Idrissa Dicko

Embedded AI for robotics, predictive world models (JEPA family) for imitation and reinforcement learning, signal and image processing for intelligent embedded systems, efficient/compressed deep learning architectures for real-time control.

Research Intern — Efficient Computer Vision - IRT SystemX / TotalEnergies - Palaiseau, France

(2026-03 - 2026-08)

• Designed resource-efficient deep learning pipelines for high-dimensional sensor and image data, targeting embedded/edge deployment for industrial monitoring — directly relevant to embedded AI constraints in humanoid robot control.
• Applied model compression (pruning, quantization) to convolutional and transformer-based architectures, optimizing the accuracy/compute trade-off for real-time, resource-constrained inference.
• Benchmarked energy and inference metrics across compressed models to quantify deployment-readiness on embedded hardware.

Research Engineer - Centre national de la recherche scientifique (CNRS) - Orsay, France

(2025-09 - 2026-02)

• Fine-tuned large Transformer architectures at scale; contributed to the CHIST-ERA FAIRClinical project on structuring heterogeneous data for downstream modeling.

M1 Research Intern - Centre national de la recherche scientifique (CNRS) - Orsay, France

(2025-05 - 2025-09)

• Built annotated datasets and implemented entity normalization pipelines for large-scale model training.

M1 Research Project - Laboratoire Interdisciplinaire des Sciences du Numérique (LISN) - Orsay, France

(2025-01 - 2025-03)

• Investigated failure modes (hallucinations) in large models and proposed mitigation strategies — relevant to reliability and safety of learned models deployed on physical systems.

Master - Data Science (Informatique) - Université Paris-Saclay (2024-09 - 2026-08)

Bachelor - Computer Science and Engineering - Université Technologique (2020-01 - 2024-06)

Baccalauréat - Collège Moderne Cheick Modibo Diarra (2018-10 - 2019-08)