Yair Carmon

Accelerated Methods for NonConvex Optimization

We present an accelerated gradient method for nonconvex optimization problems with Lipschitz continuous first and second derivatives. In a time

Partial Similarity of Shapes Using a Statistical Significance Measure

O(\epsilon^{-7/4} \log(1/ \epsilon) )

Information, Estimation, and Lookahead in the Gaussian Channel

, the method finds an

Comparison of the Achievable Rates in OFDM and Single Carrier Modulation with I.I.D. Inputs

\epsilon

Disproof of the Shamai-Laroia conjecture

-stationary point, meaning a point

Lower Bounds and Approximations for the Information Rate of the ISI Channel

x

On information, estimation and lookahead

such that

The role of lookahead in estimation under Gaussian noise

\|\nabla f(x)\| \le \epsilon

Gradient Descent Efficiently Finds the Cubic-Regularized Non-Convex Newton Step.

. The method improves upon the

Convex Until Proven Guilty: Dimension-Free Acceleration of Gradient Descent on Non-Convex Functions.

O(\epsilon^{-2} )