Wiki

POET is a novel reparameterized training algorithm

NSA combines algorithmic innovation and hardware optimization to achieve efficient long-context modeling.

Embodied navigation optimizes navigation routes by interacting with the physical world through moving objects.

DiC is a diffusion model architecture baseline that combines speed and performance

PCEvolve is a novel API-assisted algorithm

EBTs are a promising new paradigm that can simultaneously expand the learning and thinking capabilities of models.

D-MoLE is a novel method designed for continuous multimodal instruction fine-tuning

M+ significantly improves the ability to retain long-term information

AI Flow improves the intelligence, responsiveness and accessibility of AI services

SparseMM prioritizes and preserves visual semantics during decoding

MAS is a computing system consisting of multiple agents interacting in an environment.

CTC is a loss function and modeling method widely used in sequence-to-sequence learning tasks.

The shared knowledge set of the search tree is a search algorithm proposed by Google DeepMind

VeBrain enables robots to “see, think, act”

BCM plays an important role in promoting the development of industrial defect detection technology.

Machine forgetting meets needs such as privacy protection, legal requirements or copyright protection.

Grafting is a simple way to edit a pre-trained Diffusion Transformer (DiT).

PENCIL aims to allow large models to dynamically erase unnecessary intermediate results during the generation process until the final answer is obtained.

RAP is a high-resolution image perception plugin based on RAG technology that does not require training.

Word error rate is an important indicator for evaluating the performance of speech recognition or text matching systems.

Speaker Similarity aims to measure whether two speech samples come from the same speaker or how similar the two samples are.

Guided sampling is a technique used to enhance sample quality in generative models, aiming to improve the controllability of generative models.

Shallow self-reflection aims to quickly optimize the performance of the current task or behavior by making local adjustments to the model through immediate feedback.

Multimodal thinking visualization aims to provide a more intuitive and comprehensive display of thinking, decision-making and information processing processes through the collaborative work of multiple different modalities.