Path-based message passing GNN incorporating path features (angles, dihedral angles) significantly improves molecular property prediction, achieving MAE of 8.70×10^-3 eV on QM8.

Daniel Flam-Shepherd, Tony Wu, Pascal Friederich et al.

PolyGen is a Transformer-based autoregressive model that directly generates high-quality 3D polygon meshes, achieving a log-likelihood of 4.26 bits per vertex on ShapeNet.

Charlie Nash, Yaroslav Ganin, S. M. Ali Eslami et al.

Introduced Predictive Sampling, reducing ARM inference calls by 96.7% and achieving 27.6x speedup on MNIST using fixed-point iteration.

Auke Wiggers, Emiel Hoogeboom

Introduces continuous volume sampling (VS) for kernel interpolation, achieving near-optimal error bounds based on spectral properties, applicable to any Mercer kernel.

Ayoub Belhadji, Rémi Bardenet, Pierre Chainais

Proposes Bayesian deep learning with marginalization, using deep ensembles and MultiSWAG to improve accuracy, calibration, and mitigate double descent.

Andrew Gordon Wilson, Pavel Izmailov

TracIn estimates training sample influence via gradient inner products and checkpoints, applicable to any SGD-trained model.

Garima Pruthi, Frederick Liu, Mukund Sundararajan et al.

CodeBERT, a Transformer-based bimodal pre-trained model, uses MLM and RTD objectives to improve natural language-code understanding, achieving SOTA on code search and documentation.

Zhangyin Feng, Daya Guo, Duyu Tang et al.

GraSP prunes 80% weights at initialization, with only 1.6% accuracy drop, by preserving gradient flow.

Chaoqi Wang, Guodong Zhang, Roger Grosse

This paper introduces a multi-model approach combining diverse approximate posteriors with multiple importance sampling to improve decision-making over standard ELBO-based methods.

Romain Lopez, Pierre Boyeau, Nir Yosef et al.

MetaGen employs neural generation and adversarial training to synthesize theorems, boosting Metamath proof success rates by over 10%.

Mingzhe Wang, Jia Deng

Proposes directed graph hashing algorithms combining automorphism analysis, strong connectivity, and Merkle-style recursion, effectively handling cycles.

Caleb Helbling

Conditioned normalizing flow-based multivariate probabilistic forecasting, outperforming state-of-the-art on thousands of interacting time series.

Kashif Rasul, Abdul-Saboor Sheikh, Ingmar Schuster et al.

AGATHA uses graph-mining and Transformer models for hypothesis generation in medical research, achieving a recommendation accuracy of 0.901.

Justin Sybrandt, Ilya Tyagin, Michael Shtutman et al.

Proposed streaming sparse GP ensemble for implicit surface mapping, resource-efficient and comparable to exact GP.

Johannes A. Stork, Todor Stoyanov

ConvLab-2 integrates state-of-the-art models, supports multi-dataset evaluation, and provides diagnostic tools for task-oriented dialogue systems.

Qi Zhu, Zheng Zhang, Yan Fang et al.

Proposes a human-centered AI framework balancing control and automation to enhance reliability, safety, and trust.

Ben Shneiderman

VI-ULCB gives self-play in zero-sum Markov games \tilde{O}(\sqrt{H^3S^2ABT}) regret against fully adaptive opponents.

Yu Bai, Chi Jin

REALM integrates a trainable neural retriever with language models, boosting open-domain QA accuracy by 4-16%, with improved interpretability and modularity.

Kelvin Guu, Kenton Lee, Zora Tung et al.

Using continuous geodesic convolutions to learn on 3D shapes, surpassing handcrafted feature baselines.

Zhangsihao Yang, Or Litany, Tolga Birdal et al.

Using importance weighting to correct feedback shift under delayed feedback, enhancing conversion rate prediction accuracy.

Shota Yasui, Gota Morishita, Komei Fujita et al.