ProdSumNet framework employs product-of-sums matrix decomposition to drastically reduce neural network parameters, achieving 98.44% accuracy on MNIST with only 3554 parameters.

Chai Wah Wu

Empirical validation of Kearns et al.'s rich subgroup fairness algorithm on four real datasets, demonstrating quick convergence and favorable fairness-accuracy tradeoffs.

Michael Kearns, Seth Neel, Aaron Roth et al.

Models RL policy as Wasserstein gradient flow; introduces particle-based algorithms, outperforming TRPO/PPO.

Ruiyi Zhang, Changyou Chen, Chunyuan Li et al.

MAPO leverages memory buffers to improve policy gradient sample efficiency, achieving 2.6% accuracy gain on semantic parsing benchmarks.

Chen Liang, Mohammad Norouzi, Jonathan Berant et al.

Deep neural networks follow the Frequency Principle (F-Principle), capturing low-frequency components first, explaining generalization behavior.

Zhi-Qin John Xu, Yaoyu Zhang, Yanyang Xiao

Using Deep Deterministic Policy Gradient (DDPG) with monocular images, the vehicle learns autonomous lane following within 10 episodes, achieving over 200 meters of safe driving.

Alex Kendall, Jeffrey Hawke, David Janz et al.

TextWorld is a sandbox learning environment for training RL agents on text-based games, supporting handcrafted or automatically generated games.

Marc-Alexandre Côté, Ákos Kádár, Xingdi Yuan et al.

DARTS employs differentiable architecture search with bilevel optimization, achieving high efficiency and state-of-the-art results on CIFAR-10 and PTB datasets.

Hanxiao Liu, Karen Simonyan, Yiming Yang

Proposes EQL÷, a neural network with division units, for symbolic equation discovery and extrapolation, achieving high accuracy with minimal data.

Subham S. Sahoo, Christoph H. Lampert, Georg Martius

Proposes stepwise relative reachability combined with a new baseline to prevent bad incentives in impact penalties, improving safety in reinforcement learning.

Victoria Krakovna, Laurent Orseau, Ramana Kumar et al.

GamePad integrates deep learning with Coq, predicting proof steps to enhance automation and human-AI collaboration.

Daniel Huang, Prafulla Dhariwal, Dawn Song et al.

Analyzes FedAvg's performance drop on non-IID data via Earth Mover's Distance; sharing small global data improves accuracy by 30%.

Yue Zhao, Meng Li, Liangzhen Lai et al.

Proposes PETS, combining uncertainty-aware deep dynamics models with trajectory sampling, achieving high sample efficiency and asymptotic performance in model-based RL.

Kurtland Chua, Roberto Calandra, Rowan McAllister et al.

Proposes Anonymous Walk Embeddings (AWE), combining feature-based and data-driven methods, to enhance unsupervised graph classification with state-of-the-art accuracy.

Sergey Ivanov, Evgeny Burnaev

The study uses statistical physics to compute entropy and mutual information in deep neural networks, exploring the link between compression and generalization.

Marylou Gabrié, Andre Manoel, Clément Luneau et al.

A learning-based framework optimizes tensor programs for deep learning, reducing manual tuning and achieving performance comparable to hand-tuned libraries.

Tianqi Chen, Lianmin Zheng, Eddie Yan et al.

Online distillation enables parallel training of multiple models, doubling training speed and improving reproducibility on large datasets.

Rohan Anil, Gabriel Pereyra, Alexandre Passos et al.

GrBAL and ReBAL enable rapid online adaptation with only 1.5–3 hours of meta-training experience.

Anusha Nagabandi, Ignasi Clavera, Simin Liu et al.

Tensor2Tensor library enhances neural machine translation using the Transformer model, significantly improving BLEU scores.

Ashish Vaswani, Samy Bengio, Eugene Brevdo et al.

Constructs a three-layer deep neural network leveraging local manifold learning, achieving an approximation error of O(m^{-2s/(2s+d)}) based on sample size m.

Charles K. Chui, Shao-Bo Lin, Ding-Xuan Zhou