K
Kilian Q. Weinberger
Researcher at Cornell University
Publications - 241
Citations - 71535
Kilian Q. Weinberger is an academic researcher from Cornell University. The author has contributed to research in topics: Computer science & Deep learning. The author has an hindex of 76, co-authored 222 publications receiving 49707 citations. Previous affiliations of Kilian Q. Weinberger include University of Washington & Washington University in St. Louis.
Papers
More filters
Fixed neural network steganography: train
TL;DR: The proposed method, Fixed Neural Network Steganography (FNNS), yields significantly lower error rates when compared to prior state-of-the-art methods and achieves 0% error reliably for hiding up to 3 bits per pixel (bpp) of secret information in images.
I s h igh v ariance u navoidable in rl? a c ase s tudy in c ontinuous c ontrol
TL;DR: In this article , the authors investigate causes for the perceived instability in reinforcement learning experiments and propose several methods to decrease the variance of continuous control from pixels with an actor-critic agent.
On the Effectiveness of Offline RL for Dialogue Response Generation
TL;DR: This paper study the efficacy of various offline RL methods to maximize sequence-level objectives for dialogue response generation and show that offline RL shows a clear performance improvement over teacher forcing while not inducing training instability or sacrificing practical training budgets.
Proceedings Article
Characterizing the Loss Landscape in Non-Negative Matrix Factorization.
TL;DR: In this article, the authors revisited the NMF optimization problem and analyzed its loss landscape in non-worst-case settings, showing that gradients in deep networks tend to point towards the final minimizer throughout the optimization procedure.
Proceedings ArticleDOI
Unsupervised Adaptation from Repeated Traversals for Autonomous Driving
Yurong You,Cheng Perng Phoo,Katie Luo,Travis Zhang,Wei-Lun Chao,Bharath Hariharan,Mark E. Campbell,Kilian Q. Weinberger +7 more
TL;DR: In this article , the authors leverage unlabeled data collected from the end-users' environments (i.e. target domain) to adapt the system to the difference between training and testing environments.