Jesus Garcia Ramirez
Machine Learning Engineer
- +34 665 444 632
- jgarciaramirezai@gmail.com
- jesusgarciaramirez.github.io
- Leuven, Belgium
Proactive machine learning engineer with 3+ years of specialized experience in creating and implementing effective ML solutions with a proven track record of excellent communication and teamwork demonstrated through successful collaboration within interdisciplinary teams of researchers, engineers and non technical stakeholders. Confident in the ability to excel in fast-paced environments while supporting smart business decisions
Experience
PhD Researcher
- Developed an accurate (80% explained variance) CNN-based model for predicting neuron responses to images, bridging the gap between computational and biological vision
- Optimized Receptive Field estimation by designing a novel Gaussian approximation, reducing fitting parameters by thousands
- Implemented a closed-loop pipeline leveraging CNN encoding models, successfully identifying optimal stimuli for recorded neurons within a high-pressure, one-day experiment
- Created an end-to-end interactive visualization to enhance model interpretability and facilitate communication with non-technical stakeholders
Research Engineer
- Engineered a highly-accurate (92% success rate) and fast (microsecond inference) Brain-Machine Interface system using a non-linear extension of Kalman filter, enabling real-time control for individuals with reduced mobility
- Developed an innovative online retraining procedure to reduce the amount of data required resulting in a 90% data utilization reduction, paving the way for broader accessibility
- Led a cross-functional team of researchers, engineersm and non-technical stakeholders to deliver the solution 6 months early, exceeding expectations
Projects
Efficient analysis of eye tracking data via Deep Learning
- Developed an automatic labelling tool to streamline the analysis of mobile eye-tracking recordings from an art exhibition
- Finetuned a video classification model (SlowFast) using curated 10k sample dataset, reducing manual workload on 80% with 90% accuracy
- Adapted Resnet to handle multidimensional time-series data for behaviour prediction, achieving 60% automation with 80% accuracy
Conference Presentations
Speaker at Society for Neurosciences
- Presented poster: Single neuron signatures of spatial attention in the human lateral occipital complex
Speaker at Neural Control of Movement
- Presented poster: Comparing reach direction decoding in macaque PMv, PMd and M1
Speaker at Society for Neurosciences
- Presented poster: Single unit correlates of visual reasoning in the human lateral occipital complex
Speaker at FENS
- Presented poster: Object decoding with spatial attention in the human lateral occipital complex
Speaker at Society for Neurosciences
- Presented poster: Decoding reaching direction from macaque dorsal and ventral premotor and primary motor cortex