|
Yaiza Bermudez | INRIA | Past Experience
Internships
Commissariat a l'Energie Atomique et aux energies alternatives (CEA), Grenoble, France :
Adaptative predictions model's development for the urban cold and heat network's optimization
supervised by Marina REYBOZ (CEA-LIST) and Mathieu VALLEE (CEA-LITEN)- 6 months 2024.
- Description:
This internship focused on improving heat usage forecasting in district heating networks, addressing challenges like data scarcity and catastrophic forgetting in traditional models.
A Multi-Head Attention (MHA) mechanism tailored for regression was developed and integrated into the DreamNet incremental learning framework.
Key innovations included preprocessing data with the Hodrick-Prescott filter to isolate trends and incorporating future temperature data for enhanced accuracy.
The MHA model achieved significant performance improvements (MAE: 0.303 MW, RMSE: 0.423 MW) and demonstrated resilience to data challenges.
Future work will focus on refining incremental learning and exploring advanced preprocessing techniques.
The master thesis can be found here
- Description:
This internship focused on forecasting solar power generation, addressing challenges posed by weather-dependent fluctuations.
The study compared data-driven models like XGBoost and LSTM with theory-driven models such as the Ineichen-Perez and Simple Solis clear-sky models.
Data-driven approaches demonstrated superior performance, achieving a lower RMSE (9.4%) compared to theory-driven models (29.65%).
To balance performance and carbon footprint, a hybrid model was developed, combining the strengths of both approaches.
This work culminated in a publication hal-04339309, highlighting the contribution to advancing solar energy forecasting and optimizing energy management systems.
The report can be found here
Ministerial Delegation for Artificial Inteligence (DMIA), French home Ministry, Paris, France :
Discovery of the home ministry in the field of artificial inteligence with the management of national and international programs supervised by the Préfet and Tanguy Herserant - 4 months 2022
- Description:
This 4-month internship with the Ministerial Delegation for Artificial Intelligence (DMIA) provided valuable insight into AI applications for national and international programs.
The role involved contributing to varius AI projects, utilizing Python as a primary programming tool.
Additionally, a collaborative effort with an alternant focused on developing a VPN to ensure secure access to external servers.
This experience established a solid foundation for future contributions to governmental AI initiatives.
- Description:
During this internship, I focused on analyzing existing data features and identifying new ones to better characterize cases among patients with diabetes.
This included exploring common patterns such as sleep schedules, physical activity levels, and other relevant factors.
|