All Methods

Core AI methods for learning, reasoning, and intelligent decision-making.

Methods for understanding, generating, and translating natural language.

Techniques for understanding images, videos, and visual content.

Methods for speech recognition, synthesis, and audio understanding.

Methods for video understanding, generation, and temporal analysis.

Models combining text, images, audio, and video inputs.

AI methods for robot perception, planning, and interaction.

Intelligent agents learning through real-world interactions.

Methods for 3D perception, mapping, and spatial reasoning.

Learning from graph-structured data and relationships.

Forecasting and modeling patterns in sequential time-series data.

AI methods advancing scientific research and discovery.

Attention-based neural architecture for sequence modeling and generation.

State space architecture designed for efficient long-context sequence modeling.

State-space models for capturing long-range dependencies.

Neural networks specialized for extracting spatial features from data.

Neural architectures designed for sequential and time-dependent data.

Models that learn representations from graph-structured data.

Architecture that routes inputs to specialized expert networks.

Neural models that learn compact representations by reconstructing inputs.

Generative models using generator-discriminator competition.

Generative architectures that create data through iterative denoising.

Mechanisms that focus on the most relevant parts of input data.

Dense vector embeddings capturing semantic meaning.

Methods that encode sequence order for transformer-based models.

Fully connected neural layers for feature transformation and learning.

Functions introducing non-linearity to improve model learning capacity.

Techniques that stabilize training and improve model convergence.

Connections that improve gradient flow in deep neural networks.

Methods for reducing feature dimensions while preserving key information.

Neural operations that extract local patterns from structured data.

Methods for splitting text into tokens for language model processing.

Training models on large datasets to learn general-purpose representations.

Adapting pre-trained models for specific downstream tasks.

Training models to better understand and follow user instructions.

Extending pre-training using additional domain-specific datasets.

Learning meaningful representations without manually labeled data.

Combining labeled and unlabeled data to improve model performance.

Reusing knowledge learned from one task to solve another.

Training models progressively from simpler tasks to harder ones.

Learning multiple related tasks within a shared model architecture.

Methods enabling models to learn continuously without forgetting.

Transferring knowledge from larger models into smaller, efficient models.

Training sequence models using target outputs as guidance.

Aligning models with human preferences through reinforcement learning.

Preference optimization without reinforcement learning.

Optimizing model behavior using human or synthetic preferences.

Learning reward functions for optimizing machine learning models.

Aligning AI behavior using predefined principles and rules.

Improving models using feedback generated by other AI systems.

Guiding model behavior through carefully designed prompts and instructions.

Reasoning step by step to solve complex problems effectively.

Methods for logical inference, analysis, and decision-making.

Breaking complex goals into structured actions and intermediate steps.

Finding relevant information efficiently across large search spaces.

Improving responses through self-evaluation and iterative refinement.

Using external tools and APIs to complete complex tasks.

Combining reasoning and actions to solve multi-step problems.

Invoking external functions and APIs during model execution.

Maintaining context and recalling information across interactions.

Coordinating multiple intelligent agents to achieve shared goals.

Managing multi-step AI workflows and task execution pipelines.

Standardized communication between AI models, tools, and services.

Combines retrieval with language generation.

Retrieving documents using semantic vector representations.

Keyword-based retrieval using sparse text representations.

Combining dense and sparse retrieval for improved search quality.

Reordering retrieved results based on relevance and quality.

Searching data using vector embeddings and similarity matching.

Structured graphs representing entities and relationships.

Low-rank adaptation for efficient fine-tuning of large models.

Quantized LoRA for memory-efficient model fine-tuning.

Parameter-efficient methods for adapting large language models.

Learning task-specific prompts without updating all model weights.

Optimizing trainable prefix representations for downstream tasks.

Fine-tuning lightweight adapter layers while freezing base models.

Algorithms that update model parameters for efficient and stable learning.

Techniques that adjust learning rates throughout the training process.

Methods for setting initial model parameters before training begins.

Gradient-based optimization for updating machine learning models.

Randomly disabling neurons during training to reduce model overfitting.

Constraining model weights to improve generalization and robustness.

Softening target labels to improve model confidence and accuracy.

Expanding training datasets with transformed or synthetic examples.

Reducing numerical precision for faster and more efficient inference.

Removes unnecessary parameters while preserving performance.

Using sparse representations to reduce computation.

Accelerating text generation through predictive decoding strategies.

