Implementation of a deep learning architecture to create financial factors. The objective of this project is to reduce pricing errors by training neural networks to generate deep factors (intermediate features) based on firm characteristics(inputs) in order to predict portfolio returns (outputs).
In this project, we explore several approaches to optimise target re-identification (re-ID) as a re-ranking problem. Our work consisted in trying different methods to re-rank the re-ID results. Each image is represented by a node in a graph to which we have applied supervised and unsupervised link prediction methods.
The goal of this challenge is to predict the ISUP Grade using only Histopathology images. For that, we need to deal with the process of Whole Slide Images as huge gigapixel images and deal with the limited number of patients provided in the train set. This is a Multiple Instance Learning (MIL) problem which is a form of weakly supervised learning. Lean more about this in my github repository.
This project studies the Object DGCNN paper which proposes a method for object detection and velocity estimation in 3D LIDAR scans. It improves previous voxel- or pillar-based frameworks by adding a graph network and by avoiding non-maximum suppression by teaching the network to not produce duplicates through bipartite result-GT-matching.
Structure from motion is a technique of estimating the motion of the camera and recovering three-dimensional (3D) scenes from 2-dimensional (2D) image sequences taken from two or multiple different views of one camera. The objective of this project is to identify the various approaches to generating sparse 3D reconstructions using the Structure from Motion (SfM) algorithms.
Numerous super-resolution methods have been proposed in the computer vision community. In this project, we will investigate a family of models called ”denoising diffusion probabilistic models” (DDPM) which are nowadays of great interest for image generation. The goal of this project is to get familiar with this type of method and to understand how they are applied to super-resolution in this paper (SR3).