Page Not Found
Page not found. Your pixels are in another canvas.
Sitemap
A list of all the posts and pages found on the site. For you robots out there is an XML version available for digesting as well.
Pages
About Navjot Kukreja
About me
Page not in menu
This is a page not in th emain menu
Posts
portfolio
Portfolio item number 1
Short description of portfolio item number 1
Portfolio item number 2
Short description of portfolio item number 2
publications
talks
Automated fast finite difference computations
Published:
This was my first academic talk and the first time Devito was being presented to the world. The talk was about the first version of Devito - how it was built, and some performance results.
Symbolic Math for Automated Fast Finite Difference Computation
Published:
This was at the SIAM-CSE conference in Atlanta. My talk was part of a workshop called MS84 Domain-Specific Abstractions for Full-Waveform Inversion. Here I presented a high-level overview of Devito’s API.
Leveraging symbolic math for rapid development of applications for seismic imaging
Published:
Audience of ~200 people who were senior academics and executives from the Oil and Gas Industry. This talk was a sales pitch for the general idea of Devito - that end users should not be writing low-level code and should instead focus on high-level algorithmic questions.
Rapid development of seismic imaging applications using symbolic mathematics
Published:
This talk was at a smallish (30 people) workshop targeted at academia and the O&G Industry in the Middle East. I presented Devito - the core idea, some API examples, and some performance numbers.
Accelerating adjoint computations by combining checkpointing and compression
Published:
I gave this talk while visiting INRIA Bordeaux as part of a collaboration. This talk was based on my work on combining lossy compression and checkpointing. Here I presented the performance model that can predict whether lossy compression can accelerate adjoint computations.
High-level abstractions for checkpointing in adjoint-based optimisation
Published:
In this talk, I first presented pyRevolve - a high-level library for managing checkpointing for adjoint-based optimisation problems.
Lossy compression for checkpointing
Published:
In this talk I presented the work I had started while at Argonne. This was initial work about lossy compression of checkpoints.
Introduction to Devito
Published:
Here I conducted a three-day workshop teaching the attendees the basics of solving finite-difference problems using Devito.
Workshop on Full-Waveform Inversion
Published:
I conducted a full day workshop on Full-Waveform Inversion, demonstrating the mathematics and code required to invert some basic acoustic examples.
Training Neural Networks on the Edge: The why and the how
Published:
Here I presented the use of revolve-based checkpointing to enable training of neural networks on tiny devices. The use case presented was that of in-situ retraining of student-teacher pairs to address the viewpoint problem in computer vision.
Workshop on Software Technologies for Modelling and Inversion
Published:
This was a three-day workshop on Software technologies around seismic imaging. I did two hour-long presentations around my work in lossy compression and automatic differentiation.
Automatic differentiation of adjoint stencil loops
Published:
This talk focussed on automatic differentiation of stencil loops. A key feature of stencil loops is the gather operation that is converted into a scatter by conventional AD. This scatter can not be easily parallelised. This talk discussed a compiler-level transform that changes that scatter operation back into a gather operation that is easy to parallelise.
High-performance backpropagation for scientific computing
Published:
This talk was part of Supercomputing’s doctoral showcase. Here I presented the entirety of my thesis work, which consisted of checkpointing, lossy compression and automatic differentiation.
High-performance backpropagation for structured-grid solvers
Published:
This talk was based on my doctoral thesis work on checkpointing, lossy compression and automatic differentiation. Delivered this talk remotely from London.
teaching
Programming for Android
Professional Masters module, SENAI CIMATEC, 2015
My first teaching experience and it was in Brazilian Portuguese. Class size of 30, delivered in person. Developed all the teaching material myself.
Software Carpentry
Workshop, Imperial College London, 2019
Software Carpentry is a voluntary organisation that uses innovative research-informed teaching methods to teach software skills to scientists that are not computer scientists. I completed the Software Carpentry instructor training in 2017. I have taught the following modules since then:
Machine Learning (ACSE-8)
Masters module, Imperial College London, 2019
- Part of the MSc. Applied Computational Science and Engineering
- Total module contact hours: 36 hours
- I was a Graduate Teaching Assistant on this module this year before taking up lecturing responsibilities the following year
- Class size: 30 students, delivered in person.
Introduction to Machine Learning
Graduate school training, Imperial College London, 2019
- I developed the material from scratch, at request from the Graduate training program at Imperial College
- I delivered this mini course over two half-day sessions (6 hours).
- 30 PhD students from different departments. Delivered in person
Machine Learning (ACSE-8)
Masters module, Imperial College London, 2020
- Part of the MSc. Applied Computational Science and Engineering in the Department of Earth Science and Engineering at Imperial College London
- Taught half the lectures (18/36 hours)
- Module was planned to be delivered offline but had to be moved online at short notice.
- I developed 2/6 lectures, one assignment and the miniproject that was held as a Kaggle in-class competition.
- Class size of 90 + 30 PhD students from the department invited
Multicore and Multiprocessor Programming (COMP 528)
Masters module, Department of Computer Science, University of Liverpool, 2021
- Part of the MSc. Big Data and High-Performance Computing and MSc. Advanced Computing in the Computer Science Department at the University of Liverpool.
- Module coordinator as well as Lecturer from 2021 onwards.
- Average class size of 70 students.