Caching attention states to speed up autoregressive model inference.

Memory-efficient attention for serving large language models efficiently.

Highly optimized attention algorithm for faster transformer execution.

Reducing model size while preserving accuracy and performance.

Techniques that improve model inference speed and resource efficiency.

Learning action values to maximize long-term rewards through experience.

Optimizing decision-making policies for reinforcement learning agents.

Learning environment dynamics for better planning and decisions.

Training reinforcement learning agents using fixed offline datasets.

Learning through real-time interactions with environments.

Learning representations by comparing similar and dissimilar samples.

Learning distance metrics that capture semantic similarity between samples.

Creating dense feature representations for downstream tasks.

Automatically discovering useful features directly from raw input data.

Generative models that create data through iterative denoising processes.

Learning continuous transformations for efficient generative modeling tasks.

Generative models that learn score functions over data distributions.

Fast generative models designed for high-quality one-step sampling.

Detecting and localizing objects within images and video scenes.

Partitioning images into regions for detailed understanding

Generating realistic images from text, noise, or other visual inputs.

Classifying images into predefined categories using deep learning models.

Estimating human or object poses from images and videos.

Following objects consistently across multiple frames in video sequences.

Understanding 3D scenes, geometry, and spatial relationships.

Splitting text into tokens for efficient language model processing.

Learning language patterns for text prediction and generation tasks.

Translating text accurately between different natural languages.

Generating coherent and contextually relevant natural language text.

Producing concise summaries while preserving essential information.

Answering questions using context, documents, or structured knowledge.

Extracting structured facts and entities from unstructured text.

Converting spoken language into accurate text using AI models.

Generating realistic and natural-sounding speech from text inputs.

Identifying and verifying speakers from voice characteristics.

Generating speech, sound effects, and other synthetic audio content.

Creating original music compositions using generative AI models.

Analyzing video content to recognize events, actions, and scene dynamics.

Generating realistic videos from text, images, or learned visual representations.

Segmenting objects and regions consistently across video frames.

Finding relevant videos using semantic search and similarity.

Planning safe and efficient robot movements in complex environments.

Controlling robotic interaction through grasping and manipulation.

Navigating autonomous robots in dynamic environments.

Learning robot control policies from experience, demonstrations, or rewards.

Neural rendering method for reconstructing realistic three-dimensional scenes.

Fast technique for high-quality rendering of complex 3D scenes.

Simultaneous localization and mapping for robotic perception and navigation.

Representing 3D environments using spatial point collections.

Mathematical principles for optimization in machine learning.

Modeling uncertainty and randomness with probabilistic frameworks.

Analyzing data distributions, patterns, and relationships through statistical methods.

Functions that measure prediction errors during model optimization.

Foundation for vectors, matrices, and neural network computations.

Measuring model accuracy, quality, and performance using evaluation metrics.

Assessing AI outputs through human judgment and expert review.

Using language models to evaluate responses and compare output quality.

Comparing model outputs using human or AI preference judgments.

Evaluating models against standardized datasets and benchmark tasks.

Studying internal model mechanisms to understand learned behaviors.

Identifying which inputs contribute most to model predictions.

Using diagnostic probes to analyze learned representations.

Making AI decisions more transparent and understandable for users.

Reducing inaccurate or fabricated content generated by AI systems.

Embedding identifiable markers into AI-generated content for verification.

Ensuring AI systems behave safely and follow intended objectives.

Protecting models against prompt injection and jailbreak attacks.

Improving model reliability under noisy, adversarial, or changing conditions.

Protecting sensitive data throughout model training and deployment.

Scaling model training across multiple machines and computing devices.

Executing model computations to improve training efficiency.

Deploying AI models for reliable, large-scale production inference.

Systems optimized for fast and scalable model inference workloads.

Reducing memory usage while maintaining efficient model performance.

Graphics processors optimized for large-scale AI training and inference.

Google-designed accelerators for machine learning workloads.

Dedicated neural processors for efficient on-device AI computation.

Specialized hardware designed to speed up AI model execution.

Studying how model performance improves with increased scale and data.

Unexpected capabilities that appear as AI models become larger.

Learning new tasks from examples provided directly within prompts.

Using extra computation during inference to improve performance.

Optimizing prompts and context to improve model responses consistently.

Learning internal environment representations for planning.

Large pre-trained models adaptable to diverse downstream AI applications.