MLDM project - 2nd year of the Machine Learning & Data Mining Master
Made with:
Source: To be requested
From transactional data (purchase history) for 100,000 customers, the goal is to build a recommender system for product repurchase. We explored several ways to do so
Model the transactions problem as bi-partite graph and use Graph Mining technics (Graph Neural Networks, link predictions, random walks)
Build features from the data and combine them as a weighted vote (frequency, seasonality, similarity)
With these technics we were able to finish first in the class.
Particles detection in tomograms
Deep Learning course - 2nd year of the Machine Learning & Data Mining Master
Made with:
Source: To be requested
The goal is to detect particles in a cell (biologic). The project was part of a bigger competition on kaggle with more than 1,000 participants. We have at our disposal tomograms "slices" of cell and ground truth for certain cells.
We have several ways to solve the problem
Build a classic image detection model
Build a Neural Net that considers the tomograms of one cell as one 3D image and perform convolutions
Build a Neural Net that considers each tomograms independently, so each 2D image is its own sample
We explored deeply the last one to develop a 2D U-NET model that tries to predict where are the particles. There are 7 types of particles, with some being easy to detect and other more difficult. With this model we were able to detect in the image shapes of particles and then to return the coordinates of a certain particle in the cell.
Machine Learning project
1st year of the Machine Learning & Data Mining Master
Made with:
Source: To be requested
The goal was to use K-Nearest-Neighbors and various speed-up technics. For example we reduced the data via Principal Component Analysis and then learn a KNN on that, the advantage is that the dimensionality is reduced and then easier to learn, but by reducing the dimensions, we loose some informations, so we needed to chooses wisely the dimension.
The goal of the project was to render a simple 3d world and be able to move in it.
The map is a labyrinth and the player (modelised with a blue cube), can move inside:
SPACE: to rotate to the left (the step angle of the rotation is π/8)
UP ARROW: Move forward
DOWN ARROW: Move backward
LEFT ARROW: Move to the left (according to angle)
RIGHT ARROW: Move to the right (according to angle)
Tetris-like project
2nd year of the Computer Science Bachelor
Made with:
Source: None
The goal of the project was to create a tetris-like game using MLV.
