International Partner Search

Innovation & Technology Offer

COVID-19 A path towards robust, efficient and trustworthy deployment of AI models in enterprises/organisations

Country of Origin: Italy
Reference Number: TOIT20200325001
Publication Date: 26 March 2020

Summary

An Italian high-tech startup, specialised in building products and services in the field of Explainable and Trustworthy Artificial Intelligence (AI), has developed a serving library/API that can be retrofit to AI models to get human centric explanations,detect anomalies in the AI output, improve efficiency, robustness and the overall trust in AI models outputs.Technical and licensing agreements with firms, hospitals, research groups working on the COVID crisis are sought.

Description

An innovative startup based in Torino, Italy specialised in building products and services in the field of Explainable and Trustworthy Artificial Intelligence (AI) has developed a serving library/API that can be retrofit to AI models to get human centric explanations, detect anomalies in the AI output, improve efficiency, robustness and the overall trust in the outputs of AI models. It is also intended to be a monitoring tool to keep track of the performance of the AI models. The company offers technology cooperation pro bono for companies, hospitals, research groups working on the Covid crisis for non-commercial purposes.

The company is interested in licensing agreements for all other use cases and companies.



The product is designed as an MLOps (Machine Learning Operations) solution allowing ML (Machine Learning) engineers/ data scientists to package a model by creating an augmented API for model serving and a database for continuous monitoring.



- The augmented API is generated by using  proprietary eXplainable AI libraries. This API is designed to:



- Enrich single queries with local explanations of the output, enhancing interpretability of the model decisions. 



- Explanations are presented in the form of historical examples, decision rules and counterfactual examples.



- Create an intuitive feedback mechanism to monitor the model performance over time. 



- Perform robust uncertainty quantification of the model answers to increase trust in each recommendation.



- Perform anomaly detection of live data and report adversarial attacks.

Advantages and Innovations

- Efficient querying, reducing time for getting model explanations and insights



- Non-intrusive, the company doesn't look into the inner workings of the ML models



- Can be used for different types of AI models



- A handy tool for putting AI models into production either locally or on cloud



- An easy to use interface for data scientists and ML Engineers

Stage Of Development

Available for demonstration

Stage Of Development Comment

The Minimum Viable Product will be ready for early adoption from 31 March 2020. It currently works with most AI models using tabular data, and will be ready for time-series and image data  within the next three to four weeks. Our AI team is working very hard with an agile methodology to reach this objective.

Requested partner

The start up technology can be applied to any organisation/enterprise that has invested in building AI models and would like to put them in production. However a few example cases where the technology would add most value have been identified. Machine Learning Engineers and Data Scientists working in these fields and similar use cases would benefit most from our offer:

1. Healthcare diagnostics, treatment planning, outcome prediction using Artificial Intelligence: to increase trust and robustness of AI decisions. At the moment the company is most interested in this application since given the COVID emergency, they are willing to contribute pro-bono for putting COVID related AI models into production using our technology.

2. Trustworthy and compliant process automation in Financial services, banking and insurance: e.g. explanations and counterfactual explanations for a mortgage decision made by an AI system

3. Predictive maintenance: Increasing operator trust in the recommendations giving by an AI system

4. Cyber security: Anomaly detection of the prediction of AI models



The company offers technology cooperation pro bono for companies, hospitals, research groups working on the Covid crisis for non-commercial purposes.



The company is interested in licensing agreement for all other use cases and companies.