MONASH DEEPNEURON

✦

MONASH DEEPNEURON ✦

HPC Projects.

Neural Cellular Automata

Project Manager: Keren Collins

Cellular automata is a simulation technique that involves an n-dimensional (usually 2D) grid of cells, each with a state. Each cell’s state is updated iteratively based on a “ruleset”. The project will explore the different behaviours of cellular automata that emerge from different rulesets and eventually replace these rules with CNNs.

Read the project blog here.

view the live project here

On-Hit

Project Lead: Nathan Lam, Jackie Nguyen

The goal of On-hit is to develop a foundational framework for AI action recognition that can be scaled across various applications. Boxing is a great starting platform because its dynamic nature pushes the limits of these models while still being confined within a small boundary area. By leveraging already existing AI pose estimation models, we can detect the skeletons of boxers and use this data to identify actions (e.g., punches, dodges)

Join the On-HIT team

Cluster Development

Project Lead: Anthony Oon

Team members will design and implement a custom HPC cluster workload management and scheduling software on a mini-cluster made of 4 Raspberry Pi 4 SBCs. This will first involve setting up remote access (similar to MDN workstation) and performing other mini-cluster improvements (automated infra management using Ansible, etc...).

Moving forward the team is exploring options in integrating different kinds of hardware into the existing cluster, as well as investigating different operating systems.

Astrophysical Simulation

Project Lead: Erol Cemiloglu

Developing a sophisticated physics engine for astrophysical simulations, enabling in-depth investigations into celestial bodies like planets, black holes, and stars. This includes enhancing simulation efficiency through parallelisation of complex tasks like fluid dynamics and gravitational interactions across computing clusters.

Parallel Clip Training

A HPC x AI Collaboration

Create a CLIP (model which connects related images and text) from scratch, testing HPC techniques to make training it faster. Over the Summer, this involves parallelising the training of a CNN over multiple CPUs/GPUs and investigating whether this speeds up DL model training & execution.

DDoS Dissipation

An investigation into Distributed Denial of Service attacks and how HPC techniques can be used to defend a web service under such attacks. A research paper published in the HPC Asia 2018 conference introduces FlexProtect - A DDoS protection architecture for a distributed computing system. Our aim is to explore and investigate how FlexProtect performs against DDoS attacks using HPC simulations.

Past HPC Projects

INDYscc-2022

DeepNeuron will be competing at IndySCC in November 2022 in order to win a place in the Student Cluster Competition (SCC), the world championship competition held at the International Supercomputing Conference.
Job Record analysis

In order to increase productivity of M3 MASSIVE users, we are devising a new way to view job records so that better, more informed decisions can be made when submitting their next job.
Horovod

Implementing Horovod on the Monash HPC cluster MASSIVE, and developing training and documentation to enable users to easily train their deep learning models.
Monitoring the cluster

Monitoring the cluster is a visualisation project to display interesting and important data about the Monash M3 Cluster.
ASC20-21 STUDENT SUPERCOMPUTER CHALLENGE

In ASC, students are required to work with hardware and assemble a HPC cluster that is capable of running a range of scientific applications.

read more
ASC21-22 STUDENT SUPERCOMPUTER CHALLENGE
DRONE PHOTOGRAMMETRY

Parallelising the photogrammetry pipeline and implementing a scalable, cloud based solution for point cloud analysis.
Ray Tracing Research

Simulate individual light rays to render high quality 3D assets realistically. Utilise HPC techniques to parallelise simulations and increase efficiency.