CENTRALE LYON - Post-doctoral researcher (12 months) Mathematics of energy-efficient AI: diffusion models on emerging hardware
Core
Developing mathematical convergence theory for diffusion models to quantify energy gains from analog/photonic hardware by mapping device-level non-idealities to score-error budgets.
Role type
Post-doctoral researcher (applied mathematics & hardware-aware AI)
Builds
Energy-accuracy frontiers for PCM-based photonic matrix-vector engines and FeFET-based analog computing alternatives
Domain
Energy-efficient AI, analog/photonic computing, diffusion models, stochastic analysis
Deliverable
production ML models
Required skills
probability and stochastic analysis, numerical analysis, optimisation, machine learning, scientific computing (PyTorch/JAX), mathematical modelling, hardware-aware AI
Preferred skills
diffusion/score-based generative models, finite-precision arithmetic, hardware-aware ML, approximate/stochastic computing, device modelling, circuit simulation, photonics, ferroelectric devices, energy modelling
Technologies
PyTorch, JAX, PCM-based photonic matrix-vector engines, FeFET-based analog computing
Responsibilities
Measure induced score error from network arithmetic perturbations; map device physics error models to score-error budgets; extend convergence guarantees to mixed frozen/redrawn error processes; co-design energy-accuracy frontiers with hardware teams and verify predictions against calibrated measurements
Seniority
Post-doctoral, hands-on IC researcher