[ContinualAI Seminars]: "Posterior Meta-Replay for Continual Learning"
5 views
Skip to first unread message
Keiland Cooper
Feb 23, 2022, 2:26:11 AM2/23/22
to Continual Learning & AI News
Hi All,
This Thursday 24-02-2022, 17.30 CET, for the ContinualAI Seminar, Christian Henning will present the paper:
Title: “Posterior Meta-Replay for Continual Learning”
Abstract: Learning a sequence of tasks without access to i.i.d. observations is a widely studied form of continual learning (CL) that remains challenging. In principle, Bayesian learning directly applies to this setting, since recursive and one-off Bayesian updates yield the same result. In practice, however, recursive updating often leads to poor trade-off solutions across tasks because approximate inference is necessary for most models of interest. Here, we describe an alternative Bayesian approach where task-conditioned parameter distributions are continually inferred from data. We offer a practical deep learning implementation of our framework based on probabilistic task-conditioned hypernetworks, an approach we term posterior meta-replay. Experiments on standard benchmarks show that our probabilistic hypernetworks compress sequences of posterior parameter distributions with virtually no forgetting. We obtain considerable performance gains compared to existing Bayesian CL methods, and identify task inference as our major limiting factor. This limitation has several causes that are independent of the considered sequential setting, opening up new avenues for progress in CL.
---------------
- YouTube link: [click here]
- Microsoft Teams: [ click here to join ]
- YouTube recordings of the previous sessions: https://www.youtube.com/c/ContinualAI
---------------
Feel free to share this email to anyone interested and invite them to subscribe this mailing-list here: https://groups.google.com/g/continualai
Please also contact me if you may be interested in speaking at one of our next sessions!
Looking forward to seeing you all there!
All the best,
Keiland Cooper
This Thursday 24-02-2022, 17.30 CET, for the ContinualAI Seminar, Christian Henning will present the paper:
Title: “Posterior Meta-Replay for Continual Learning”
Abstract: Learning a sequence of tasks without access to i.i.d. observations is a widely studied form of continual learning (CL) that remains challenging. In principle, Bayesian learning directly applies to this setting, since recursive and one-off Bayesian updates yield the same result. In practice, however, recursive updating often leads to poor trade-off solutions across tasks because approximate inference is necessary for most models of interest. Here, we describe an alternative Bayesian approach where task-conditioned parameter distributions are continually inferred from data. We offer a practical deep learning implementation of our framework based on probabilistic task-conditioned hypernetworks, an approach we term posterior meta-replay. Experiments on standard benchmarks show that our probabilistic hypernetworks compress sequences of posterior parameter distributions with virtually no forgetting. We obtain considerable performance gains compared to existing Bayesian CL methods, and identify task inference as our major limiting factor. This limitation has several causes that are independent of the considered sequential setting, opening up new avenues for progress in CL.
---------------
- YouTube link: [click here]
- Microsoft Teams: [ click here to join ]
- YouTube recordings of the previous sessions: https://www.youtube.com/c/ContinualAI
---------------
Feel free to share this email to anyone interested and invite them to subscribe this mailing-list here: https://groups.google.com/g/continualai
Please also contact me if you may be interested in speaking at one of our next sessions!
Looking forward to seeing you all there!
All the best,
Keiland Cooper
Reply all
Reply to author
Forward
0 new messages