13th TAILOR WP7 Seminar
We are pleased to announce the outline of the thirteenth event in our TAILOR WP7 seminar series.
Please let us know if you plan to participate by entering your name in this:
https://docs.google.com/document/d/1ESGppvYKrGPRBrn7l5Yo6Q2R03u6kJghLJieoLDrv5E/edit
The seminar will consist of two parts: a scientific and a social part. Below you can find a detailed overview of each part for the ninth seminar on 11 March from 15:30-17:00.
Scientific part: Online meeting
Meeting ID: 693 1745 3723 – Passcode: qu5!zD*m
https://universiteitleiden.zoom.us/j/69317453723?pwd=aEJEcE03QVFPQmtqUHowUFJuM2Mwdz09
15:30-16:00 — Carolin Benjamins About Tools, Exploration and Landscapes in Hyperparameter Optimization
Automated Machine Learning provides methods and processes to make machine learning more accessible and to improve efficiency of machine learning systems. By now, many subfields emerged.
This talk will take a tour starting from our Hyperparameter Optimization (HPO) tool SMAC based on Bayesian Optimization (BO) techniques. SMAC, short for Sequential Model-Based Algorithm Configuration, features a clever intensification routine to efficiently compare hyperparameter configurations and integrates methods for multi-fidelity and multi-objective optimization.
From here, we notice BO has its own HPs where it might be beneficial to not only set them statically, but dynamically throughout the optimization. Thus we step up a meta-level and aim to dynamically adjust the exploration-exploitation trade-off with SAWEI. SAWEI shows to adjust online to different domains and HP landscapes without meta-learning, without prior knowledge.
The landscapes and dynamically adjusting algorithms finally lead us to wondering how the HP landscapes change for iterative algorithms? We present a first approach to create dynamic landscapes for Reinforcement Learning algorithms, showing a change over time which underlines the importance of Dynamic Algorithm Configuration.
16:00-16:30 — Round table updates by all WP7 participants (one speaker per group)
Slides: https://docs.google.com/presentation/d/1i28GwpYxjniglvJFpHCvM4yuKehuiwdKicqza8iMO18/edit#slide=id.g1347b5b076f_0_0
- Overview highlighting ongoing and planned work focusing on the following:
- Collaboration between different tasks;
- Collaboration with other work packages;
- New collaborations (with people with whom that group has never worked with before).
Social part: GatherTown
16:30-17:00
- Optional: Provide an extra slide, poster, or other material to be shown in gather town (submit a PDF by email to a.w.bosman@liacs.leidenuniv.nl if what you want to present something)
- Suggestions for improving the seminar format
- Informal chats about possible collaborations
- Brainstorm/discuss ideas and mechanism for making our WP7 more impactful, more efficient and/or more enjoyable
- Chat about anything else you feel worthwhile!
Let us know on mattermost in the WP7 Automated AI channel if there are any questions, and please get in contact with us if you would like to volunteer as speaker for a future seminar.
We hope to see you at the seminar!
Best regards,
Annelot, Adem, Holger and Manuel (the TAILOR WP7 Seminar Organising Committee)
Join mattermost: https://talk.vision4ai.eu/signup_user_complete/?id=7k6tnj9zytfpbjpboa798im7dr