Welcome to SPoC, Space Proof of Concept, a creative and explorative event on how to train AI models in space - the ultimate edge!
AI Sweden, the Swedish National Space Agency and the RISE Space Data Lab together with you and partners aim to generate PoC:s for space, utilizing the infrastructure in Edge Lab as a simulated space setting.
The event will include inspirational speakers including a keynote by Christer Fuglesang; access to edge devices (simulated satellites); and space data from Copernicus Satellites in the Edge Lab.
The main focus of the event will be a 'Collaboraton' - a fusion between hackathon and collaboration. This will provide a creative setting for generating interesting ideas and future projects together. You can test grand ideas with others and accelerate our way to space!
Schedule can be found here and speakers here
The first part of the event is an open webinar 8-10:45AM 7th of September, if you wish to join please register here
Day 1
8.15-8.25 Intro by AI Sweden & the Swedish National Space Agency
8.25-8.40 Fredrik Bruhn - Unibap - About Spacecloud & more
8.40-8.55 Kim Henrikson - Zenseact/AI Sweden - Egde Lab
8.55-9.00 Short break
9.00-9.15 Evgenia Novikova - Smartilizer - About Federated frameworks and tools
9.15-9.45 Christer Fuglesang - KTH - The difference between biological and artificial intelligence in space
9.45-9.50 Short break
9.50-10.10 Tina & Tom Sjögren - Pythomspace
10.10-10.20 Massimiliano Pastena - European Space Agency
10.20-10.35 Sorin Cheran - Hewlett Packard Enterprise - Swarm Learning
10.35-10.45 Q&A and sum-up
Open webinar transitions into invite only session
10.45-12.00 Collaboraton kick-off brainstorming session (on site & remote)
12.00-13.00 LUNCH BREAK
13.00-16.30 Start of Collaboraton with free experimentation and exploration in the Egde Lab.
13.00-13:30 Deep dive into the edge lab infrastructure and Q&A to get everybody started
On demand: HPE Swarm learning walkthrough
16.30-17.30 After Work session - with Tom and Tina from Pythomspace calling in live at 17.00
Day 2
08.00-17.00 Collaboraton continues (By invite and application only, on site & remote)
09.00-09.45 Tobias Edman - Rymdstyrelsen giving a data walkthrough.
During the day, representatives from Smartilizer, HPE, Unibap and AI Sweden will be available for brainstorming and support on Slack.
Day 3
08:00-15:00 Continuation of Collaboraton - by invite and application only (on site & remote)
15.00 the event opens up to the public again
15.00-17.00 Conclusion, wrap up and sharing of the PoC:s (open webinar)
Please note that the collaboration part of the event, 10:45AM 7th to 5PM 9th of September is invite only and specific information about how to attend on-site/remote, slack channels, discord servers, VPN account etc has been sent out separately
At the end of the brainstorming session we will ask you to pair up into groups of 3-7. Our hope is that the brainstorming will lead to a discovery of shared interest and ideas and therefore aim to leave the pairing up to you. If you wish to change group during the event, please notify us in slack (or in person if you are attending on-site).
VPN account credentials has been sent out via mail, if you have not recieved them or has questions, reach out to edgelab@ai.se
Please refer to the gudies linked bellow for more detailed information for setting up the VPN connection.
NOTE: After you have logged in for the first time trough the vpn, you will need to change from the “default” password. After changing password, you will have the new password on both the VPN and services assigned to you.
Steps:
- Go to here - requires VPN connection.
- Login with the password given to you over Email.
- Go to Signing In and then press Update
- Change password.
- Done!
We've set up 3 baseline projects for the participants/collaborators to use freely during the event. Each machine has a "fresh" install of either the vendors suggested OS or when that does not applicable, Ubuntu 20.04 LTS. The hardware and networking setup for each project can be modified upon request, e.g. add/remove resources, limit bandwith etc.
All the HW are actually part of the same network infrastructure, but are segmented into vlans. (sidenote: It is possible to configure all communication paths including routing federation communication via 5G, 4G, Wifi etc. But for the event we have skipped that part to make exploration easier.)
