Machine Learning, PyTorch, PyTorch Lightning & Digital Entrepreneurship
Machine Learning, PyTorch, PyTorch Lightning & Digital Entrepreneurship
Please reach out to siciliano AT diag DOT uniroma1 DOT for late enrollment or any inquiries regarding the course.
Intensive course (8 weeks) funded under SoBigData.it — PNRR, designed to train researchers and technicians in the operational and reproducible use of newly acquired computational infrastructures. We provide skills in machine learning, advanced practice in PyTorch and PyTorch Lightning for distributed training, and application of technologies to real-world use cases with a digital entrepreneurship module. The exercises include remote access to GPU-equipped machines, versioning workflow via GitHub, and monitoring tools to ensure experiment replicability.
All course materials are available in the following Google Classroom:
https://classroom.google.com/c/ODMxNzg2NzczMjI4?cjc=nwkoquxg
Course Code: nwkoquxg
You can join the online sessions using this link:
https://meet.google.com/wzf-ynep-fjk
The schedule for the course is provided at the bottom of this page.
For those attending the lectures in person, the first week will take place at the Department of Computer, Control, and Management Engineering of Sapienza University of Rome, located at Via Ariosto 25.
Calendar
Abstract: The course is a theoretical–practical introduction covering the fundamentals of modern machine learning: problem formulation (classification, regression, ranking), preprocessing pipelines, evaluation (metrics and experimental protocols), regularization, optimization, linear models, trees, ensembles, and an introduction to neural models for deep learning. Particular attention is given to good experimental practices, data management, and principles for making experiments reproducible, including practical guidance on using the newly acquired machines: copying and synchronizing code via GitHub on the purchased machines, and creating and managing isolated Python environments (virtualenv/conda) to ensure reproducibility and consistency of software installations.
Abstract: A practical course aimed at teaching the PyTorch ecosystem: tensors and operations, building custom models, managing datasets and dataloaders, optimization strategies, debugging, saving and loading checkpoints, acceleration techniques (mixed precision), and integration with monitoring tools (TensorBoard, Weights & Biases). Special attention will be dedicated to using the GPUs of the purchased machines: managing multiple devices in PyTorch, memory allocation and best practices for CPU↔GPU data transfer, and using torch.cuda for profiling.
Abstract: A course focused on PyTorch Lightning as an abstraction layer to accelerate and make GPU training reproducible: structure of the LightningModule and Trainer, checkpointing, callbacks, logging, learning-rate scheduling strategies, and job orchestration (e.g., batch scripts / scheduler usage). Emphasis will be placed on the correct use of the GPUs available on the purchased machines, particularly the use of multiple GPUs within a single machine, with practical examples on configuration, profiling, and best practices to ensure stability and efficiency of experiments.
Abstract: A practical course aimed at applying the technologies learned in the previous modules to a business case, with the involvement of tech startup founders. The first week begins with an introduction to what it means to build a startup, focusing on reasoned selection of an industrial domain in which to apply the technology, analysis of the problem faced by a specific group of users, and identification of the solution offered. The second week encourages participants to actually interact with potential customers, starting to study their behavior and presenting them with the solution concept.
SoBigData.it - PNRR, Strengthening the Italian Research IInfrastructure for Social Mining and Big Data Analytics
Sito web del progetto: https://pnrr.sobigdata.it