GDN uses graph neural networks with structure learning to detect anomalies in multivariate time series, outperforming baselines with 99% precision on SWaT and 97.5% on WADI.

Ailin Deng, Bryan Hooi

TD3+BC adds behavior cloning regularization and normalization to TD3, achieving SOTA performance on D4RL with half the computational cost.

Scott Fujimoto, Shixiang Shane Gu

HR-NAS integrates lightweight transformers and multi-scale search space to optimize high-resolution dense prediction networks.

Mingyu Ding, Xiaochen Lian, Linjie Yang et al.

Proposes IXOMD, a model-free algorithm for high-probability convergence to NE in two-player zero-sum partially observable Markov games, with rate O(1/√T).

Tadashi Kozuno, Pierre Ménard, Rémi Munos et al.

Using measure representation and sparse regularization, proves gradient descent can recover teacher network parameters with high probability in two-layer ReLU models.

Shunta Akiyama, Taiji Suzuki

Proposes scalable conformal Bayesian prediction intervals using importance sampling, ensuring finite-sample coverage with reduced computational cost.

Edwin Fong, Chris Holmes

Introduces sparse MoE into Vision Transformer, achieving 90.35% on ImageNet with 15B parameters, halving inference compute compared to dense models.

Carlos Riquelme, Joan Puigcerver, Basil Mustafa et al.

EMDR2 introduces end-to-end training of multi-document retriever and reader using EM algorithm, improving accuracy by 2-3 points on benchmarks.

Devendra Singh Sachan, Siva Reddy, William Hamilton et al.

SimpleView method excels on ModelNet40 with half the parameters of PointNet++.

Ankit Goyal, Hei Law, Bowei Liu et al.

STCN uses L2 similarity for broader memory coverage, reaching 85.4 J&F and 20.2 FPS on DAVIS 2017.

Ho Kei Cheng, Yu-Wing Tai, Chi-Keung Tang

GCE-GNN combines session and global graphs with attention mechanisms to improve session-based recommendation accuracy.

Ziyang Wang, Wei Wei, Gao Cong et al.

RobustNav benchmarks embodied navigation robustness under visual and dynamics corruptions, revealing significant performance drops and highlighting the need for improved adaptation.

Prithvijit Chattopadhyay, Judy Hoffman, Roozbeh Mottaghi et al.

Efficient parameter allocation in large sparse models using Hash Layers, improving performance.

Stephen Roller, Sainbayar Sukhbaatar, Arthur Szlam et al.

Proposes GFlowNet, a flow network-based generative model for non-iterative diverse candidate sampling, ensuring probabilities are proportional to rewards.

Emmanuel Bengio, Moksh Jain, Maksym Korablyov et al.

Proposes inverse model and contrastive learning-based Markov state abstraction, improving sample efficiency in deep RL.

Cameron Allen, Neev Parikh, Omer Gottesman et al.

Transferring 2D pretrained models to 3D point-cloud understanding via weight inflation and minimal fine-tuning achieves state-of-the-art results.

Chenfeng Xu, Shijia Yang, Tomer Galanti et al.

Spectral Attention Network (SAN) leverages Laplacian spectrum for node positional encoding, outperforming traditional GNNs with full connectivity.

Devin Kreuzer, Dominique Beaini, William L. Hamilton et al.

Neural Monge Map learns unpaired, general-cost optimal maps; CelebA inpainting reaches FID 3.7109 versus WGAN-GP 18.7942.

Jiaojiao Fan, Shu Liu, Shaojun Ma et al.

This paper proves self-attention matrices can approximate arbitrary sparse patterns with input adjustment, requiring hidden size d = O(log L).

Valerii Likhosherstov, Krzysztof Choromanski, Adrian Weller

Introduces geometric cyclical invariance to prove stability of entropic OT solutions, applicable even with infinite costs, advancing Schrödinger bridge theory.

Promit Ghosal, Marcel Nutz, Espen Bernton