Blog

「What I have done in the past」
2026
Line Graphs: When Edges Become Vertices

Line Graphs: When Edges Become Vertices

Whitney's theorem, Beineke's forbidden graphs, and more

Take a graph $G$ and swap the roles of its vertices and edges: every edge becomes a node, and two of these nodes are linked whenever the edges they came from used to share an endpoint. That’s the line graph $L(G)$. It sounds like a cheap notational trick, but it turns out to be a real change of p...

Posted by Clement Wang on July 8, 2026

Verdant Depths

Verdant Depths

A roguelite dungeon crawler built at a game jam

Verdant Depths is a roguelite dungeon crawler I built in Python for a game jam in May 2026. Fight through 7 floors of procedurally generated forest ruins, collect perks and relics, and defeat a boss on each floor. Game trailer The game Clear all rooms on each floor, defeat ...

Posted by Clement Wang on June 4, 2026

Decoding Mouse Maze Position from Brain Signals

Decoding Mouse Maze Position from Brain Signals

A 48-hour proof of concept at Hacktion Potential

Last weekend I took part in Hacktion Potential, a 48-hour computational neuroscience hackathon at the Institut du Cerveau in Paris. Our team worked on decoding mouse maze position from brain signals: mapping high-dimensional neural activity to the (x,y) position of a mouse in a closed environment...

Posted by Clement Wang on March 2, 2026

2025
Delivery Date Estimation Model at Mirakl

Delivery Date Estimation Model at Mirakl

Building a Time Series Forecasting Model for Marketplace Orders

Mirakl builds enterprise marketplace software, letting retailers, manufacturers, and B2B companies launch and run their own online marketplaces. Joining Mirakl I joined Mirakl as a Data Scientist in December 2024, and got put on one of the company’s machine learning initiatives: estimating deli...

Posted by Clement Wang on October 7, 2025

For How Much?

For How Much?

A party game we built in 48 hours during a weekend in Chartres

The story This one’s a bit different from my usual posts. Sometimes you just do something ridiculous with friends and see what happens. I had my permanent job for already 6 months as a data scientist and I’d been feeling the need to actually build something, so two friends and I planned a weeke...

Posted by Clement Wang on May 18, 2025

2024
Training a Foundation Model for EEG Time Series at Beacon Biosignals

Training a Foundation Model for EEG Time Series at Beacon Biosignals ⭐⭐

Reaching over 80% Sleep Staging Accuracy with 5% of the annotated data

Project Overview This work was conducted during my Master’s thesis internship at Beacon Biosignals, an American company developing healthcare solutions for analyzing sleep brain activity. My research focused on training a foundation model for sleep EEG time series using self-supervised learning ...

Posted by Clement Wang on October 15, 2024

Nash Equilibrium Finding at Google DeepMind

Nash Equilibrium Finding at Google DeepMind ⭐⭐

Accelerating FoReL Convergence with Population-Based Methods

Project Overview During my final year at CentraleSupélec, I had the opportunity to collaborate with Google DeepMind on a research project (Oct 2023 – Apr 2024) focused on improving the convergence of algorithms for computing Nash equilibria in two-player zero-sum games. Finding Nash equilibria ...

Posted by Clement Wang on May 15, 2024

Playing Flappy Bird with Reinforcement Learning

Playing Flappy Bird with Reinforcement Learning

Q-Learning and Monte Carlo Control for Text Flappy Bird

Project Overview As part of the Reinforcement Learning course by Stergios Christodoulidis, I implemented reinforcement learning algorithms to train an agent to play Text Flappy Bird (TFB), a simplified version of the classic game. The challenge was to design agents that could learn to survive as...

Posted by Clement Wang on April 15, 2024

Graph Neural Network Benchmark on Twitch Dataset

Graph Neural Network Benchmark on Twitch Dataset

Benchmarking Different GNNs on Multiple Classification and Regression Tasks

Project Overview This project was carried out as part of the Machine Learning on Network Science course taught by Fragkiskos Malliaros. Our goal was to benchmark several Graph Neural Network (GNN) architectures on the Twitch Gamers dataset, a large-scale social network of streamers. Previous st...

Posted by Clement Wang on April 15, 2024

Aspect-Based Sentiment Analysis

Aspect-Based Sentiment Analysis

Understanding Sentiment with Respect to Specific Aspects

Project Overview As part of the NLP course at Naver Labs Europe, we tackled aspect-based sentiment analysis. The task is to classify a sentence’s sentiment (positive / negative / neutral) with respect to a specific aspect-term (e.g., food quality, service, ambiance). This makes the task more ch...

Posted by Clement Wang on April 15, 2024

Prey Predator Simulation with MADDPG

Prey Predator Simulation with MADDPG ⭐

Multi-Agent Reinforcement Learning for Cooperative Behavior

Project Overview This project simulates a prey-predator ecosystem using multi-agent reinforcement learning. We apply the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm to learn cooperative behaviors among predators while preys try to escape. The environment is a 2D world wher...

