Publications
Can Small Language Models Handle Context-Summarized Multi-Turn Customer-Service QA? A Synthetic Data-Driven Comparative Evaluation
Customer-service question answering (QA) systems increasingly rely on conversational language understanding. While Large Language Models (LLMs) achieve strong performance, their high computational cost and deployment constraints limit practical use in resource-constrained environments. Small Language Models (SLMs) provide a more efficient alternative, yet their effectiveness for multi-turn customer-service QA remains underexplored, particularly in scenarios requiring dialogue continuity and contextual understanding. In this study, we evaluate whether instruction-tuned SLMs, fine-tuned using parameter-efficient finetuning, can effectively handle context-summarized multi-turn customer-service QA while preserving contextual consistency, response quality and task relevance under computational constraints. We further investigate instruction-tuned SLMs for context-summarized multi-turn customer-service QA using a history summarization strategy to preserve essential conversational state and introduce a conversation stage-based qualitative analysis to evaluate model behavior across different phases of customer-service interactions. The main contributions of this work include the application of parameter-efficient fine-tuning to adapt SLMs for context-summarized multi-turn customer-service QA, a synthetic data construction pipeline for generating a context-summarized multi-turn QA dataset, and a structured evaluation framework combining quantitative metrics with human and LLM-as-a-judge assessments for customer-service QA evaluation. Nine instruction-tuned SLMs are evaluated against three commercial LLMs using lexical and semantic similarity metrics alongside qualitative assessments, including human evaluation and LLM-as-a-judge methods. Results show notable variation across SLMs, with some models demonstrating near-LLM performance, while others struggle to maintain dialogue continuity and contextual alignment. These findings highlight both the potential and current limitations of low-parameter language models for real-world customer-service QA systems.
Monocular Depth Estimation for Autonomous Driving: A Survey on Deep Learning Methods, Multi-Task Learning and Robustness in Real-World Conditions
Comprehensive survey covering supervised, self-supervised, transformer-based, and multi-task depth estimation methods for autonomous driving systems. The survey examines recent advances in convolutional neural networks, vision transformers, and attention mechanisms for monocular depth prediction. Includes detailed KITTI benchmark analysis and evaluation protocols under standard and adverse conditions (rain, fog, night driving).
From LLMs to SLMs: Advancements, Challenges and Future Directions of Language Models for Customer Interactions
Over the past decade, digital customer services have become increasingly common, spanning automated call centres, chat applications, email support and social media. The advancements in natural language processing have made it possible to migrate from rule-based approaches to conversational agents that can interpret and reason about user queries. This paper discusses language model research in customer service mainly based on client-agent question and answering implementations. Furthermore, usage in customer service scenarios in tasks such as intent detection, customer sentiment analysis and dialogue summarisation will be discussed. State-of-the-art Large language models (LLMs) like GPT-4, Claude and Gemini have showcased high performance in customer service applications. However, LLMs require substantial computational resources and frequently rely on cloud APIs, creating concerns about latency and privacy. Small language models (SLMs), ranging from one to ten billion parameters, address these issues by combining pre-training with parameter-efficient fine-tuning using methods such as LoRA and QLoRA. When coupled with retrieval-augmented generation and reinforcement learning from human or AI feedback methods, SLMs can approach the accuracy of LLMs in domain-specific tasks while supporting on-premise deployment and low-latency inference. This review is distinctive in that it surveys both LLMs and SLM-based research in customer service, current SLM families, methods that can be used for domain adaptation,current implementation of customer services in several domains and methods used to evaluate language models in customer service tasks. It also highlights critical gaps, particularly limited evaluations of SLMs against LLMs and proposes recommendations for developing efficient, secure and domain-aware customer service models.
Deep Federated Learning: A Systematic Review of Methods, Applications, and Challenges
Federated Learning (FL) represents a paradigm shift in machine learning, enabling collaborative model training on decentralized data while preserving user privacy. However, the transition from theory to real-world application is impeded by significant challenges, including high communication costs, statistical and system heterogeneity and persistent privacy vulnerabilities. These barriers critically limit the performance, scalability and security of FL systems. This paper provides a systematic review of the state-of-the-art solutions developed to address these fundamental obstacles. The review analyzes core methodological advancements, including advanced model aggregation methods, techniques to enhance communication efficiency such as model compression and decentralized training and strategies to combat statistical heterogeneity arising from non-IID data. Furthermore, it delves into emerging paradigms like Federated Meta-Learning and Federated Reinforcement Learning, alongside advanced architectural models such as hierarchical and blockchain-based systems. The practical impact of these advancements is contextualized through a review of key application domains, including healthcare, vehicular networks and the Internet of Things. A benchmark analysis is presented to assess the practical efficacy of these diverse techniques. In conclusion, this work synthesizes the critical trade-offs inherent in FL systems and highlights key directions for future research, offering a comprehensive guide for researchers and practitioners in this rapidly evolving field.