Headshot of Allamaprabhu Ani, smiling, brown background

Allamaprabhu Ani

ಅಲ್ಲಮಪ್ರಭು ಅಣಿ

PhD candidate in computational mechanics and scientific machine learning at City St George's, University of London. I build differentiable simulation systems for fracture, inverse analysis, and AI for engineering.

About Me

I develop matrix-free, differentiable solvers for PDEs at the intersection of scientific computing and machine learning. I am completing my thesis while continuing the validation and development of PhAST.

I created and maintain an open tutorial series covering PyTorch, neural networks, convolutional neural networks, and Physics-Informed Neural Networks (PINNs). The tutorials were developed to assist Dr Sathiskumar Anusuya Ponnusami's teaching at Queen Mary University of London and are maintained here for wider use.

Before my PhD, I completed an MTech at the National Institute of Technology Silchar and founded Aeroknacks, where I built aerospace structural-analysis automation tools based on established hand-calculation references, including Bruhn, Niu, Boeing Design Manuals, and ESDU data.

In 2025, I worked with EPFL's Computational Solid Mechanics lab and Prof. Jean-François Molinari. That collaboration contributed to our review of machine learning for fracture mechanics.

Outside the lab, I've practiced Hindustani classical singing since I was 3. The ragas are familiar; the riyaz is the lifelong part.

Research Software & Projects

Machine Learning Tutorials

Runnable notebooks and visual explainers for the path I keep using in my own work.

  1. Noisy sine data with train and validation split and a trained MLP fit

    01 intro 45 min

    Machine Learning Training from Scratch: Loss, Gradients, and Overfitting

    Reproduce the classic train-loss-down, validation-loss-up overfitting curve and learn what fixes it.

  2. Pearson correlation heatmap for the UCI auto-MPG regression dataset

    02 intro 40 min

    scikit-learn Regression Tutorial: Explore, Fit, Evaluate, Diagnose

    Build the full tabular ML loop and see why a pretty regression line is not enough.

  3. One-neuron and MLP models fitting a sine wave in PyTorch

    03 intro 55 min

    Neural Networks from One Neuron to a PyTorch MLP Classifier

    See exactly when one neuron fails, why a hidden layer works, and how the same pattern becomes a classifier.

  4. MLP versus CNN training loss and test accuracy on CIFAR-10

    04 intermediate 60 min

    Convolutional Neural Networks from Pixels to Feature Maps

    Train a small CNN, compare it with an MLP, then open the model and inspect the filters.

  5. Data-only neural network, physics-only PINN, and hybrid PINN on a damped oscillator

    05 advanced 70 min

    PINN Tutorial in PyTorch: Damped Oscillator, Autograd, and Inverse Problems

    Use torch.autograd.grad to make a neural network obey an ODE, then turn the same idea into an inverse problem.

Open the full tutorial path →

Recent Updates

Publications & Recognition

  1. A. S. Ani, J.-F. Molinari, G. Subhash, S. A. Ponnusami. A matrix-free, differentiable PyTorch solver for phase-field fracture: Formulation, benchmarks, and inverse analysis. arXiv:2606.23458 [cs.CE], 2026. Preprint introducing the matrix-free PyTorch phase-field fracture solver, dynamic and quasi-static benchmarks, and gradient-based inverse recovery of fracture energy.
  2. A. S. Ani, R. Nakka, G. Subhash, J.-F. Molinari, S. A. Ponnusami. Machine learning for computational fracture and damage mechanics: status and perspectives. Engineering Fracture Mechanics, Vol 332, Art 111778, 2026. Lead-authored review of machine learning methods, validation needs, and open research questions across computational fracture and damage mechanics.
  3. A. S. Ani, A. B. Deoghare. Leveraging machine learning for enhanced fatigue life prediction in aluminium alloys. Lecture Notes in Mechanical Engineering, Dec 2024.