mxochicale
commented
12 months ago From: sioni.paris.summers Sent: 18 September 2023 16:10 To: Xochicale, Miguel Subject: [Indico] Fast Machine Learning for Science workshop - instructions to presenters
Dear Miguel,
We’re looking forward to welcoming you to the Fast Machine Learning for Science workshop 2023 next week and hearing your presentation "Towards lightweight transformer-based models with multimodal data for low-latency surgical applications"! Here we have some information to help with the smooth running of your talk. General information for all attendees will be sent separately later.
Presentation Material (slideshow)
Please attach your presentation material to the indico agenda well before the start of the session in which you will present. The material will be downloaded before the session to the PC in the room, and to help us run on time it needs to be uploaded in advance. We recommend you upload your presentation in PDF format.
You will need to login to indico using the account you used to submit the abstract, navigate to “My Conference -> My Contributions” in the menu on the left-hand side of the event page, select the talk for which you are the assigned speaker, click the “edit materials” pencil icon button, and upload the file.
The full event agenda can be accessed here.
Remote Presentations
If you will be presenting remotely, you will be able to share the presentation from your computer, but you still need to upload the presentation to the indico agenda. Please switch your video on for presenting if you are able, it really helps to engage the audience.
Timing
You have been allocated either 15 minutes (standard) or 5 minutes (lightning) for your talk. Please check the indico agenda to make sure which talk type you have been allocated. These times both include the time for questions! We recommend that you aim for 12 minutes for presentation + 3 minutes for questions (standard) or 4 minutes for presentation + 1 minute for questions (lightning). The session chairperson will notify you when you have presented for 10 minutes (standard) or 3 minutes (lightning) to inform you that your time is nearly up and you should move towards your conclusions. We really want to allow time for questions and discussion after your talk, so we appreciate your help with sticking to the allocated time. Practise your talk in advance!
Monday HEP Lightning Talks
The last session of Monday will be dedicated to the lightning talks on the topic of High Energy Physics. If your talk is scheduled in this session, we need you to please upload your slideshow by the Monday morning at the latest, in PDF format. We will be compiling the contributions into one document to run the session more smoothly. Since there will be many presentations on topics like the LHC experiments and trigger systems, you can avoid introducing these concepts in your own talks, giving your more time to present your own results. Our colleague Thea Aarrestad will be presenting in the Monday morning session on Fast ML at the LHC experiments and will introduce common concepts to everyone.
Thanks, and we look forward to hearing from you next week, Alex and Sioni for the organisers
"Towards lightweight transformer-based models with multimodal data for low-latency surgical applications" has been accepted for 15 minutes presentation at the Fast Machine Learning for Science! Abstract: Surgical data technologies have not only been successfully integrated inputs from various data sources (e.g., medical devices, trackers, robots and cameras) but have also applied a range of machine learning and deep learning methods (e.g., classification, segmentation or synthesis) to data-driven interventional healthcare. However, the diversity of data, acquisitions and pre-processing methods, data types, as well as training and inference methods has presented a challenging scenario for implementing low-latency applications in surgery. Recently, transformers-based models have emerged as dominant neural networks, owing to their attention mechanisms and parallel capabilities when using multimodal medical data. Despite this progress, state-of-the-art transformers-based models remain heavyweight and challenging to optimise (with 100MB of parameters) for real-time applications. Hence, in this work, we concentrate on a lightweight transformer-based model and employ pruning techniques to achieve a balance in data size for both training and testing workflows, aiming at enhancing real-time performance. We present preliminary results from a machine learning workflow designed for real-time classification of surgical skills assessment. We similarly present a reproducible workflow for data collection using multimodal sensors, including USB video image and Bluetooth-based inertial sensors. This highlights the potential of applying models with small memory and parameter size, enhancing inference speed for surgical applications. Code, data and other resources to reproduce this work are available at https://github.com/mxochicale/rtt4ssa
Further details of the talk: https://indico.cern.ch/event/1283970/contributions/5550640/ Full program: https://indico.cern.ch/event/1283970/timetable/#20230926.detailed