Using AlphaZero and RNNs for strategic decision-making in Hex and Counter Wargames, showing promising results across terrains.

Guilherme Palma, Pedro A. Santos, João Dias

RG-VFM extends variational flow matching to Riemannian manifolds, improving protein/material generation by capturing curvature effects.

Olga Zaghen, Floor Eijkelboom, Alison Pouplin et al.

SWE-Lancer benchmark evaluates 1,400 real freelance software tasks; current models achieve under 45% success, far from earning $1 million.

Samuel Miserendino, Michele Wang, Tejal Patwardhan et al.

Transferability-aware hypernet framework with H-embedding significantly improves continual learning by capturing task relationships, achieving top performance on benchmarks.

Yanru Wu, Jianning Wang, Xiangyu Chen et al.

OctoTools is a training-free, multi-agent framework with standardized tools, boosting accuracy by 9.3% across 16 complex reasoning tasks.

Pan Lu, Bowen Chen, Sheng Liu et al.

MasRouter unifies collaboration mode, role allocation, and LLM routing, improving multi-agent system performance by 3.51% and reducing costs by over 52%.

Yanwei Yue, Guibin Zhang, Boyang Liu et al.

Using LLMs to generate interpretable mechanisms with constraint fixing, achieving competitive performance.

Jiayuan Liu, Mingyu Guo, Vincent Conitzer

This paper introduces frequency-guided noise scheduling in diffusion models, enhancing inductive biases and improving generation quality by manipulating spectral content.

Thomas Jiralerspong, Berton Earnshaw, Jason Hartford et al.

GNN+ framework enhances classic GNNs, outperforming SOTA Graph Transformers across 14 datasets with superior efficiency.

Yuankai Luo, Lei Shi, Xiao-Ming Wu

TANTE employs neural Taylor expansion for time-adaptive operator learning, reducing error by 60-80% and speeding inference by 30-40%.

Zhikai Wu, Sifan Wang, Shiyang Zhang et al.

Proposes CAM-UV-X, integrating residual and conditional independence tests, to identify causal directions amid hidden paths with theoretical guarantees.

Thong Pham, Takashi Nicholas Maeda, Shohei Shimizu

DMWM couples RSSM-S1 with LINN-S2 logic, improving long-horizon logical consistency by 14.3%.

Lingyi Wang, Rashed Shelim, Walid Saad et al.

This survey unifies federated and continual learning, mapping heterogeneity, forgetting, communication, and privacy challenges; it reports no unified numerical experiments.

Parisa Hamedi, Roozbeh Razavi-Far, Ehsan Hallaji

Proposes a latent recurrent-depth language model with 35B parameters, enabling test-time reasoning scale via iterative latent space inference, significantly boosting performance.

Jonas Geiping, Sean McLeish, Neel Jain et al.

Proposes the concept of coexistence for embodied agents, emphasizing continuous adaptation leveraging situated knowledge for long-term human interaction.

Hannah Kuehn, Joseph La Delfa, Miguel Vasco et al.

Proposes iterative importance fine-tuning of diffusion models via self-supervision, optimizing control for conditional sampling with theoretical guarantees.

Alexander Denker, Shreyas Padhy, Francisco Vargas et al.

Proposes IEE, an iterative exploitation-exploration framework, boosting importance scores for sparse pruning, achieving +1.3% Top-1 accuracy on ImageNet ResNet50.

Xinglong Sun, Maying Shen, Hongxu Yin et al.

Introduces ‘Platinum Benchmarks’ to reduce label noise, assesses LLM reliability, revealing even state-of-the-art models fail on simple tasks with 5% error rate.

Joshua Vendrow, Edward Vendrow, Sara Beery et al.

Flow Q-Learning (FQL) uses flow matching to model complex action distributions, avoiding recursive backpropagation, and achieves state-of-the-art results on 73 tasks.

Seohong Park, Qiyang Li, Sergey Levine

Proposes a similarity evaluation framework for pretraining data selection, revealing off-the-shelf embeddings underperform compared to specialized models.

Dylan Sam, Ayan Chakrabarti, Afshin Rostamizadeh et al.