EXPLORE / 06
Terms index
Specialized AI and software-development terms with concise explanations
RECORDS
Entity records
3387 records · 63/85
- 2481
Intrinsic Reward
A non-environmental reward signal driven by the agent's internal state uncertainty or novelty.
Website pending - 2482
Random Network Distillation (RND)
An exploration method generating intrinsic rewards by minimizing the difference between a target random network and a predictor network.
Website pending - 2483
Cross-Domain Transfer
Applying knowledge learned in one domain or task to a different but related domain or task.
Website pending - 2484
Energy-Based Model
A generative framework that assigns an energy score to data points rather than explicitly modeling probability distributions.
Website pending - 2485
Offline Learning
Method of training policies using historical data without real-time environment interaction
Website pending - 2486
Model-Based RL
Reinforcement learning approach that learns environment dynamics to aid decision-making and planning
Website pending - 2487
Batch Size
The number of samples used in one iteration of parameter update.
Website pending - 2488
Gradient Noise Scale
A statistical measure quantifying the difference between mini-batch gradient estimates and full-data gradients.
Website pending - 2489
Long-term AI Safety
Research field focusing on existential risks posed by future superintelligence
Website pending - 2490
Computational Complexity
A theoretical measure of resources (time, space) required to solve a specific problem
Website pending - 2491
Robust Classification
The ability of a model to maintain accurate predictions despite input perturbations or noise
Website pending - 2492
Exploration
The process where an agent tries new actions to discover unknown states or rewards.
Website pending - 2493
MARL
A method where multiple agents learn policies via trial-and-error in a shared environment.
Website pending - 2494
Neuron Interaction
The mechanism of signal transmission and collaboration between nodes in a neural network
Website pending - 2495
Capped-Profit Company
A corporate form that allows profit-making but caps returns or directs surplus to social good
Website pending - 2496
Mode Coverage
The ability of a generative model to cover all major modes in the data distribution.
Website pending - 2497
Energy-Based Model (EBM)
A generative model that defines a data distribution via an energy function.
Website pending - 2498
Next Token Prediction
Task of predicting the probability distribution of the next discrete unit based on preceding sequences.
Website pending - 2499
Adversarial Perturbation
Carefully crafted small noise added to inputs to mislead model outputs
Website pending - 2500
Generative Modeling
Machine learning method that learns data distributions to generate new samples.
Website pending - 2501
Sparse Attention
A mechanism that computes attention only between selected positions to reduce complexity.
Website pending - 2502
Collective Action Problem
A situation where individual rational choices lead to suboptimal group outcomes.
Website pending - 2503
Supercomputing
High-performance computing systems for massive complex tasks
Website pending - 2504
Co-adaptation
The process where multiple entities mutually adjust their behaviors during interaction to optimize performance
Website pending - 2505
Emergence
Properties or behaviors arising from complex systems that are not present in individual components
Website pending - 2506
Adversarial Attack
Techniques that use small perturbations to mislead ML models into making incorrect predictions.
Website pending - 2507
Sim-to-Real Transfer
The process of applying models trained in computer simulations directly to robots in the physical world.
Website pending - 2508
Automatic Domain Randomization
A training technique that improves policy robustness by progressively expanding the distribution of simulation environment parameters.
Website pending - 2509
Model Weights
Parameters obtained after neural network training that determine specific model behavior
Website pending - 2510
Staged Release
Strategy of gradually releasing model capabilities or weights to assess risks and refine safety measures
Website pending - 2511
Safety Constraints
Hard conditions limiting agent behavior to prevent harm or rule violations.
Website pending - 2512
Safe Exploration
Strategies to avoid high-risk states during trial-and-error processes
Website pending - 2513
Procedural Generation
Technique using algorithms to automatically generate content or environments rather than manual design
Website pending - 2514
Regularization
Techniques to prevent overfitting, such as weight decay or dropout.
Website pending - 2515
Double Descent
Phenomenon where test error decreases, increases, then decreases again as model complexity grows.
Website pending - 2516
Deep Reinforcement Learning
A framework combining deep learning with reinforcement learning
Website pending - 2517
Dynamic Computational Graph
Defines network structure at runtime, allowing more flexible debugging and control flow
Website pending - 2518
Deep Learning Framework
Software libraries or platforms used to build, train, and deploy neural network models
Website pending - 2519
Verifiability
The ability to confirm the truthfulness of system attribute claims through independent methods or data.
Website pending - 2520
Neural Audio Synthesis
Technology using deep neural networks to learn and reconstruct audio signals
Website pending