MARGE introduces a retrieval-based pretraining method for multilingual multi-document paraphrasing, achieving strong zero-shot performance on translation and summarization.

Mike Lewis, Marjan Ghazvininejad, Gargi Ghosh et al.

Proposes epistemic uncertainty-aware robust imitative planning (RIP) for detecting and recovering from distribution shifts in autonomous driving.

Angelos Filos, Panagiotis Tigas, Rowan McAllister et al.

This survey reviews three categories of NLG evaluation metrics: human, automatic, and machine-learned, analyzing their strengths, weaknesses, and future directions.

Asli Celikyilmaz, Elizabeth Clark, Jianfeng Gao

This paper proves that invalid action masking corresponds to valid policy gradients, scaling efficiently in large action spaces, outperforming penalty methods.

Shengyi Huang, Santiago Ontañón

Spectral bias and task-model alignment explain generalization in kernel regression and infinitely wide neural networks.

Abdulkadir Canatar, Blake Bordelon, Cengiz Pehlevan

Proposed a calibration-free, real-time 3D object tracking system combining neural detection and planar tracking on mobile devices, achieving 26FPS.

Adel Ahmadyan, Tingbo Hou, Jianing Wei et al.

Pix2Vox++ employs multi-scale context-aware fusion for accurate multi-view 3D reconstruction, outperforming SOTA with IoU of 85.2%.

Haozhe Xie, Hongxun Yao, Shengping Zhang et al.

Proposed neural models (Transformers and multilingual RNNs) with data augmentation achieved over 90% accuracy across 45 typologically diverse languages.

Ekaterina Vylomova, Jennifer White, Elizabeth Salesky et al.

wav2vec 2.0 achieves speech recognition via self-supervised learning, reaching 1.8/3.3 WER.

Alexei Baevski, Henry Zhou, Abdelrahman Mohamed et al.

CenterPoint uses point-based detection with heatmaps and regression, achieving 65.5 NDS on nuScenes with a single model.

Tianwei Yin, Xingyi Zhou, Philipp Krähenbühl

Proposes NPAS and SSNs for automatic parameter-efficient neural network training under arbitrary budgets.

Bryan A. Plummer, Nikoli Dryden, Julius Frost et al.

Extends NNGP and NTK to multi-head attention, proving Gaussianity in the infinite width limit with kernel expressions.

Jiri Hron, Yasaman Bahri, Jascha Sohl-Dickstein et al.

PatientBandits algorithm excels in handling stochastic bandits with arm-dependent delays.

Anne Gael Manegueu, Claire Vernade, Alexandra Carpentier et al.

BlazePose is a lightweight CNN for real-time human pose estimation on mobile devices, outputting 33 keypoints at over 30fps.

Valentin Bazarevsky, Ivan Grishchenko, Karthik Raveendran et al.

This paper introduces semi-supervised learning with Big Self-Supervised Models (SimCLRv2), achieving 73.9% top-1 accuracy on ImageNet with only 1% labels, outperforming previous methods by 10x.

Ting Chen, Simon Kornblith, Kevin Swersky et al.

SwAV introduces a clustering-based contrastive method for unsupervised visual representation, achieving 75.3% top-1 accuracy on ImageNet without memory banks.

Mathilde Caron, Ishan Misra, Julien Mairal et al.

Proposed Global-Local Attention Transformer (GLAT) learns structured visual commonsense, improving scene graph robustness without external knowledge.

Alireza Zareian, Zhecan Wang, Haoxuan You et al.

CRISS iteratively mines pseudo-parallel sentences using pretrained sentence embeddings, boosting unsupervised translation BLEU by 2.4 on average.

Chau Tran, Yuqing Tang, Xian Li et al.

Proposes PipeDream-2BW, a memory-efficient pipeline training system for large-scale DNNs, achieving up to 20× speedup via double-buffered weight updates and automated model partitioning.

Deepak Narayanan, Amar Phanishayee, Kaiyu Shi et al.

AWAC combines offline datasets with online fine-tuning, boosting robotic skill learning efficiency.

Ashvin Nair, Abhishek Gupta, Murtaza Dalal et al.