Kumar, Akshi. 2025. 'From Large to Small: The Rise of Small Language Models (SLMs) in Text Analytics'. In: Online Faculty Development Programme Text Analytics with Deep Learning. National Institute of Technology (NIT), Patna (Online), India 1 - 9 March 2025. [Conference or Workshop Item]

	Text From LLM to SLM.pdf - Presentation Available under License Creative Commons Attribution Non-commercial. Download (31MB)
	Text 1740633389080.pdf - Supplemental Material Download (411kB)

Abstract or Description

This document presents the inaugural lecture delivered during the Online Faculty Development Programme (FDP) on "Text Analytics with Deep Learning" held on 1st March. The lecture explores the transition from Large Language Models (LLMs) to Small Language Models (SLMs), emphasizing efficiency, adaptability, and domain-specific intelligence in text analytics. It provides a comparative analysis of LLMs and SLMs, highlighting their respective advantages in computational cost, deployment, and real-world applications.

The session discusses key SLM methodologies, including pruning, quantization, and knowledge distillation, which optimize model size and performance while maintaining accuracy. It also addresses fine-tuning strategies, ethical considerations such as bias and fairness, and the importance of explainability in AI for regulated industries like healthcare and finance.

Furthermore, the lecture examines the evolving landscape of text analytics, illustrating how SLMs are revolutionizing AI-driven applications, from real-time translation and customer service to on-device AI and IoT systems. The discussion underscores the future trajectory of SLMs, positioning them as sustainable, privacy-friendly, and efficient alternatives to LLMs.

This session aims to equip faculty members with insights into the practical applications of deep learning in text analytics and the growing relevance of SLMs in AI-driven decision-making.

Item Type:	Conference or Workshop Item (Keynote)
Departments, Centres and Research Units:	Computing
Dates:	Date Event 1 March 2025 Completed
Event Location:	National Institute of Technology (NIT), Patna (Online), India
Date range:	1 - 9 March 2025
Item ID:	38523
Date Deposited:	03 Mar 2025 11:41
Last Modified:	05 Mar 2025 18:29
URI:	https://research.gold.ac.uk/id/eprint/38523

View statistics for this item...

Edit Record (login required)