cv
Basics
Name | Giovanni Dispoto |
Label | PhD Student |
{name} DOT {surname} (AT) polimi DOT it | |
Summary | PhD Student involved in Reinforcement Learning research applied to financial trading |
Education
-
2023.11 - Present Milano, Italy
PhD
Politecnico di Milano
Reinforcement Learning for Financial Trading
- Algorithimic Trading
- Reinforcement Learning
- Machine Learning
-
2020.09 - 2023.05 Milano, Italy
MSc in Computer Science and Engineering
Politecnico di Milano
Reinforcement Learning for Financial Trading
- Machine Learning
- Reinforcement Learning
- Applied AI in Biomedicine
- Artificial Neural Networks and Deep Learning
- Foundation of Operations Research
Skills
Artificial Intelligence | |
Machine Learning | |
Reinforcement Learning |
Tech | |
Python | |
Java | |
C | |
Docker | |
Swift/SwiftUI |
Languages
Italian | |
Native speaker |
English | |
Upper Intermediate |
Interests
Artificial Intelligence | |
Machine Learning | |
Reinforcement Learning | |
Reinforcement Learning for Trading and Market Making | |
Deep Learning |
Projects
- 2021.02 - 2021.09
Performance Benchmarking of Deep Learning Applications
Research Project. Supervisors: Prof. Danilo Ardagna, Federica Filippini
- This framework is part of the AI-SPRINT Project funded by the European Union Horizon 2020 research and innovation programme
- Adaptation of the tf-slim by Google on Tensorflow 2
- a-GPUBench allows training a model on a target machine (e.g.a Server or AWS service) remotely
- The Architecture, the hyperparameters and the dataset are specified using an XML file
- 2021.12 - 2022.03
Premature Ventricular Complexes (PVCs) and Premature Atrial Complexes (PACs) detector using an ECG-based Deep Learning approach
Course Teamwork
- 1-D Convolutional Neural Network (1D-CNN) that classifies a beat as Normal Sinusoid Beat (N),Premature Ventricular Complexes (S) and Premature Atrial Complexes (V) relying on the ECG signal
- The starting dataset was composed by about 92% of N, 4.5% of V and 3.5% of S samples
- Our model reached 95% of accuracy on N, 83% on V and 76% on S on a separated test set with unseed patients
- We applied XAI techniques as LIME and GradCam to cross-check that the model was learning meaningful features as R-R peak distance.