activity
20242026
collaborators

6 papers

cs.LG2026

Non-normal spectral signatures of instability in neural network training dynamics

Souvik Ghosh

Training instabilities in deep networks - loss spikes, oscillatory convergence, and gradient pathologies - are empirically prevalent but lack a rigorous operator-theoretic explanat…

math.FA2026

On anti-coproximinal and strongly anti-coproximinal subspaces of function spaces

Shamim Sohel, Souvik Ghosh, Debmalya Sain +1

The purpose of this article is to study the anti-coproximinal and strongly anti-coproximinal subspaces of the Banach space of all bounded (continuous) functions. We obtain a tracta…

math.FA2025

On symmetricity of the norm derivatives orthogonality in operator spaces

Souvik Ghosh, Kallol Paul, Debmalya Sain

We investigate -orthogonality and its local symmetry in the space of bounded linear operators. A characterization of Hilbert space operators with symmetric numerical range is e…

math.FA2025

Smoothness in the space of bounded linear operators on semi-Hilbert space

Somdatta Barik, Souvik Ghosh, Kallol Paul +1

Given a nonzero positive operator on a Hilbert space , a semi-inner product is naturally induced on . In this work, we introduce the notion of \emph{

math.FA2025

On some subspaces of vector-valued continuous function space, from the perspective of Best coapproximation

Souvik Ghosh, Kallol Paul, Debmalya Sain +1

This article explores anti-coproximinal and strongly anti-coproximinal subspaces in the spaces of vector-valued continuous functions and operator spaces. We provide a complete char…

math.FA2024

Orthogonality induced by norm derivatives : A new geometric constant and symmetry

Souvik Ghosh, Kallol Paul, Debmalya Sain

In this article we study the difference between orthogonality induced by the norm derivatives (known as -orthogonality) and Birkhoff-James orthogonality in a normed linear spac…