> For the complete documentation index, see [llms.txt](https://ark-defai.gitbook.io/ark/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://ark-defai.gitbook.io/ark/5.-layer-2-or-ark-consensus-algorithmic-layer/5.3-model-structure-or-computational-module-cluster.md).

# 5.3 Model Structure | Computational Module Cluster

ARK's consensus layer is not just a single model, but an AI collective system combining multiple neural architectures and learning paths, mainly including:

* **LSTM Time Series Predictor** — Long Short-Term Memory model, specialized in predicting market price trends, premium fluctuations, and price trajectory directions under policy changes.
* **Transformer Decoder** — Deep semantic model, used to understand governance proposal content, distinguish semantic structures, and generate execution recommendations and anomaly alert messages.
* **Reinforcement Learning Decision-maker (RL Agent)** — Simulates all possible parameter combinations in the policy space, and through reward functions, self-learns the optimal module configurations and adjustment strategies.
* **Anomaly Detection and Sentiment Network** — Used to capture community changes, voting fluctuations, investor sentiment, and early signals of FOMO, panic, trust collapse, etc., for defensive deployment.