From Epic to Dramatic Tradition: Literary Contribution of Poet Kalidas – A Critical Study

1st Author

• Name: Piyush Junghare

• Department: CSE (Cyber Security)

• Institution: GHRCEM, Nagpur, India

• Email: piyushjungharecyb@ghrietn.raisoni.net

2nd Author

• Name: Jay Chheniya

• Department: CSE (Cyber Security)

• Institution: GHRCEM, Nagpur, India

• Email: jaychheniyacyb@ghrietn.raisoni.net

3rd Author

• Name: Mohit Behare

• Department: CSE (Cyber Security)

• Institution: GHRCEM, Nagpur, India

• Email: mohitbeharecyb@ghrietn.raisoni.net

4th Author

• Name: Pratik Kashte

• Department: CSE (Cyber Security)

• Institution: GHRCEM, Nagpur, India

• Email: pratikkashtecyb@ghrietn.raisoni.net

5th Author

• Name: Saniya Belekar

• Department: CSE (Cyber Security)

• Institution: GHRCEM, Nagpur, India

• Email: saniya.belekar.cyb@ghrietn.raisoni.net

6th Author

• Name: Vaishnavi Dhoble

• Department: CSE (Cyber Security)

• Institution: GHRCEM, Nagpur, India

• Email: vaishnavi.dhoble.cyb@ghrietn.raisoni.net

7th Author

• Name: Shalini Kumari

• Department: CSE (Cyber Security)

• Institution: GHRCEM, Nagpur, India

• Email: shalini.kumari@ghrau.edu.in

Abstract :Abstract This paper presents a comprehensive review of Google's Titans architecture, a groundbreaking neural memory framework designed to address the limitations of traditional Transformer models in handling long-term context. Titans introduces a novel neural long-term memory module that learns to memorize at test time, enabling language models to retain, update, and recall information over millions of tokens without the quadratic computational cost associated with standard attention mechanisms. The architecture combines three interconnected memory systems: short-term attention-based memory, a deep neural long-term memory module, and persistent memory for task-specific knowledge. This review examines the theoretical foundations, architectural components, and design principles that enable Titans to achieve superior performance on long-context tasks while maintaining computational efficiency. The paper provides detailed analysis of the three architectural variants Memory as Context (MAC), Memory as Gate (MAG), and Memory as Layer (MAL) and discusses their respective strengths and trade-offs in different application scenarios.

Keywords: Deep learning, long-term memory, neural architecture, sequence modelling , test-time learning, Transformers.