with gradient descent on two-layer networks in low-rotation regimes

Figure 2 for Feature selection with gradient descent on two-layer networks in low-rotation regimes

Figure 3 for Feature selection with gradient descent on two-layer networks in low-rotation regimes

Access Paper or Ask Questions

Convex Analysis at Infinity: An Introduction to Astral Space

May 06, 2022
Miroslav Dudík, Ziwei Ji, Robert E. Schapire, Matus Telgarsky

Access Paper or Ask Questions

Stochastic linear optimization never overfits with quadratically-bounded losses on general data

Feb 14, 2022
Matus Telgarsky

Figure 1 for Stochastic linear optimization never overfits with quadratically-bounded losses on general data

Figure 2 for Stochastic linear optimization never overfits with quadratically-bounded losses on general data

Access Paper or Ask Questions

Actor-critic is implicitly biased towards high entropy optimal policies

Oct 21, 2021
Yuzheng Hu, Ziwei Ji, Matus Telgarsky

Access Paper or Ask Questions

Fast Margin Maximization via Dual Acceleration

Jul 01, 2021
Ziwei Ji, Nathan Srebro, Matus Telgarsky

Figure 1 for Fast Margin Maximization via Dual Acceleration

Figure 2 for Fast Margin Maximization via Dual Acceleration

Figure 3 for Fast Margin Maximization via Dual Acceleration

Figure 4 for Fast Margin Maximization via Dual Acceleration

* ICML 2021

Access Paper or Ask Questions

Early-stopped neural networks are consistent

Jun 10, 2021
Ziwei Ji, Justin D. Li, Matus Telgarsky

Figure 1 for Early-stopped neural networks are consistent

Access Paper or Ask Questions

Generalization bounds via distillation

Apr 12, 2021
Daniel Hsu, Ziwei Ji, Matus Telgarsky, Lan Wang

Figure 1 for Generalization bounds via distillation

Figure 2 for Generalization bounds via distillation

Figure 3 for Generalization bounds via distillation

Figure 4 for Generalization bounds via distillation

* To appear, ICLR 2021

Access Paper or Ask Questions

Gradient descent follows the regularization path for general losses

Jun 19, 2020
Ziwei Ji, Miroslav Dudík, Robert E. Schapire, Matus Telgarsky

Figure 1 for Gradient descent follows the regularization path for general losses

* To appear, COLT 2020

Access Paper or Ask Questions

Directional convergence and alignment in deep learning

Jun 11, 2020
Ziwei Ji, Matus Telgarsky

Figure 1 for Directional convergence and alignment in deep learning

Figure 2 for Directional convergence and alignment in deep learning

Figure 3 for Directional convergence and alignment in deep learning

Access Paper or Ask Questions

Neural tangent kernels, transportation mappings, and universal approximation

Oct 15, 2019
Ziwei Ji, Matus Telgarsky, Ruicheng Xian

Access Paper or Ask Questions