Avatar

Freddy Jiang

I am a research engineer at the Applied AI Lab, The Trade Desk. I graduated from Imperial College London with a master's degree in Artificial Intelligence and a bachelor's degree in Joint Maths and Computer Science.

I am passionate about all things AI and data science. I love to write code as I consider it a form of art. I also love learning, building and automating things with new tools and technologies.

Portfolio

NVAE Medical Imaging

Cardiac Shape Analysis with Nouveau Variational Autoencoder

2024. Master's dissertation. Supervised by Prof. Loïc Le Folgoc, Prof. Elsa Angelini. Winton Capital Applied Computing MSc Prize issued by Department of Computing, ICL.

paper / code / models will be made available mid-2025

I demonstrated that Nouveau VAE, a state-of-the-art hierarchical non-autoregressive generative framework, can learn from clinically annotated cardiac segmentation masks to encode anatomic shapes efficiently. As part of this investigation, I proposed a novel metric, the Fréchet ResNet Distance with SimCLR, for measuring the similarity between synthetic and real cardiac masks. I used the learned embeddings as an anatomical constraint to improve U-Net segmentation performance, which also generalises well in domain adaptation.

VARIATIONAL AUTOENCODER

MEDICAL IMAGING

PYTORCH

PYTHON

SimCLR Medical Imaging
Evaluating SimCLR for Medical Image Classification

2023. Bachelor's dissertation. Supervised by Prof. Ben Glocker.

paper info repo zenodo

I explored SimCLR, a state-of-the-art self-supervised learning framework, for classifying a wide range of medical imaging modalities, including colon pathology, dermatology, blood cells and retina fundus. I proposed a novel augmentation sequence and other modifications that yielded up to 30.6% improvement in accuracy over baseline supervised methods.

CONTRASTIVE LEARNING

MEDICAL IMAGING

PYTORCH

PYTHON

LLMs for PCL
LLMs for Detecting PCL

2024. ICL NLP coursework. Group of 2.

paper info repo zenodo

We devloped an ensemble model of finetuned RoBERTa and DistilBERT models to detect Patronising and Condescending Language in text (subtask 1), achieving strong performance and attaining full marks in the coursework.

LLM

ROBERTA

PYTORCH

PYTHON

CYOA

Create Your Own Adventure - Automated Gamebook Generation

2022. ICL group project. Distinguished project recognised by Department of Computing, ICL.

Create Your Own Adventure is a web application that offers an extensive set of tools for authors to easily create, fully customise and print their gamebooks.

REACT

NEXT.JS

TYPESCRIPT

GPT

SBERT

Peer Assessment
Peer Assessment Platform

2021. Internship project at DoC EdTech, ICL.

The Peer Assessment web application is used by the Department of Computing at Imperial College. It allows team members to evaluate each other's contributions and provide anonymous feedback. It provides staff with ability to view and moderate submissions, as well as automatically generate and persist form templates and reports.

FLASK

PYTHON

POSTGRESQL

DoC EdTech Site
DoC EdTech, Imperial College Website

2021. Volunteer work.

I designed and developed the official site for the Department of Computing's Educational Technology team at Imperial College London.

JEKYLL

BOOTSTRAP 5

HTML

SCSS

Data Animations
Data Animations App

2023. Hobby project.

A lightweight Python application that allows you to render highly customisable bar chart animations from CSV files into HD videos.

PYGLET

PYTHON

PANDAS

Duel!
Duel! - 2P Shooter Game

2020. Game jam submission.

Duel! is a local 2-player platformer shooter game where players fight to the death on a small, floating island. The game features weapons and a rewind mechanic to surprise your opponent. Developed in 1 week for Brackeys Game Jam 2020.2.

UNITY

C#

Ten Minutes of Existence
Ten Minutes of Existence - Puzzle Adventure Game

2020. Game jam submission.

Ten Minutes of Existence is a single-player puzzle adventure game where you play as an ephemeral entity and explore the underwater world. The goal is to collect 5 gems while avoiding obstacles and other entities, to unlock a passage to the final destination before time runs out. Developed in 1 month for GameDev.tv Community Jam.

UNITY

C#

SHADERLAB

View more projects
Portal