Overview of Howso’s Trustworthy AI#
Key Features of the Howso Engine#
Perfect attribution traces decisions directly back to the source data, enabling complete auditability
Challenge misleading results from current interpretability approaches with breakthroughs in feature importance analysis
Howso is robust against several common challenges (correlated features, redundant features, difference in scale between features, and multiple distinguishing features) faced by other feature importance tools, including the SHAP metric, which often lead to misleading results.
Fix bias and debug
Easily add, edit, and remove data without spending more costly re-training time and computational resources.
Identify counterfactuals and perform “what if?” scenario analyses
On average, Howso outperforms LightGBM classification and regression accuracy, and demonstrates notable success with small datasets. Additionally, Howso intelligently handles and imputes sparse and missing data.
Perform multiple tasks automatically, including…
Supervised and unsupervised learning
Generative and discriminative analysis
Bias detection and prediction monitoring
Utilize structured and semi-structured data
Flexibly handle a variety of data types, including time-series data and nominals
Calculate reliable uncertainties for your predictions
Avoid hallucinations with complete and accurate knowledge of uncertainties so you can deliver accountability and know when there is not enough information to make good decisions.
Remain robust to adversarial attacks that exploit AI in action, and reliably detect them
Howso outperforms other AI methdos and maintaining the integrity of your data and decisions, even after an attack.
Howso values gracious intellectual honesty. In that spirit, we’re telling you up front where we struggle and how we are planning to improve.
Additional data types
Currently, we only work with structured data. We are in beta testing for semi-structured data, tiptoeing toward generative language, and images handling is on the distant horizon.
We work well with hundreds of features, but we are not yet able to handle thousands of features in practical applications.
Very large datasets
Handling very large datasets with subtle signals (e.g., datasets requiring tens of millions of records and/or thousands of features to capture the complex relationships within the data) currently requires manual work from engineering, data science, and subject matter expert teams. However, currently available Howso tools, including ablation and non-robust feature contribution calculations, can be used to help identify subsamples of large datasets that contain enough signal to be used for data science analysis.
What’s next? How to use these guides…#
If you have not already installed Howso Engine, that’s your first step! You can find our installation guide here.
Once you’re installed, you can try out Howso using a variety of pre-built Jupyter notebook Recipes. These notebooks will provide “recipes” for how to utilize Understandable AI in many applications.
And, as always, we welcome your participation and feedback on our github page!