- Homepage
- >
- Scientists
- >
- Scientific events
- >
- Seminars
- >
- 2025
- >
- Deep Learning and Hardware Acceleration for AI Inference
| Type | Seminar |
| Date | August 27, 2025 - 11:00 |
| Time | 11:00 |
| Location | Room 105, GANIL, Caen | France |
Deep Learning and Hardware Acceleration for AI Inference
Frédéric Druillole (LP2I Bordeaux)
Deep learning is a subfield of machine learning where models, inspired by the human brain, learn from large datasets through neural networks. The training process relies on two key elements:
•Loss function: It measures the difference between the model’s prediction and the actual target. The goal of training is to minimize this loss.
•Gradient Descent Algorithm: This optimization technique updates the model’s weights by calculating the gradient (slope) of the loss function with respect to each weight.
Once trained, deep learning models—such as DNNs (Deep Neural Networks), CNNs (Convolutional Neural Networks), RNNs (Recurrent Neural Networks), LNNs (Liquid Neural Networks), or Neural ODEs—can bedeployed for inference on a variety of hardware platforms. To meet real-time and power-efficient constraints, AI inference is increasingly performed on specialized hardware.
•FPGAs (Field Programmable Gate Arrays) ,GPUs (Graphics Processing Units), Edge Processors (suchas RISC-V AI cores, STM32, MAX78000).
Modern Edge AI computing and Spatial Accelerators (logic-based architectures) bring inference capabilitiescloser to the data source, enabling fast, efficient, and secure decision-making at the edge. These architectures are optimized for executing AI models with minimal latency and power consumption, making them ideal for robotics, autonomous systems, IoT, and real-time industrial or scientific applications.
The R&T Think initiative is dedicated to exploring and evaluating emerging AI-capable components by testing them with simple, interpretable models. This early-stage research helps assess their computational performance, energy efficiency, and integration potential.
Once validated, the most promising hardware–model combinations are used to design advanced AI solutions tailored to the specific needs of scientific instruments in experimental physics.