EXPLORE / 06
Terms index
Specialized AI and software-development terms with concise explanations
Entity records
4517 records · 72/113
- 2841
AVX512
An Intel extension supporting 512-bit wide vector operations for accelerated computing.
Website pending - 2842
VNNI
CPU instruction set extension designed to accelerate integer-quantized neural network inference
Website pending - 2843
Pretrained Model
Model pre-trained on large datasets to learn general features
Website pending - 2844
oneAPI Deep Neural Network Library
Intel's library for accelerating deep learning inference and training
Website pending - 2845
oneDNN
An open-source math kernel library provided by Intel for optimizing deep neural networks.
Website pending - 2846
EDA
Initial investigation of data characteristics via visualization and statistics
Website pending - 2847
Dataset Documentation
Standardized metadata describing a dataset's source, structure, and intended use.
Website pending - 2848
normalized Pointwise Mutual Information
A normalized statistical measure quantifying the strength of co-occurrence between two words.
Website pending - 2849
ML-Agents
An open-source toolkit from Unity for training intelligent agents.
Website pending - 2850
Latent Bottleneck
An architectural design that compresses high-dimensional inputs into a fixed-size latent representation to decouple input scale from computation cost
Website pending - 2851
Cross-Attention
An attention mechanism allowing query vectors to attend to key-value pairs from different sources
Website pending - 2852
ML Interface
Tools for rapidly building frontends to interact with and showcase ML models
Website pending - 2853
N-gram LM
A statistical language model predicting the next word based on the previous n-1 words
Website pending - 2854
WER
Metric measuring ASR accuracy by summing substitutions, insertions, and deletions
Website pending - 2855
LM-boosted decoding
Technique introducing an external language model during decoding to correct probability distributions
Website pending - 2856
Sliding Window
A strategy processing continuous data streams by moving a fixed-size window step-by-step.
Website pending - 2857
CTC
A loss function for sequence labeling that allows output length to differ from input without explicit alignment.
Website pending - 2858
Bidirectional Encoder Representations
A language model architecture that considers both left and right context simultaneously.
Website pending - 2859
Constrained Decoding
Applying rules during generation to ensure output meets specific syntactic, lexical, or semantic requirements
Website pending - 2860
Beam Search
A heuristic search algorithm that retains the top-k most probable sequences to balance efficiency and quality
Website pending - 2861
Multimodal Retrieval
Technology matching similarity across different modalities like text and images
Website pending - 2862
Semantic Segmentation
Task of classifying each pixel in an image
Website pending - 2863
Ethical AI
Ensuring AI systems are designed and used in accordance with moral standards and social values
Website pending - 2864
Offline Reinforcement Learning
Training policies using pre-collected datasets without real-time environment interaction
Website pending - 2865
AI Ethics
Study of moral principles, fairness, and social impacts in AI development and application
Website pending - 2866
Data Bias
Systematic errors or imbalances in training data that cause unfair model outputs
Website pending - 2867
Code Abstraction
A programming technique that hides complex implementation details behind simpler interfaces
Website pending - 2868
DRY Principle
A software development principle aiming to reduce repetition of software patterns
Website pending - 2869
CO2 Equivalent
A metric converting various greenhouse gases into their carbon dioxide equivalent impact
Website pending - 2870
AI ASIC
Integrated circuits custom-designed for AI workloads, such as Gaudi.
Website pending - 2871
Carbon Footprint
Total greenhouse gas emissions generated during computation or training processes
Website pending - 2872
BERT
Bidirectional Encoder Representations from Transformers for NLP tasks
Website pending - 2873
Model Reliability
The ability of a model to consistently perform tasks correctly under specific conditions.
Website pending - 2874
Model Trustworthiness
User confidence in the accuracy, fairness, and robustness of model outputs.
Website pending - 2875
Data Representativeness
The degree to which training data distribution mirrors the real-world application environment.
Website pending - 2876
Fully Sharded Data Parallel (FSDP)
A technique that shards model parameters, gradients, and optimizer states across GPUs to save memory
Website pending - 2877
Table Question Answering
NLP task extracting answers from structured tables
Website pending - 2878
Q-Learning
A model-free reinforcement learning method that finds the optimal policy by updating state-action value functions
Website pending - 2879
Deep Q-Learning (DQN)
A reinforcement learning algorithm combining deep learning with Q-learning to handle high-dimensional state spaces
Website pending - 2880
IPU
A processor architecture designed by Graphcore specifically for machine learning workloads.
Website pending