Proposes Bebop with TV loss and rejection sampling to stabilize MTP acceptance rate, achieving up to 95% and 1.8× RL training acceleration.

Yucheng Li, Huiqiang Jiang, Yang Xu et al.

This study compares xLSTM, Mamba-2, and Gated DeltaNet architectures, demonstrating xLSTM's superior performance in complex sequence tasks due to its robust state tracking and memory accumulation.

Anamaria-Roberta Hartl, Levente Zólyomi, David Stap et al.

Proposes Latent World Recovery (LWR), a robust multimodal learning framework that aligns modality-specific embeddings in a shared latent space, handling missing modalities without imputation.

Hui Wang, Tianyu Ren, Joseph Butler et al.

Fourier features improve point-cloud imitation, raising RoboCasa success from 13% to 34% and real-world normalized score from 14.8% to 40.2%.

Balázs Gyenes, Emiliyan Gospodinov, Jan Frieling et al.

This paper offers a machine learning perspective on uncertainty in dynamical systems, distinguishing aleatoric and epistemic uncertainties, and analyzing their propagation across tasks.

Yusuf Sale, Christopher Bülte, Felix Czaja et al.

RLCSD combines contrastive learning with reinforcement on-policy self-distillation, effectively mitigating style drift and improving reasoning model performance.

Leyi Pan, Shuchang Tao, Yunpeng Zhai et al.

Proposes GLACIER, a multimodal foundation model using contrastive learning and Finsler geometry for molecular property prediction, achieving SOTA results.

Emily Nguyen, Yongchan Hong, Harsh Toshniwal et al.

EEVEE framework employs a router-conditioned prompt set with co-evolution to enhance LLM robustness across heterogeneous task streams, improving scores by 10.38-24.32 points.

Weixian Xu, Shilong Liu, Mengdi Wang

Unified analysis of GP-UCB and DEC in RKHS bandits via MAIR framework, revealing their fundamental differences and advantages.

Yunbei Xu

COGENT integrates Graph Neural Networks with Neural ODEs for continuous long-term physical forecasting on irregular meshes, outperforming traditional autoregressive models.

Zesheng Liu, Maryam Rahnemoonfar

Proposes an efficient online algorithm for drifting halfspaces under Massart noise, achieving an error bound of η + ˜O(Δ^{1/3}/γ), nearly matching theoretical limits.

Mingchen Ma, Guyang Cao, Jelena Diakonikolas et al.

This paper introduces flexible sequence kernels based on evolutionary substitution matrices, leveraging Gaussian processes for data-efficient protein property prediction, outperforming embedding-based methods.

Martin Jankowiak, Yerdos Ordabayev, Rudraksh Tuwani et al.

Proposes a multi-scale residual-aware framework combining Transformer models, achieving significant reduction in systematic bias for time series forecasting.

Amrijit Biswas, Mustafa Kamal, Robin Krambroeckers et al.

Proposes high-temperature sampling to generate synthetic canaries for privacy auditing, improving detection of model memorization.

Nicole Mitchell, Galen Andrew, Arun Ganesh et al.

Proposed a continuous-time Markov chain framework to enhance insertion language models' flexibility and sampling efficiency.

Dhruvesh Patel, Benjamin Rozonoyer, Soumitra Das et al.

DRPO introduces smooth advantage-weighted quadratic regularization to improve stability and efficiency in LLM RL training, replacing hard masks with continuous gradient weights.

Jiarui Yao, Xiangxin Zhou, Penghui Qi et al.

Introducing Topological Neural Operators (TNO), a framework leveraging cell complexes and discrete exterior calculus to improve PDE modeling on complex geometries, achieving over 20% accuracy gains.

Lennart Bastian, Samuel Leventhal, Mustafa Hajij et al.

Proposes an fully automated time-series forecasting framework combining high-frequency dataset TimeTrack with dynamic local telemetry, using NAS to generate accurate models, effectively addressing cold-start issues.

Abd Elghani Meliani, Arora Sagar, Adlen Ksentini et al.

Muon optimizer outperforms Adam in robustness and transferability, with larger logit margins and higher spectral rank in features.

Tianyu Ruan, Fengzhuo Zhang, Shuche Wang et al.

Reasoning Arena improves reasoning models via adaptive routing and trace tournaments, achieving a 7.6% average performance boost.

Han Zhou, Adam X. Yang, Laurence Aitchison et al.