(1/N) Happy to share my first machine learning conference paper as a Ph.D. student. "Auto-Transfer: Learning to Route Transferrable Representations" accepted at ICLR 2022 (@iclr_conf). @IBMResearch @csatrpi @rpi #MachineLearning #iclr2022
arxiv: arxiv.org/abs/2202.01011
arxiv: arxiv.org/abs/2202.01011
(2/N) Knowledge transfer between heterogeneous source and target networks/tasks has received a lot of attention in recent times. We answer the following questions in this paper. 1) What knowledge to transfer? 2) Where to transfer? 3) How to transfer the source knowledge?
(3/N) To address "What knowledge to transfer?" and "Where to transfer?", we propose an adversarial multi-armed bandit (AMAB) that learns the parameters of our routing function.
(4/N) To address "How to transfer the source knowledge?", we propose to meaningfully combine feature representations from source and target networks. This is in contrast with existing methods that force the target representation to be similar to source representation.
(5/N) The benefits of the proposed method are demonstrated on multiple datasets. Significant improvements are observed over seven existing benchmark transfer learning methods, particularly when the target dataset is small.
(6/N) Thanks to all the mentoring and support by co-authors at IBM Research @keerthi166 @superluss @Karthik, @pinyuchenTW, and Amit. Thanks to @iclr_conf reviewers for constructive feedback that improved the work. Looking forward to presenting at ICLR.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
