Dr. Keven Ren (He/Him)

Computational Astrophysicist
Data Scientist Aspiree
Machine Learning Enthusiast

Hi, I'm Keven, a Melbourne-based PhD graduate in astrophysics with a keen interest in data and deep learning.
My research uses simulations and modeling to investigate how rare early Universe quasars and galaxies are linked to the distribution of the underlying dark matter. I create semi-analytical models to predict how the distribution of galaxies and quasars evolve across cosmic time.
I've recently finished an internship on web-based data-visualisation with the Astronomy Compute and Data Services (ADACS) and am currently looking to find new work in data science/machine learning!
Please find my CV/Publications on the right or scroll down to read about a few of my recent data-related projects.

Key technologies: python (numpy/pandas/tensorflow/keras/scikit-learn)
Familiarity: javascript, React, django, SQL, Google Cloud Platform

Sep 2019 - Current

Segmenting coral features with MultiRes-UNet

Pixel-based segmentation of the different stages of coral life using a MultiRes-UNet architecture (UNet with modified Inception-like blocks). Pretraining done on extensive labeled data of a tile taken on a single day, with subsequent fine-tuning using select sections of tiles taken on other days.
Model performance of the coral class yields an F1-score of 0.94, highlighting the future potential for this model to be used as passive means of monitoring coral health.

Key Technologies: python, numpy, tensorflow/keras

May 2022

BirdCLEF22 - A deep learning method to identify birds from their audio samples

Classification of bird species from audio samples using a ResNet50v2 architecture. Converting the audio spectrum into a spectrogram via a Short-time Fourier Transform suggests a CNN architecture could be used to extract features in the time-frequency domain. Various augmentation methods were employed such as white/pink noise and audio mixup to both facilitate generalization and combat class imbalance from underrepresented species.
Due to a lack of computing resources, our team was not able to experiment as much as we would've liked. Additional details for this project can be found below.

Key Technologies: python, numpy, pandas, tensorflow/keras

Aug 2021 - Nov 2021

Interactive web-based visualisation for large-scale astronomical data products

Creating a new React-based plotting component to replace the previous bokeh-based charting tool. Current React-based libraries do not offer the functionality that we require, such as 2d-histogram, contour plots or corner plots. The only library that did react-vis, was deprecated in the late-2020.
I have constructed the desired plotting components using the popular recharts library as a base, and have also coded the necessary pipeline integrating between simulation and the django backend, as well as django to the ReactJS frontend.
One of the coded components can be accessed through the github given below.

Key Technologies: python, django, javascript, react, recharts

2020 - 2021

Semi-empirical model of the evolution of quasar densities across cosmic time

Quasars are luminous objects powered by the accretion of cold gas onto the central supermassive black hole. The population of extremely high-redshift quasars (z > 6) are not well understood due to the inherent rarity of such objects. Knowing how this population evolves across time would provide insight to growth mechanisms of supermassive black holes during the early Universe.
Motivated by the most recent observations for the population of high-redshift quasars, we develop a novel semi-empirical model with only a single free parameter, the quasar duty cycle, and the remaining parameters being fixed by calibration.
We provide functional forms for the population distribution and forecast the expected number of objects observed for an upcoming wide-area sky survey.

Key Technologies: python, analytical model building, MCMC

Socials