Loading article...
Loading article...
Stream/Area:
-
Subject:
-
Paper ID:
IJSREM68859
Volume:
10 |
11 |
Nov 2026
Pages:
1-24
Publication Frequency:
Monthly (12 issues per year)
License:International Journal of Scientific Research in Engineering and Management follows the Creative Commons Attribution 4.0 (CC BY 4.0) License, allowing free use and distribution with proper attribution.
Department of Computer Science
Vidyapeeth Junior College
Artificial intelligence (AI) systems appear complex, yet their core operations rest on a small set of mathematical ideas that are accessible to students with Class 11–12 mathematics. This narrative review explains those ideas and shows where each one appears in practice. Vectors and matrices represent data and parameters; derivatives and gradients measure how a model’s error changes; optimization uses gradients to learn; probability and statistics describe uncertainty and prediction; eigenvalues and principal component analysis compress information; and entropy and cross-entropy supply the loss functions of classification and language modelling. The review then shows how these tools combine in embeddings, vector search, retrieval-augmented generation and Transformer attention. Each topic follows one pattern: concept, formula, symbol definitions, a fully worked numerical example, an AI application and a statement of limitations. All calculations were recomputed independently, and advanced material is deliberately simplified or labelled as an extension. The paper is not a systematic review and reports no original experiments.
KEYWORDS
Main insights. This review showed that the working of modern AI can be followed with school-level mathematics. Data become vectors and matrices (Section 2, Section 3). Calculus converts the question “how wrong is the model, and which way should it change?” into a gradient, and the chain rule makes that gradient computable layer by layer (Section 4). Gradient descent repeats small steps along the negative gradient, and its behaviour depends strongly on the learning rate (Section 5). Probability turns scores into predictions and measures uncertainty (Section 6), eigenvectors of the covariance matrix give the principal directions of data (Section 7), and cross-entropy gives the standard loss for classification and language models (Section 8). Embeddings, retrieval-augmented generation and Transformer attention are then compositions of these same tools (Section 9). Why mathematical understanding matters. A reader who knows these ideas can interpret what a model’s outputs mean, anticipate why training may fail (for example, with a learning rate that is too large), judge the limits of similarity scores and probabilities, and read technical papers without treating the equations as a barrier. Mathematical understanding also supports responsible use: a probability is an estimate, a similarity score is not proof of relevance, and a fluent answer is not a verified one. Opportunities for further study. Natural next steps are: singular-value decomposition as a generalisation of PCA; convergence theory and adaptive optimisers; Bayesian inference and maximum-likelihood estimation in depth; the mathematics of convolutional and recurrent networks; and graph-based, reinforcement-learning and generative models. Practising each worked example with different numbers, and then implementing it in a short program, is the most reliable way to consolidate the material.
[1] I. Goodfellow, Y. Bengio, and A. Courville, Deep Learning. Cambridge, MA, USA: MIT Press, 2016. [2] C. M. Bishop, Pattern Recognition and Machine Learning. New York, NY, USA: Springer, 2006. [3] K. P. Murphy, Machine Learning: A Probabilistic Perspective. Cambridge, MA, USA: MIT Press, 2012. [4] T. Mikolov, K. Chen, G. Corrado, and J. Dean, “Efficient estimation of word representations in vector space,” in Proc. International Conference on Learning Representations (ICLR), 2013. [5] J. Pennington, R. Socher, and C. D. Manning, “GloVe: Global vectors for word representation,” in Proc. 2014 Conf. Empirical Methods in Natural Language Processing (EMNLP), pp. 1532–1543, 2014. [6] J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “BERT: Pre-training of deep bidirectional transformers for language understanding,” in Proc. 2019 Conf. North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pp. 4171–4186, 2019, doi: 10.18653/v1/N19-1423. [7] N. Reimers and I. Gurevych, “Sentence-BERT: Sentence embeddings using Siamese BERT-networks,” in Proc. 2019 Conf. Empirical Methods in Natural Language Processing and 9th Int. Joint Conf. Natural Language Processing, pp. 3982–3992, 2019, doi: 10.18653/v1/D19-1410. [8] H. Jégou, M. Douze, and C. Schmid, “Product quantization for nearest neighbor search,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 33, no. 1, pp. 117–128, 2011, doi: 10.1109/TPAMI.2010.57. [9] Y. A. Malkov and D. A. Yashunin, “Efficient and robust approximate nearest neighbor search using hierarchical navigable small world graphs,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 42, no. 4, pp. 824–836, 2020, doi: 10.1109/TPAMI.2018.2889473. [10] J. Johnson, M. Douze, and H. Jégou, “Billion-scale similarity search with GPUs,” IEEE Transactions on Big Data, vol. 7, no. 3, pp. 535–547, 2021, doi: 10.1109/TBDATA.2019.2921572.
Publication Date
11/10/2026
Peer Review Type
Double-Blind Peer Review
Review Rounds: 2
Documents
Article Views
4
Total views since publication
PDF Downloads
6
Total downloads since publication
Geographic Reach
5
Digital Preservation: This article is permanently archived on Zenodo and complies with all data preservation standards across major indexing platforms.
Plagiarism Screening
This manuscript has been checked for plagiarism using TurnitinPlagiarismChecker software.
Conflict of Interest
The authors declare no conflict of interest related to this research work.
Funding Information
No funding information available for this research.
Related research in the same subject area
Loading similar articles...