simon-benigeri Goto Github PK

🔭 I’m Simon. I collaborate with physician-scientists to research applications of AI and machine learning for cardiovascular disease, with a particular focus on heart failure. We work with multimodal and longitudinal data. I graduated from the Northwestern University MSAI program in March 2022. I'm from Paris, France, where I worked for IBM's Watson and AI consulting team for 3 years.

⚡ I’m searching for full time AI/ML engineering roles. If you're hiring, please reach out!

The ideal role:

• Focuses on solving clear problems, i.e., What problem are we solving? Can you give me an example? Why is this important?
• Requires creative problem solving. I enjoy reading research papers to find new ideas to implement.

📫 How to reach me:

• [email protected]
• https://www.linkedin.com/in/simon-benigeri-30993338/

🌱 At Northwestern I worked on research projects in statistical language modeling.

• Evaluating out of domain robustness of neural abstractive summarization models (e.g., Pegasus, BART, T-5).
• Delivering machine teaching teaching functionalities for an information extraction system.

🏫 I'm always trying to learn. Here are some of my recent personal projects:

• Dope image classifier. My friend mkobbi and I chose a simple project, like image classification on CIFAR10, and we focus on the ML engineering aspects. It's a good way to to get experience with technologies like Pytorch-lightning, optuna, weights and biases, RedisAI, ONNX, and streamlit. (in progress - we need to write documentation!)
• Improving financial trading decisions with deep RL and transfer learning is a project where my colleagues and I implement a Deep Q learning agent to trade stocks. We "made profit" on past data but don't use this agent for your own investments.
• LSTM language model is a project where my colleagues and I learned to train a word-level language model with LSTM. We train on 2 corpora: Wikitext-2 and NY Times articles on covid-19.
• Low Precision Machine Learning. I set up a code base to run experiments that measure error due to training ML algorithms in low precision floating point representations. I also simululate stochastic rounding to see if it helps. The current repo uses a very simple model and toy datasets so we don't notice any error due to quantization. However, you can replace the model, the dataset, and run your own experiment.