Posted by Clement Wang on April 1, 2024

Averaging Weights Leads to Wider Optima and Better Generalization

Averaging Weights Leads to Wider Optima and Better Generalization

Experimenting with Stochastic Weight Averaging (SWA)

Project Overview This project was part of the Bayesian Machine Learning course taught by Rémi Bardenet. Together with Thomas Lemercier, I reimplemented the paper “Averaging Weights Leads to Wider Optima and Better Generalization” by Izmailov et al. (2019). The core idea of Stochastic Weight Ave...

Posted by Clement Wang on March 15, 2024

Lymphocytosis Classification

Lymphocytosis Classification

Detection of Tumoral Lymphocytosis from Blood Smears

Project Overview This project aimed to build a classifier to detect tumoral lymphocytosis using blood smear images and patient metadata. The dataset, collected at Lyon Sud University Hospital, included 204 patients (142 for training and 42 for testing). The goal was to support clinicians in iden...

Posted by Clement Wang on March 1, 2024

Implicit Acceleration by Overparameterization

Implicit Acceleration by Overparameterization

Stacking Linear Layers Improves Training?

Project Overview This project was part of the Theoretical Principles of Deep Learning course taught by Hedi Hadiji. Together with Guillaume Levy, I reimplemented the experiments of the paper: Arora, S., Cohen, N., & Hazan, E. (2018). On the Optimization of Deep Networks: Implicit Accelerat...

Posted by Clement Wang on February 15, 2024

ABBA Symbolic Representation of Time Series

ABBA Symbolic Representation of Time Series

Time Series Forecasting with Adaptive Brownian Bridge-based Symbolic Aggregation

Project Overview This project was part of the Machine Learning for Time Series course by Laurent Oudre. Together with Guillaume Levy, I reimplemented two papers on symbolic representations of time series, focusing on the ABBA method (Adaptive Brownian Bridge-based symbolic Aggregation). Our obj...

Posted by Clement Wang on January 15, 2024

Text-based Molecule Retrieval

Text-based Molecule Retrieval ⭐

Learning to Match Molecular Graphs with Text Queries

Project Overview This project, part of the Advanced Learning for Text and Graph Data course, tackled a challenging task: retrieve the correct molecular structure from a text description. We combined textual descriptions and molecular graphs using a contrastive learning framework, aligning repre...

Posted by Clement Wang on January 15, 2024

2023
Mixture Density Networks

Mixture Density Networks

Modeling Multi-Modal Regression with Probabilistic Neural Networks

Project Overview As part of the Probabilistic Graphical Models and Deep Generative Models course by Pierre Latouche and Pierre-Alexandre Mattei, I implemented Mixture Density Networks (MDNs) to model multi-modal regression problems with Antoine Debouchage and Valentin Denée. MDNs allow neural ne...

Posted by Clement Wang on December 15, 2023

Satellite Images Competition and Publication at BiDS 2023

Satellite Images Competition and Publication at BiDS 2023 ⭐

Predicting Leaf Area Index from Sentinel Satellite Data

Project Overview In April 2023, I participated in a 3-week competition to predict the Leaf Area Index (LAI) for each pixel in satellite images from Sentinel-1 and Sentinel-2. The project combined remote sensing, computer vision, and deep learning, and led to a publication at the Big Data from Sp...

Posted by Clement Wang on November 15, 2023

Spiking Neural Networks

Spiking Neural Networks

Exploring Bio-Inspired Neural Networks

Project Overview This was a one-week project exploring Spiking Neural Networks (SNNs)—a bio-inspired neural network paradigm that mimics the behavior of biological neurons. The work was conducted in collaboration with Timothée Masquelier, one of the most cited researchers on SNNs. Unlike traditi...

Posted by Clement Wang on September 15, 2023

Analyzing Business Opportunities at Etandex

Analyzing Business Opportunities at Etandex

Commercial Opportunity Prediction as a Freelance Business Intelligence Analyst

Project Overview Before returning from Germany, I took on a two-month freelance mission as a Business Intelligence Analyst at Etandex, one of France’s largest construction and infrastructure companies. The goal was to predict the likelihood of converting a commercial opportunity into a signed co...

Posted by Clement Wang on August 15, 2023

Photogen AI - Founding a Startup on Diffusion Models

Photogen AI - Founding a Startup on Diffusion Models ⭐⭐

From Research to Production: AI-Generated Images for Customers

Startup Journey The idea for Photogen AI came during my gap year. After an internship in a US startup, I wanted to try building something by myself. At the same time, Generative AI was exploding. When a friend suggested starting a project together, we jumped in and decided to launch a company ar...

Posted by Clement Wang on July 15, 2023

3D Surgery Tools Tracking from Single RGB Camera at Stryker

3D Surgery Tools Tracking from Single RGB Camera at Stryker ⭐⭐

Computer Vision for Medical Device Tracking in Surgical Environments

About Stryker Before returning to university, I did another six-month internship in the R&D team at Stryker in Freiburg, Germany. Stryker is a global MedTech company that develops surgical equipment and navigation systems. The work felt meaningful because it directly aimed to improve operati...