Generic user guides for the various hw can be found here:
Nvidia Jetson AGX Xavier documentation and tutorials
Google Coral documentation and tutorials
Raspberry Pi 4 documentation and tutorials
(To access the projects you will have to connect via VPN (this also applies to those whom are attending on-site))
IP Range 172.25.17.32/27
Gateway: 172.25.17.33
| IP Address | Type | CPU | RAM | Storage |
|---|---|---|---|---|
| 172.25.17.34 | Data VM | 4 vCPU | 8GB | 50GB + Data |
| 172.25.17.35 | VM 1 | 4 vCPU | 8GB | 50GB |
| 172.25.17.36 | VM 2 | 4 vCPU | 8GB | 50GB |
| 172.25.17.47 | Raspberry 1 | 4 CPU | 4GB | 16GB |
| 172.25.17.48 | Raspberry 2 | 4 CPU | 4GB | 16GB |
| 172.25.17.49 | Coral 4 | 4 CPU | 8GB | 32GB |
| 172.25.17.50 | Xavier 1 | 4 CPU | 8GB | 32GB |
| 172.25.17.51 | Xavier 2 | 8 CPU | 32GB | 32GB |
| 172.25.17.52 | Coral 3 | 8 CPU | 32GB | 32GB |
IP Range 172.25.17.64/27
Gateway: 172.25.17.65
| IP Address | Type | CPU | RAM | Storage |
|---|---|---|---|---|
| 172.25.17.66 | Data VM | 4 vCPU | 8GB | 50GB + Data |
| 172.25.17.67 | VM 1 | 4 vCPU | 8GB | 50GB |
| 172.25.17.68 | VM 2 | 4 vCPU | 8GB | 50GB |
| 172.25.17.79 | Coral 6 | 4 CPU | 4GB | 16GB |
| 172.25.17.80 | Coral 5 | 4 CPU | 4GB | 16GB |
| 172.25.17.81 | Raspberry 3 | 4 CPU | 8GB | 32GB |
| 172.25.17.82 | Raspberry 4 | 4 CPU | 8GB | 32GB |
| 172.25.17.83 | Xavier 4 | 8 CPU | 32GB | 32GB |
| 172.25.17.84 | Xavier 3 | 8 CPU | 32GB | 32GB |
IP Range 172.25.17.96/27
Gateway: 172.25.17.97
| IP Address | Type | CPU | RAM | Storage |
|---|---|---|---|---|
| 172.25.17.98 | Data VM | 4 vCPU | 8GB | 50GB + Data |
| 172.25.17.99 | VM 1 | 4 vCPU | 8GB | 50GB |
| 172.25.17.100 | VM 2 | 4 vCPU | 8GB | 50GB |
| 172.25.17.111 | Coral 8 | 4 CPU | 4GB | 16GB |
| 172.25.17.112 | Coral 7 | 4 CPU | 4GB | 16GB |
| 172.25.17.113 | Raspberry 5 | 4 CPU | 8GB | 32GB |
| 172.25.17.114 | Raspberry 6 | 4 CPU | 8GB | 32GB |
| 172.25.17.115 | Xavier 5 | 8 CPU | 32GB | 32GB |
| 172.25.17.116 | Xavier 6 | 8 CPU | 32GB | 32GB |
Nvidia Jetson AGX Xavier & Drive PX2
Google Coral
Raspberry Pi 4
HPE EdgeLine1000 with Tesla T4 GPU generic specs
Comino Grando RM with Nvidia A100 / 3975WX / 256GB / 2TB NVMe generic specs
Virtual Machines (mainly CPU based)
On the Virtual Machine marked with Data in your project you'll find the Copernicus datasets under /mount/copernicus
For more details on the dataset see data/README.md in data directory.
FedBird - powered by Scaleout FEDn
The HW setup each team has been provided is compatible with FedBird, a proof of concept utilizing Scaleouts FEDn framework for the federation part together with YOLO3tiny on bird observation images.
For a step by step tutorial on how to set ut up on your project, please see the "AI_Sweden-Fedbird_tutorial_xxx.pdf" in this repository
Swarm by HPE
This is a great github repository to check out! HewlettPackard/swarm-learning: A simplified library for decentralized, privacy preserving machine learning
Recommendations from attendees
Link archive of Satellite datasets
Here is the labelled forest fire dataset
ML on Google Coral
Here are some nice demos for the Google Coral's thats nice to start with
Open source frameworks tutorials and introductions
Tensorflow Federated: Federated Learning for Image Classification
Pysyft: Intro and examples
A survey of current available open-source FedML FWs, done by AI Sweden partner Smartilzer
Open-Source Federated Learning Frameworks for IoT: A Comparative Review and Analysis:
Some recomended "light" reading
Federated Learning on Non-IID Data Silos: An Experimental Study
Advances and Open Problems in Federated Learning
A Survey on Federated Learning Systems: Vision, Hype and Reality for Data Privacy and Protection
Federated Learning for Mobile Keyboard Prediction
The strategy of the Swedish National Space Agency is a good reference on how to get funding for space endevours.
Communities
Openminded is an open-source community whose goal is to make the world more privacy-preserving by lowering the barrier-to-entry to private AI technologies