Advances in Trustworthy, Efficient, and Scalable Machine Learning: From Privacy to Telecom Applications

Abstract

Over the past several years, machine learning has rapidly expanded into domains in which sensitive and large-scale environments demand solutions that are not only accurate, but also privacy-aware, efficient, and adaptable. This talk explores a sequence of ideas aimed at making learning systems more trustworthy and more adaptable to real‑world constraints. First, I will talk about our contributions in the field of privacy-preserving machine learning, and machine unlearning.

Then we will revisit one of the most fundamental components of deep learning: learning rate scheduling. Unlike fixed or periodic policies, we present a convergence-aware scheduler that monitors training dynamics and adaptively increases the learning rate when stagnation is detected, enabling the optimizer to escape sharp local minima and achieve improved generalization. Next, I discuss the importance of small, efficient models for telecom network deployment, where constraints on latency, memory, and energy consumption are critical. I present a mixed-precision, neuron-level quantization framework that assigns different numerical precisions to individual neurons based on their sensitivity and contribution to model performance. Finally, I outline our ongoing work on tabular foundation models, motivated by the structured, large-scale data central to telecom systems, and discuss their potential for network intelligence and automation.

Biography

From kth.se:

"I am an industrial postdoctoral researcher at KTH and Ericsson Research in Sweden. My work focuses on distributed machine learning and unlearning, generative AI, and data privacy. Currently, my goal is to miniaturize foundation models for telecom use cases.

I earned my Ph.D. from Umeå University in the area of privacy-preserving machine learning, with a dissertation titled “Navigating Model Anonymity and Adaptability.” I hold an M.Sc. from South Asian University, India and a B.Sc. from the University of Delhi (ANDC), India."