Horovod uses ring-allreduce for efficient GPU communication, reducing code changes and speeding up distributed TensorFlow training.

Alexander Sergeev, Mike Del Balso

Transformers with local self-attention improve ImageNet negative log-likelihood from 3.83 to 3.77, surpassing PixelCNN++.

Niki Parmar, Ashish Vaswani, Jakob Uszkoreit et al.

Proposes an attention-based CNN model (CAML) for ICD code prediction with 0.54 micro-F1 and interpretable text snippets.

James Mullenbach, Sarah Wiegreffe, Jon Duke et al.

Learn deep disentangled embeddings using F-statistic loss to enhance few-shot learning performance.

Karl Ridgeway, Michael C. Mozer

Introduces a compression-based generalization bound for deep nets leveraging noise stability, outperforming naive parameter counting.

Sanjeev Arora, Rong Ge, Behnam Neyshabur et al.

L4 loss-based stepsize adapts dynamically, improving Adam and Momentum optimizers across multiple architectures.

Michal Rolinek, Georg Martius

Combining color and depth to improve SSCnet did not surpass the original method.

Andre Bernardes Soares Guedes, Teofilo Emidio de Campos, Adrian Hilton

SRL survey: AE, E2C, ICM and priors learn compact control states that speed RL.

Timothée Lesort, Natalia Díaz-Rodríguez, Jean-François Goudou et al.

Proposes Tree SHAP, an efficient algorithm for consistent feature attribution in tree models, reducing complexity from exponential to polynomial time.

Scott M. Lundberg, Gabriel G. Erion, Su-In Lee

Deep domain adaptation leverages adversarial, statistical, and reconstruction methods to improve cross-domain visual tasks, achieving up to 89.5% accuracy on benchmarks.

Mei Wang, Weihong Deng

UMAP leverages Riemannian geometry and topology to produce scalable, high-quality low-dimensional embeddings that preserve both local and global data structures.

Leland McInnes, John Healy, James Melville

ENAS uses parameter sharing with a controller RNN to search neural architectures efficiently, reducing GPU time by 1000x, achieving CIFAR-10 error of 2.89%.

Hieu Pham, Melody Y. Guan, Barret Zoph et al.

DeepLabv3+ integrates ASPP and a simple decoder, using Xception backbone with depthwise separable convolution, achieving 89.0% mIOU on Pascal VOC 2012 without post-processing.

Liang-Chieh Chen, Yukun Zhu, George Papandreou et al.

This paper proposes a systematic classification of black-box explanation methods based on model type, problem context, and interpretability goals.

Riccardo Guidotti, Anna Monreale, Salvatore Ruggieri et al.

Texygen integrates multiple models and metrics, enabling comprehensive evaluation of text generation quality.

Yaoming Zhu, Sidi Lu, Lei Zheng et al.

Proposes regularized evolution to discover architectures surpassing human designs, achieving 83.9% top-1 ImageNet accuracy.

Esteban Real, Alok Aggarwal, Yanping Huang et al.

MotifNet employs motif-based multivariate polynomial filters to handle directed graphs, outperforming spectral methods on node classification tasks.

Federico Monti, Karl Otness, Michael M. Bronstein

This study demonstrates that fixing most parameters in deep CNNs and training only a small subset achieves near state-of-the-art performance.

Amir Rosenfeld, John K. Tsotsos

Proposed NoDeF, a nonparametric delayed feedback model, improves CVR prediction by capturing complex delay distributions.

Yuya Yoshikawa, Yusaku Imai

Generating Wikipedia articles by summarizing long sequences using a decoder-only architecture and ROUGE scores.

Peter J. Liu, Mohammad Saleh, Etienne Pot et al.