Sep 2022 - Jan 2023

Can we improve the performance of self-supervised models in the presence of a small amount of labels?

The improvement in the output representations can be achieved by emphasizing the quality of the self-supervised models' produced representations. This can be accomplished through a two-step approach. Firstly, incorporating a linear supervision layer in the initial stage. Secondly, enforcing a correlation close to one among inputs belonging to the same class. By employing these techniques, we can enhance the overall quality of the output representations. Report here 

Language: Python

Tools: PyTorch, PyTorch Lightning, WandB, Colab

Notions: Computer vision, self semi and supervised learning

Feb 2022–Sep 2022

Self-supervised learning aims to learn useful representations of input data. In continuous learning, data is fed to the model sequentially, its efficacy is drastically diminished. We can prevent severe forgetfulness and continue to train our models by adding a prediction layer that forces the current representations to precisely match the frozen learned representations.

Language: Python

Tools: PyTorch, PyTorch Lightning, WandB, Colab

Notions: Computer vision, self-supervised learning, supervised learning, continual learning, distillation model.

Oct 2021–Dec 2021

In the last two decades, the increasing number of shocks and financial crises has been a major issue for the financial risk management teams. Among the wide range of exercises in this field, stress tests have become a main guideline for the regulator in order to assess the banking system resilience against the realizations of various categories of risk (market, credit, operational, climate, etc). The main challenge is to simulate unfavorable extreme (but plausible) negative returns similar to a historical dataset.

Task: This is an unsupervised learning problem: Given real data from stock market indexes that will act as a train dataset, the task is to learn a generative model that simulates synthetic stock market indexes. Code is make available here

• Time Series Generation of financial data.

• Language: Python

• Task: Implement Losses, Implement Generative Models, LU-factorization, Gaussian Model

Feb 2021 - Sep 2021

• Mathematical modeling of the concept of resources utilization for a resource management platform

• Optimization of a scheduler for fair resource allocation.

• Implementation of schedule for fair allocation of resources available.

Language: Python, Java, Google OR-Tools

Task: Mathematical modelling of resources usages, Scheduler Implementation, Simulator Implementation

Apr 2021 - Jun 2021

• Collecting and labelling data. Generate some using patterns of commands online.

• Implementation of a deep learning model to extract commands from a user's queries.

• Use of ktrain framework and fine-tuning of BiLSTM-Bert model to obtain a language processing model capable of extracting commands.

Jul 2020 - Sep 2020

• Research, design and implement machine learning applications to solve user misunderstanding problems on the web. Code is made available here

• Process and analyze users' queries to extract the real need of the user

• Deploy the resulting solution on a Windows server

May 2020 - Jun 2020

• Develop a computer vision model for classification of images of people's wearing mask. Code is made available here

• Docker configuration for launching an instance of the server deployed on Render

• Model from scratch, then use of PyTorch framework and ResNet architecture to classify my images

• Push solution on GitHub and deployment on Render

May 2017 - May 2020

• Game based on how the current markets work.

• On the one hand, we have sellers who offer services and on the other hand buyers who take products and finally banks who make loans

• Technologies: Java, JavaFx