Independent Researcher · Computer Science

Caxton Emerald S, PhD

Mechanistic interpretability, explainable artificial intelligence, and understanding the computations performed by neural networks.

Open to Postdoctoral Researcher and Research Scientist opportunities

About

I am an independent researcher with a PhD in Computer Science from Pondicherry University (awarded February 2026). My research investigates how artificial neural networks represent and approximate mathematical functions, with particular attention to minimal architectures, neuron-level computations, and model optimisation. I have developed research software for inspecting and visualising neural networks, and collaborated on applications of neural networks to colourimetric sensing.

My long-term objective is to connect theoretical understanding of learning systems with practical, efficient implementations. I am interested in learning across the full development process—from algorithms and experimentally validated models to specialised hardware and custom chip implementation—and in interdisciplinary questions involving biology, cognitive neuroscience, physics and astrophysics.

Research interests and directions

Established research

Understanding neural network mechanisms

Mechanistic interpretability; minimal neural architectures; function approximation; analysis of neurons, layers and learned transformations; explainable AI.

Model compression and optimisation, including inactive-neuron removal and neuron merging in simple feedforward architectures.

Research directions to explore

From explanations to efficient systems

Extending interpretability to deep networks and language models; causal analysis of representations and circuits; connections to computational and cognitive neuroscience.

Interpretable model-to-hardware co-design, neuromorphic computing, computational biology, quantum approaches to AI, and AI-enabled scientific discovery, including questions in astrophysics.

Selected peer-reviewed publications

  • Neural network optimization via inactive neuron removal and neuron merging in single input single output architectures.
    Caxton Emerald S & T. Vengattaraman · International Journal of Information Technology (2025)
  • NNVisualiser: An Emerging Framework for Visualising Artificial Neural Networks.
    Caxton Emerald S & T. Vengattaraman · IEEE ICDSAAI (2025)
  • Neural Network Coordinates: A Python Package for Studying Artificial Neural Networks.
    Caxton Emerald S & T. Vengattaraman · Proceedings of the International Conference on AI Systems and Sustainable Technologies (2026)
  • Interpretable Neural Network Design for Approximating the Generalized Absolute Value Function.
    S. Caxton Emerald & T. Vengattaraman · Springer ASSET 2025 proceedings (2026)
  • Approximating the Sine Function with the Universal Approximator: An Experimental Study with Visualisations.
    S. Caxton Emerald & T. Vengattaraman · Springer ICACS 2025 proceedings
  • Understanding the Role of Neurons and Layers in One-Dimensional Three-Class Classifiers.
    S. Caxton Emerald & T. Vengattaraman · Springer ICACS 2025 proceedings
  • Explainable Artificial Intelligence with Single Layer Feedforward Neural Network and Improved Crowned Porcupine Optimization Algorithm for Classification Problems.
    S. Caxton Emerald & T. Vengattaraman · Engineering, Technology & Applied Science Research 15(2), 21593–21598 (2025)
  • Bimetallic Silver-Copper Micro Flower Nanozyme-Based Colorimetric Sensor Assisted by Artificial Neural Network for Point-of-Care Uric Acid Detection.
    N. Alungal, P. Govindharaj, Caxton Emerald S & S. Kannan · Langmuir 42(37), 27381–27394 (2026)
  • Au@CeO₂ nanozyme based smart colourimetric sensor for cholesterol: A neural network powered point of care solution model.
    P. Govindharaj, N. Alungal, Caxton Emerald S & S. Kannan · Biochemical Engineering Journal 225, 109908 (2026)

Additional publications and research outputs are available through Google Scholar and ORCID.

Research software, models and demonstrations

Python research package

NeuralNetworkCoordinates

Provides coordinates for intermediate neural-network transformations to support neuron- and layer-level analysis and visualisation.

Python research package

NNVisualiser

Visualises neural network computations and intermediate transformations, including flow plots, using Matplotlib and coordinates from NeuralNetworkCoordinates.

Interactive model explorer

Compact ANN model collection

Explore trained neural networks approximating mathematical functions, including models associated with the JeevanRuth optimisation methodology.

Published Research · Biomedical AI

Cholesterol Concentration Predictor

An interactive research application demonstrating the prediction of cholesterol concentration from RGB colourimetric measurements using artificial neural networks and linear regression.

Associated with research on an Au@CeO₂ nanozyme-based colourimetric biosensor, the application supports comparison of neural network models trained using different hidden-layer configurations, activation functions, and random initialisations.

The published study investigated approximately 2,500 ANN models and reported a best validation mean squared error of 0.0577.

Publication: Govindharaj, P., Alungal, N., Emerald, C. S., and Kannan, S. (2026). Biochemical Engineering Journal, 225, 109908.

Research demonstration only. Not intended for clinical diagnosis or medical decision-making.

Interactive scientific application

Uric acid colourimetric ANN demo

Demonstrates ANN-based estimation of uric acid concentration from smartphone RGB measurements with blank correction and model comparison.

The sensing applications are research demonstrations. Their outputs should not be used as a substitute for validated clinical measurements or medical diagnosis.

Education and experience

Education

PhD in Computer Science
Pondicherry University · Awarded 6 February 2026

Thesis: Minimal Architecture Conjectures and Optimisation for Mechanistic Interpretability of Low-Dimensional Universal Approximator ANNs.

Integrated M.Sc. in Computer Science
Pondicherry University · 2015
CGPA: 9.57/10

Academic and industry experience

Assistant Professor, Department of Statistics and Data Science, CHRIST (Deemed to be University), Bengaluru · Oct 2025 – Sep 2026

Assistant Professor, Theivannai Ammal College for Women · Apr 2018 – Apr 2019

Software Development Engineer, HTC Global Services · Dec 2016 – Mar 2018

Guest Faculty, Pondicherry University · Nov 2015 – May 2016

Contact and research profiles

I welcome enquiries concerning postdoctoral positions, research scientist opportunities, scientific collaboration, and research software.

For professional enquiries, please contact me through LinkedIn.