SRMC has been my first conference ever. I went with mixed expectations and had a great deal of excitement. Most of the conference attendees were fellow PhD students, which helped a lot. It is always nice when you find your people.
For the workshop, I presented our latest paper, "An Empirical Analysis of IDS Approaches in Container Security". The presentation is here if you would like to check it out. We also got awarded the best paper of the workshop.
There were two papers that really stood out from the rest, say, from a newcomer's perspective. "D-Cliques: Compensating for Data Heterogeneity with Topology in Decentralized Federated Learning" and "Democratizing Machine Learning".
"D-Cliques" had a clear movie-like three act structure.
- We are doing federated learning
- Different nodes have access to/train on datasets with different compositions (non-uniform labels)
- We found the best way to connect those nodes
It's this one!
"Democratizing Machine Learning" was presented beautifully. Again, the gist of the work can be summarized by just one figure;
The paper proposed an optimization algorithm for machine learning pipelines (e.g. Adam), but one that can work for low power devices. The green line on the figure above is the SGD optimizer, being run on devices "that are essentially as powerful as smartphones" (as I've learned, after asking the presenter directly after the session). The proposed algorithm ran on the same device set is not visible, unfortunately, because it's hidden behind the lines at top, because it's as good as them!
Listening to around 20 papers in the span of 4 days, presenting our paper and my PhD topic in the PhD forum, the poster sessions, meeting with everyone all together were exhausting. I cannot wait for my next conference!