… Balanced bi-modal SFT gives limited and uneven gains, RLVR provides the first broad improvement, and self-distillation reshapes the benchmark profile. …
Papers arxiv:2605.31535 RayDer: Scalable Self-Supervised Novel View Synthesis from Real-World Video Published on May 29 Submitted by Stefan Baumann on Jun 1 CompVis Authors: , Stefan Andreas Baumann , , Abstract RayDer is a unified feed-forward transformer that consolidates camera estimation, scene… …
… AdaState replaces this static anchor with a self-evolving one. …
Papers arxiv:2604.25154 Prior-Aligned Data Cleaning for Tabular Foundation Models Published on Apr 28 Submitted by Laure Berti-Equille on May 5 Authors: Abstract Deep reinforcement learning framework L2C2 addresses prior mismatch in tabular foundation models by sequentially applying data cleaning o… …
… We introduce the Real-time Autoregressive Video Extrapolation Network RAVEN , a training-time test framework that repacks each self rollout into an interleaved sequence of clean historical endpoints and noisy denoising states. …
Papers arxiv:2606.20404 FlowBender: Feedback-Aware Training for Self-Correcting Conditional Flows Published on Jun 18 Submitted by Daniel Gilo on Jun 19 Technion Israel institute of technology Authors: , , , Abstract FlowBender is a closed-loop framework that addresses constraint satisfaction in di… …
… Cross-Modal Self-Attention CMSA shares keys and values across modalities while keeping modality-specific queries, facilitating information exchange and inter-modal alignment. …
… By aligning intermediate diffusion features with clean-image representations from self-supervised vision encoders , existing methods improve convergence and generation quality. …
… This concatenation inevitably doubles the sequence length, quadrupling the computational complexity of the self-attention mechanism and introducing prohibitive overhead. …
… The following papers were recommended by the Semantic Scholar API Skill is Not One-Size-Fits-All: Model-Aware Skill Alignment for LLM Agents 2026 Decoupling Knowledge and Task Subspaces for Composable Parametric Retrieval Augmented Generation 2026 SkillGraph: Skill-Augmented Reinforcement Learning … …