EXPLORE / 06
Terms index
Specialized AI and software-development terms with concise explanations
Entity records
4517 records · 69/113
- 2721
Sparse MoE
A model structure that improves training and inference efficiency by dynamically routing inputs to activate only a subset of expert networks.
Website pending - 2722
Cross-Embodiment Learning
Method transferring skills across robots of different morphologies
Website pending - 2723
Image Generation Foundation Model
A general-purpose model pre-trained to support various downstream visual tasks.
Website pending - 2724
Multimodal Alignment
The process of making data from different modalities correspond in feature space.
Website pending - 2725
Cross-embodiment Training
Joint training on data from diverse robot morphologies to enable generalization across different hardware
Website pending - 2726
Heterogeneous Data Alignment
Process of mapping data from different sources or formats into a unified space
Website pending - 2727
Long-Horizon Autonomous Execution
The ability of a model to maintain goal consistency and logical coherence across multi-step, extended-duration tasks.
Website pending - 2728
Zero-shot Generalization
The ability of a model to perform tasks in unseen scenarios without prior training on them.
Website pending - 2729
Vision-Language Navigation (VLN)
Task where an agent navigates to a target location in a visual scene based on natural language instructions.
Website pending - 2730
Embodied Question Answering (EQA)
An agent navigates through physical or virtual environments to answer questions about them.
Website pending - 2731
Simultaneous Interpretation
The process or technology of translating speech in real-time while the speaker is talking
Website pending - 2732
Streaming Output
Technology that returns results chunk-by-chunk during generation rather than waiting for completion
Website pending - 2733
Continual Pre-Training (CPT)
Further unsupervised pre-training of a base model on domain-specific data
Website pending - 2734
Language World Model (LWM)
A language model trained to explicitly predict and simulate environmental state transitions
Website pending - 2735
Multi-View Geometric Consistency
Ensuring generated images or videos from different angles maintain logical unity in 3D structure and object relationships.
Website pending - 2736
Visual Reference Coding
A technique that generates corresponding frontend code directly from provided images or video screenshots.
Website pending - 2737
MMDiT
Multimodal Diffusion Transformer for image and video generation
Website pending - 2738
Emergent Abilities
New capabilities that suddenly appear when model scale increases, which smaller models lack
Website pending - 2739
Cross-Embodiment Alignment
Enabling data from robots with different morphologies to coexist and mutually enhance in a unified feature space
Website pending - 2740
Cross-Embodiment Transfer
The capability to apply policies learned on one robot morphology to another with different hardware.
Website pending - 2741
OOD Generalization
The model's ability to adapt to new scenarios or tasks not covered in the training data.
Website pending - 2742
Cross-framework Generalization
The ability of a model to maintain stable performance across different software architectures or development frameworks without significant adjustments.
Website pending - 2743
Dual-Stream Diffusion Model
Diffusion architecture processing parallel streams of modalities or features to enhance conditional control
Website pending - 2744
Visual Code Generation
Technology that automatically generates corresponding code based on visual cues from images or videos.
Website pending - 2745
Multimodal Agent
An AI system capable of processing multiple input modalities and autonomously planning task execution.
Website pending - 2746
Trajectory Prediction
Inferring a sequence of future actions or positions based on current state and goals
Website pending - 2747
Robot Embodiment
The specific physical structure, sensor configuration, and kinematic properties of a robot.
Website pending - 2748
Tensor Core
Hardware units in NVIDIA GPUs dedicated to accelerating matrix multiplication and accumulation
Website pending - 2749
Cross-Embodiment Generalization
The ability of a policy model to transfer strategies across different types of robot hardware or embodiments.
Website pending - 2750
Sparse Model
Models that utilize only a fraction of their total parameters during inference to reduce computational overhead.
Website pending - 2751
Linear Attention
Approximation methods reducing attention computational complexity from quadratic to linear.
Website pending - 2752
Gated Delta Network
A linear attention architecture variant combining gating mechanisms with delta updates.
Website pending - 2753
Multimodal Reasoning
The ability of a model to perform logical deduction using combined inputs like text and images
Website pending - 2754
S2TT
A multimodal translation task directly converting source-language speech into target-language text
Website pending - 2755
Gating Mechanism
Neural network component controlling information flow or activation via learnable parameters.
Website pending - 2756
Gated DeltaNet
An efficient sequence modeling method combining linear recurrence with gating mechanisms.
Website pending - 2757
Residual Connection
A mechanism allowing gradients to propagate directly across layers to mitigate deep network training issues
Website pending - 2758
Gated DeltaNet (GDN)
An architecture combining linear attention and gating mechanisms to efficiently compress historical information
Website pending - 2759
Semantic Adherence
The model's ability to accurately understand and execute complex logic in prompts.
Website pending - 2760
Streaming Processing
Processing data as it arrives without waiting for complete input, suitable for real-time voice interaction.
Website pending