title: "Multi-modal models, processing simultaneously different type of data"
description: "Raissi et al. (2019) introduced PINNs (Physics-Informed Neural Networks) in their article entitled Physics-Informed Neural Networks: A Deep Learning Framework for Solving Forward and Inverse Problems Involving Nonlinear Partial Differential Equations. This work shows that PINNs can efficiently solve a wide range of problems in fluid dynamics, in solid mechanics, or in quantum mechanics. PINNs represent an advanced IA-based modeling approach that allows to solves physics or engineering problems by using neural networks. Concretely, PINNs can be used to simulate physical phenomenon or to predict experiment results by using input data as initial conditions and physical parameters. Using PINNs generally requires sme knowledge about the underlying physics related to the task at hand and about IA modeling techniques."
description: "In the era of massive and diverse data, the ability for an AI model to process and integrate information from various sources is crucial. In deep learning, multi-modality allows to integrate various data, including text, image, audio, video into the same model to create more robust and more versatile AI systems. This session will cover various fundamental concepts and advanced techniques required to process these different types of information."
title: "Inference engineering and operational setup"
description: "Raissi et al. (2019) introduced PINNs (Physics-Informed Neural Networks) in their article entitled Physics-Informed Neural Networks: A Deep Learning Framework for Solving Forward and Inverse Problems Involving Nonlinear Partial Differential Equations. This work shows that PINNs can efficiently solve a wide range of problems in fluid dynamics, in solid mechanics, or in quantum mechanics. PINNs represent an advanced IA-based modeling approach that allows to solves physics or engineering problems by using neural networks. Concretely, PINNs can be used to simulate physical phenomenon or to predict experiment results by using input data as initial conditions and physical parameters. Using PINNs generally requires sme knowledge about the underlying physics related to the task at hand and about IA modeling techniques."
description: "After model training, model fine-tuning, pre-trained model selection, how to deploy it and make it easily usable? while trying to minimize its footprint? The answers to these questions correspond to particular engineering task: requiring to choose a suitable system, then transforming and compressing the model, accounting for energy and storage costs, expected performance level, latency. Then, it will require to embrace DevOps practice for operational setup, using several technologies, such container, Kubernetes orchestration, inference servers, Edge App deployment. We will focus on simple and dedicated solutions such as Gradio, BentoML."
