Technology
PyramidCLIP: HierarchicalFeatureAlignmentfor Vision-languageModelPretraining AnonymousAuthor(s) Affiliation Address email
Zhuang, K. Li, H. Cheng, X. Guo, F. Huang, R. Ji, and X. Sun, "Disco: Remedy213 self-supervised learning on lightweight models with distilled contrastive learning,"arXiv preprint214 arXiv:2104.09124,2021.215
Improved Algorithms for Collaborative PAC Learning
Neural Information Processing Systems http://nips.cc/
NearNeighbor
We show that LSH based algorithms can be made fair, without a significant loss in efficiency. Specifically, we show an algorithm that reports a point in the rneighborhood of a query q with almost uniform probability.
ForkMerge: Mitigating Negative Transfer in Auxiliary-Task Learning
Even in large language models, negative transfer problems may still exist.
ForkMerge: Mitigating Negative Transfer in Auxiliary-Task Learning
Even in large language models, negative transfer problems may still exist.
DATA: Differentiable ArchiTecture Approximation
Jianlong Chang, xinbang zhang, Yiwen Guo, GAOFENG MENG, SHIMING XIANG, Chunhong Pan
Neural Information Processing Systems http://nips.cc/
Playing hard exploration games by watching YouTube
Yusuf Aytar, Tobias Pfaff, David Budden, Thomas Paine, Ziyu Wang, Nando de Freitas
Deep reinforcement learning methods traditionally struggle withtaskswhere environment rewards are particularly sparse.
e97d1081481a4017df96b51be31001d3-Supplemental-Conference.pdf
Kinetics action classification.Our settings mainly follow[31,77]. The settings mainly follow [39, 77]. We report top-1 and top-5accuracyonthevalidation set. Entriesusingspatialresolution >2242 are noted in gray; entries using in-house data for supervision are in light blue. Importantly, our method is muchsimpler than many other entries.
e97d1081481a4017df96b51be31001d3-Paper-Conference.pdf
We observe that MAE can outperform supervised pre-training by large margins.