Seminar talk by Dr. Alexander Lim (Online)

【Date&Time】
6 February, 2026 (Friday) 13:30-14:30
Admission Free, No Booking Necessary
【Place】
Zoom
Link: https://us06web.zoom.us/j/81105467072?pwd=oK1av7JOKZVoPpsU3MwNIciNObXmmo.1
Meeting ID: 811 0546 7072
Passcode: 528457
【Speaker】
Alexander Lim
【Title】
Faithful-Newton Framework: Bridging Inner and Outer Solvers for Enhanced Optimization
【Abstract】
Inexact Newton methods are widely used for unconstrained minimization due to their exceptional empirical performance. Unfortunately, this empirical efficiency does not carry over to their theoretical worst-case convergence guarantees.
Moreover, in large-scale settings, the high computational cost of solving the inner linear system at every iteration of inexact Newton’s methods remains a significant bottleneck. In this talk, we will address these challenges by introducing Faithful-Newton, a framework that tightly integrates the inner solver with the outer minimization process. We demonstrate that this integration improves upon the inexact Newton’s methods in terms of both the theoretical and empirical performance. In particular, under strongly convex and Hessian Lipschitz setting, Faithful-Newton can achieve either a superlinear rate or a condition-number-independent linear rate globally.
【Bio】
Alexander (Alex) Lim completed his PhD in Mathematics at the University of Queensland (UQ), under the supervision of Prof. Fred Roosta. His research focuses on developing and analysing second-order type optimization methods for large-scale problems. Prior to his PhD, he obtained a Bachelor of Mathematics from UQ in 2020 and a Bachelor of Arts (Honours), major in philosophy of mathematics and physics, from the University of Tasmania in 2016.