Posted by Clement Wang on July 15, 2023

Diagnostic Learning Differences with AI at Polygon Technologies

Diagnostic Learning Differences with AI at Polygon Technologies

Analysing video sessions with AI

About Polygon Technologies I spent six months at Polygon Technologies, a psychology startup based in Santa Monica, California. Polygon was trying something new: offering remote diagnostics for learning differences such as dyslexia and ADHD. The approach was innovative and ambitious, but after th...

Posted by Clement Wang on January 15, 2023

2022
Being a Machine Learning Consultant at Paris Digital Lab

Being a Machine Learning Consultant at Paris Digital Lab

Various Computer Vision projects for companies

Experience Overview After a year and a half of general engineering studies, I wanted to step into the professional world before finishing my degree. That’s when I started my gap year and joined the Paris Digital Lab. During my time at Paris Digital Lab, I worked on three Data science projects, e...

Posted by Clement Wang on July 15, 2022

Being President of Automatants AI Student Association

Being President of Automatants AI Student Association ⭐⭐

Leading 20 People, Organizing Events, and Promoting Machine Learning at CentraleSupélec

About Automatants Automatants is the AI student association of CentraleSupélec. It promotes machine learning by sharing knowledge, organizing events, competitions, and projects. I first joined in my 3rd year out of curiosity, and half year later, from January 2021 to January 2022, I became it...

Posted by Clement Wang on February 15, 2022

Tabular Data Competition: Building Classification from Geodata

Tabular Data Competition: Building Classification from Geodata

Ranked 2nd over 72 in Course Competition on Tabular Data

Project Overview Ranked 2nd out of 72 participants in a machine learning course competition focused on tabular data classification. The challenge involved classifying buildings into 5 distinct categories using a combination of geographical data and building metadata. The competition required pa...

Posted by Clement Wang on January 15, 2022

2021
Solving a Maze without looking

Solving a Maze without looking

Genetic Algorithm for Maze Solving

Introduction Genetic algorithms are an interesting class of optimization methods inspired by the process of natural selection. In this project, I applied a genetic algorithm to solve a maze with a fun twist: the algorithm never sees the maze layout. Instead, it provides a list of moves of a fixe...

Posted by Clement Wang on December 15, 2021

Quick, Draw! Competition at Automatants

Quick, Draw! Competition at Automatants ⭐

Winning a Competition on Imbalanced Image Classification

Introduction This blog post details my experience winning the Automatants Quickdraw! competition on imbalanced image classification. This competition was an excellent opportunity to apply everything I had learned over a year of deep learning study and practice, and to experiment with a broad ran...

Posted by Clement Wang on November 15, 2021

Neural Style Transfer and Perceptual Loss

Neural Style Transfer and Perceptual Loss

From Vanilla NST to Real-Time Style Transfer

Introduction Neural Style Transfer (NST) is a fascinating application of computer vision. The ability to separate and recombine the content and style of images using perceptual loss opens new possibilities for how we can use neural networks. In this project, I explored both the original NST algo...

Posted by Clement Wang on September 15, 2021

French Robotic Cup - Autonomous Robot Development

French Robotic Cup - Autonomous Robot Development

Building an Autonomous Robot for Object Manipulation in Defined Environment

Project Overview The French Robotic Cup (Coupe de France de Robotique) is a year-long robotics competition. Our team of 11 engineering students split into three sub-teams: one on the robot’s movement, one on the interaction system (getting cups onto the board), and one on vision, where I worked,...

Posted by Clement Wang on July 15, 2021

Infering Genes Links with Unsupervised Methods

Infering Genes Links with Unsupervised Methods

Collaboration with Pasteur Institute on Gene Expression Analysis

Project Overview This one-week project was conducted in collaboration with the Pasteur Institute, focusing on inferring gene relationships from expression data using unsupervised learning methods in R. Our team consisted of 6 students working with a comprehensive dataset containing gene expre...

Posted by Clement Wang on June 15, 2021

ViaRézo - Tech Student Association of CentraleSupélec

ViaRézo - Tech Student Association of CentraleSupélec

My first steps in a tech environment

About ViaRézo ViaRézo is the tech student association of CentraleSupélec. It provides internet access and many web services (mailing lists, VMs, social media, etc.) to more than 2000 students. About 20 members with 4 board members and 4 roots. Each year, ViaRézo gives beginner courses on variou...

Posted by Clement Wang on May 1, 2021

To The Ground - Unity Game Jam

To The Ground - Unity Game Jam ⭐

My First Group Coding Project: 3D Horror Game in Unity

Project Overview This was my very first group coding project with a team of 5 people. We designed and developed a 3D horror game in Unity using C# within one week. This intensive game jam experience taught me the fundamentals of game development and team collaboration. Game Trailer Gam...

Posted by Clement Wang on February 15, 2021

2020
My First Deep Learning Project: Cat Generator

My First Deep Learning Project: Cat Generator

From DCGAN to Progressive GAN for cat image generation

Project Overview This project marked my introduction to deep learning. As my very first deep learning project, I chose to explore the world of Generative Adversarial Networks (GANs) by training models to generate realistic cat images. I was accompanied by a more experienced member of Automatants...

Posted by Clement Wang on October 15, 2020