Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{206408,
        author = {Shrikant. S. Bhagat and Prathmesh. S. Kolte},
        title = {Techniques in Machine Learning after Large Language Model Integration: A Comprehensive Survey of Emerging Paradigms, Methods, and Open Challenges},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {13},
        number = {2},
        pages = {1241-1250},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=206408},
        abstract = {The emergence of Large Language Models (LLMs) has fundamentally reorganized the theory and practice of machine learning (ML), shifting the field from narrow, task-specific model training toward general-purpose foundation models that are adapted, augmented, and orchestrated rather than built from scratch. This survey provides a comprehensive and structured account of the techniques that have emerged, matured, or been substantially redefined as a direct consequence of LLM integration into ML pipelines. We organize the surveyed techniques into five functional categories: (i) adaptation techniques that specialize a frozen or lightly modified LLM for a task, including prompt engineering, in-context learning, and parameter-efficient fine-tuning methods such as LoRA, QLoRA, and adapters; (ii) knowledge-augmentation techniques, principally retrieval-augmented generation (RAG) and its graph and hybrid-search variants; (iii) alignment techniques, including supervised fine-tuning, reinforcement learning from human feedback (RLHF), direct preference optimization (DPO), and constitutional AI approaches; (iv) reasoning and orchestration techniques, including chain-of-thought prompting, self-consistency, tree and graph-of-thought search, and agentic tool-use frameworks built on planning-execution-observation loops; and (v) efficiency techniques, including knowledge distillation, quantization, pruning, and model merging, which make LLM-derived capability deployable under real-world latency, memory, and cost constraints. For each technique we describe its underlying mechanism, mathematical or algorithmic intuition, representative use cases, and trade-offs relative to classical ML approaches. We further present a multi-dimensional comparative analysis across computational cost, data requirements, controllability, and interpretability; a discussion of evaluation methodologies specific to generative and agentic systems; a survey of applied case studies drawn from enterprise, educational, and public-service domains; and a detailed treatment of open challenges including hallucination, evaluation validity, security vulnerabilities such as prompt injection, bias propagation, and the environmental cost of large-scale training and inference. The paper concludes with a research agenda covering hybrid symbolic-neural reasoning, continual and incremental post-training, standardized agentic benchmarks, and sustainable small-model specialization. This survey is intended as a rigorous reference for students, researchers, and practitioners transitioning from classical ML system design to LLM-integrated architectures.},
        keywords = {Large Language Models, Retrieval-Augmented Generation, Parameter-Efficient Fine-Tuning, LoRA, QLoRA, Reinforcement Learning from Human Feedback, Direct Preference Optimization, Prompt Engineering, Chain-of-Thought Reasoning, Agentic AI, Multimodal Learning, Knowledge Distillation, Model Quantization, Model Merging.},
        month = {July},
        }