Topic Pages

Files

• action-conditioned-time-series-datasets.md - Comparison of temporal datasets with actions, interventions, treatments, or logged decisions for action-conditioned world models.
• ai-north-star-terminology.md - High-level AI goal terms mapped back to measurable time-series and world-model interfaces.
• byte-level-language-models.md - Language modeling directly over bytes or byte-derived latent units.
• causal-time-series.md - Causal structure, reasoning, and synthetic generation for temporal data.
• context-aided-forecasting.md - Forecasting from numeric history plus essential context, especially text-conditioned time series.
• diffusion-language-models.md - Tracker for masked diffusion, continuous-flow, and self-correcting language-model progress.
• distribution-priors-in-self-supervised-learning.md - How SSL anti-collapse and clustering objectives impose embedding-distribution priors, especially under class imbalance.
• digital-world-models.md - Software-defined world models over code, web, GUI, games, and operational digital systems.
• efficient-recurrent-sequence-models.md - Mamba-family SSMs and ParaRNN-style nonlinear recurrent models as efficient sequence-model backbones.
• energy-based-models.md - Energy-based and latent-variable framing for autonomous intelligence.
• evolution-strategies.md - Black-box population optimization as an RL and LLM post-training alternative.
• extra-long-context-time-series.md - Toolbox of sparse attention, compression, recurrent state, memory, hierarchy, and serving mechanisms for extra-long time-series context.
• fast-slow-thinking-robotics-time-series.md - Layered fast/slow system design across robotics, telecom-style signal stacks, and action-conditioned time-series world models.
• foundation-time-series-model-research-agenda.md - Central research agenda and slotting frame for evaluating progress toward foundation time-series models.
• graph-observability-benchmarks.md - Comparison of graph/system observability datasets and RCA benchmarks for ChronoGraph-adjacent experiments.
• gpu-inference-optimization.md - Simulators, emulators, forecast-aware serving control, and related systems papers for GPU LLM inference optimization.
• graph-structure-as-transformer-context.md - Ways to encode graph structure as Transformer context for graph time series and action-conditioned world models.
• high-dimensional-time-series-forecasting.md - Forecasting over multivariate time series with thousands of aligned channels and structured channel dependencies.
• intermediate-layer-representations.md - Cases where the best transferable representation is an internal layer rather than the final model output.
• iterative-dataset-bootstrapping.md - Model-assisted data engines that iteratively expand and repair annotation layers.
• jepa.md - Joint Embedding Predictive Architecture as a recurring predictive-representation pattern.
• latent-action-models.md - Inferring action-like control codes from raw observation transitions when explicit action labels are missing.
• latent-space-predictive-learning.md - Training by predicting future representations rather than raw observations.
• latent-tokenization.md - Learned chunking, concepts, and token alternatives.
• llm-post-training.md - SFT, RL, preference optimization, and ES viewed through post-training objective and parameter-update geometry.
• looped-transformers-and-test-time-memory.md - Universal Transformer descendants spanning looped depth, latent reasoning, recursive small models, residual-stream state, and test-time memory.
• mixture-of-experts.md - MoE as compute allocation and multimodal specialization.
• nepa.md - Next-embedding prediction as a predictive-representation objective adjacent to JEPA.
• number-tokenization.md - Scalar numeric value tokenization and embedding interfaces for time-series and language models.
• observability-time-series.md - Observability metrics, telemetry, and the boundary between passive forecasting and operator-action world models.
• rank-and-flow-methods.md - Rank structure and compression in time-series Transformers.
• representation-collapse.md - Collapse avoidance and embedding-distribution regularization.
• representational-similarity-calibration.md - Null-calibrated protocol for comparing representation spaces across model sizes, layers, modalities, checkpoints, and time-series dependence structures.
• robotics-text-conditioning.md - How robotics models use language instructions, plans, prompts, metadata, and subgoals.
• robotics-time-series-modeling.md - How robotics and physical-intelligence models encode sensor trajectories, actions, time, and attention.
• self-supervised-representation-learning.md - SSL methods for visual and general representation learning.
• latent-state-time-series-modeling.md - Position spine for reading time series as evolving system state rather than only future observation forecasting.
• streaming-latent-state-updates.md - Always-on state update, retained memory, abstain/trigger behavior, and real-time serving contracts.
• synthetic-data-for-time-series.md - Synthetic temporal data for pretraining, alignment, and reasoning.
• tabular-foundation-models.md - Tabular in-context foundation models and their relationship to time-series priors.
• terminology.md - Canonical terminology for the whole wiki, prioritizing time-series and world-model vocabulary.
• time-series-benchmark-hygiene.md - Evaluation caveats for time-series foundation-model benchmarks, leakage, and adaptation modes.
• time-series-classification-foundation-models.md - Classification and representation-learning branch of time-series foundation models.
• time-series-foundation-models.md - Foundation models for forecasting, understanding, generation, and reasoning over time series.
• time-series-generation.md - Time-series generation interfaces spanning prototype prompts, text conditioning, downstream utility, causal generation, irregular continuous generation, and market simulation.
• time-series-scaling-and-efficiency.md - Scaling, compactness, architecture, and inference-efficiency tradeoffs for time-series foundation models.
• tokenizer-transfer.md - Transfer and distillation between tokenization regimes.
• training-dynamics.md - Optimizer, loss, curvature, compression, controlled forgetting, and noise effects that shape model training.
• unified-multimodal-models.md - Models that combine understanding and generation across modalities.
• vision-foundation-models.md - Scaled vision encoders, pixel representations, and dense features.
• vision-language-models.md - Vision-language learning beyond standard autoregressive VLMs.
• world-models.md - Predictive models of environment dynamics for planning and control.