Leading a team of 4 students on a pioneering research and engineering project
on real-time 3D multi-person human pose estimation using a 360° camera and
three mmWave radar sensors.
Developed a pioneering real-time 3D multi-person human pose estimation
algorithm using a 360° camera and three mmWave radar sensors combining
traditional computer vision techniques and deep learning.
2022-05 - 2022-08
Causal Reasoning Research Internship
University of Southampton, Southampton, United Kingdom
Advisor: Dr. Srinandan Dasmahapatra
Keywords: Causal reasoning, Causal Inference, Causal Discovery, Causal
Representation Learning, Neural-Causal AI, Machine Learning, Deep Learning
Conducted literature research on Causal Machine Learning, Causal
Representation Learning, Causal Discovery and Neural-Causal AI
Developed a score-based, iterative and Gumbell-softmax-based continuous
optimisation causal discovery algorithm under the assumption of Gaussian
structural equation models with equal error variances. Due to the latter
assumption, no interventional data is necessary
Developed a variational inference Expectation-Maximisation-like causal
discovery algorithm from unknown interventions using an LSTM as an
autoregressive distribution over the space of DAGs
Implemented the above algorithms using the Pytorch library
Presented my work at the ECS research seminar on 30 September 2022
2021-09 - 2022-05
Bachelor's Dissertation
University of Southampton, Southampton, United Kingdom
Title: "CC3DVAEWGAN: a controllable conditional 3D point cloud VAE-WGAN
based on shrinking layers and extending layers"
Advisor: Dr. Jonathon Hare
Keywords: Computer Vision, Deep Learning, Geometric Deep Learning
Proposed a novel two-stage framework (CC3DVAEWGAN) based on Shrinking
Layers and Extending Layers to achieve conditional and controllable
generation of point clouds
Proposed an innovative architecture combining a VAE and a WGAN, named
VAE-WGAN, capable of learning a latent space distribution for each training
class and a function from a latent vector to a point cloud blueprint
Proposed two novel architectures, named Extending Layer and Stack Extending
Layer, capable of exploring self, local and global correlations between points in
a semantic-based manner for the generation of point clouds
Implemented the proposed architectures and their training procedures using
the Pytorch and Pytorch Geometric libraries. The implementation supports multi-
GPU utilisation on cluster computing
Conducted experiments to verify the validity of the proposed solutions
2021-07 - 2021-09
Deep Learning Research Internship
Mainz University of Applied Sciences, Mainz, Germany