Introduction

Example where deep neural network activation path information is captured and included in the Case data design.

The Tikos Python library provides functions to build and run the Tikos Reasoning Platform and generate transparency and explainability logs for model outputs.

💎

Transparency

Build - generate Cases

1. Select trained model (any class of standard machine learning, deep-learning, or LLM)
2. Run test set through trained model (to create the initial Case-base)

Run

1. At inference time new Cases are created, added to the Case-base and analysed to produce:
2. Transparency insights such as:
   1. Neural activation path data
   2. Similarity, proximity and adaptation data
   3. Causality data

🧊

Extend for explainability

Build - generate Context

1. Combine model features with additional problem domain information
2. Review, edit, and approve dynamically created knowledge graph

Run

1. At inference time, the model output (Case data) is explained by reasoning over the contextual information (knowledge graph) to produce:
2. Reasoning and explainability insights such as:
   1. Comparisons to known prior problem/solution pairs
   2. Adaptations and causality data at both the Case-item and whole Case level
   3. Mathematical and natural language outputs

🚧

The library is intended for evaluation and prototyping only

Please keep in mind:

• Your data will be sent to and processed on our servers. (Production deployments are in our roadmap, where you can install Tikos in your environment and retain control of all data processing.)
• The package utilizes LLM functionality. In every case we only use open-source models, no data is transmitted to commercial model providers.
• The code is in active development and will likely have bugs.

For production deployments join our waitlist by emailing [email protected].