The model approximates the Gödel Machine in a coding agent environment and guides the expansion through Thompson sampling with adaptive scheduling.

The first framework to successfully apply distribution matching distillation to MDM-based text generation, setting a record in few-step language sequence generation.

MultiPL-MoE is an effective method for extending low-source programming languages in the post-pre-training stage.

The Tongyi Qianwen team systematically studied the role of gating mechanisms in standard softmax attention.

The Lancelot framework incorporates fully homomorphic encryption into BRFL to achieve robust privacy protection.

By jointly aligning global and local features, adversarial examples can be effectively guided toward the target feature distribution and transferability can be enhanced.

The receptive field is an important concept for understanding visual information processing and provides a reference for designing, analyzing and optimizing visual models.

SVG enables faster diffusion training, efficient few-step sampling, and improved generation quality.

RewardMap enhances the capabilities of multimodal large language models in structured vision tasks.

A novel principle-based discriminative constraint optimization framework avoids difficulty bias and training instability.

ReinFlow features a lightweight implementation, built-in exploration capabilities, and broad applicability to various streaming strategy variants.

FHE is widely used in scenarios such as cloud computing security, federated learning, medical data analysis, and financial data collaboration.

BRFL is designed to address the Byzantine attack problem that occurs during model aggregation.

EGMN successfully captured the potential interaction effects between user preferences and video features.

SAC Flow achieves state-of-the-art performance in continuous control and robot operation benchmarks.

UserBench aims to assess and enhance an agent’s ability to understand, interact with, and adapt to real-world user communication.

PLACER is fast and stochastic, and can easily generate prediction sets to map conformational heterogeneity.

With its significant advantages, RAE is poised to become the new default choice for training diffusion Transformers.

Given the limitations of existing fine-tuning techniques such as GRPO, GVPO has emerged as a reliable and versatile post-training paradigm.

ReCA has generalization capabilities in terms of application scenarios and system scale, and the success rate of tasks has been improved by 4.3%.

DexFlyWheel is a scalable and self-improving data generation paradigm for agile operations.

NovaFlow is able to handle rigid, articulated, and deformable objects in different robot forms.

TreeSynth demonstrates exceptional robustness and scalability in large-scale data synthesis.

GTA significantly outperforms standard SFT baselines and state-of-the-art RL methods in multiple text classification benchmarks.

The model approximates the Gödel Machine in a coding agent environment and guides the expansion through Thompson sampling with adaptive scheduling.

The first framework to successfully apply distribution matching distillation to MDM-based text generation, setting a record in few-step language sequence generation.

MultiPL-MoE is an effective method for extending low-source programming languages in the post-pre-training stage.

The Tongyi Qianwen team systematically studied the role of gating mechanisms in standard softmax attention.

The Lancelot framework incorporates fully homomorphic encryption into BRFL to achieve robust privacy protection.

By jointly aligning global and local features, adversarial examples can be effectively guided toward the target feature distribution and transferability can be enhanced.

The receptive field is an important concept for understanding visual information processing and provides a reference for designing, analyzing and optimizing visual models.

SVG enables faster diffusion training, efficient few-step sampling, and improved generation quality.

RewardMap enhances the capabilities of multimodal large language models in structured vision tasks.

A novel principle-based discriminative constraint optimization framework avoids difficulty bias and training instability.

ReinFlow features a lightweight implementation, built-in exploration capabilities, and broad applicability to various streaming strategy variants.

FHE is widely used in scenarios such as cloud computing security, federated learning, medical data analysis, and financial data collaboration.

BRFL is designed to address the Byzantine attack problem that occurs during model aggregation.

EGMN successfully captured the potential interaction effects between user preferences and video features.

SAC Flow achieves state-of-the-art performance in continuous control and robot operation benchmarks.

UserBench aims to assess and enhance an agent’s ability to understand, interact with, and adapt to real-world user communication.

PLACER is fast and stochastic, and can easily generate prediction sets to map conformational heterogeneity.

With its significant advantages, RAE is poised to become the new default choice for training diffusion Transformers.

Given the limitations of existing fine-tuning techniques such as GRPO, GVPO has emerged as a reliable and versatile post-training paradigm.

ReCA has generalization capabilities in terms of application scenarios and system scale, and the success rate of tasks has been improved by 4.3%.

DexFlyWheel is a scalable and self-improving data generation paradigm for agile operations.

NovaFlow is able to handle rigid, articulated, and deformable objects in different robot forms.

TreeSynth demonstrates exceptional robustness and scalability in large-scale data synthesis.

GTA significantly outperforms standard SFT baselines and state-of-the-art RL methods in multiple text classification benchmarks.