IA Neurosymbolic AI use cases
Neurosymbolic AI has been in operation for over 10 years. Our teams have deployed artificial intelligence solutions in different sectors. On a daily basis, this AI processes billions of data to provide decision support in real time.
Here 4 use cases: AI for healthcare, AI for security, AI for aircraft et AI for supply chain.
Healthcare AI
IA supports resuscitation service
Needs:
Improve the management of head trauma, particularly during the crucial first 3 days.
Analyze Acute Respiratory Syndrome Sepsis.
Provide preventive maintenance with detection of faulty ventilator valves.
Tools and technologies:
Neuro-symbolic neural networks (knowledge base).
Decision trees.
Data weighting with degradation of the parameter value over time (fuzzy logic).
Results for 10 years:
Real-time signal analysis system with 254 digital and analog real-time signals per patient for 50 patients (1 md of data / D).
Helps in the management and diagnosis of head trauma patients via the monitoring of intracranial pressure. Proposal for a decision within 1.5 sec.
Real-time monitoring vs. monitoring smoothed over several minutes.
Measuring the impact of exogenous inputs (eg visit from a loved one)
Identification of variations between modeled Acsos and practical reality.
Asynchronous loops to provide decision support with parcel data (and evolution of the score according to the data flow).
Cardiac arrest detected 30 minutes before it happens.
Security AI
Security unit monitoring by AI
Needs:
Identify nuisance alarms and alarms not taken into account by the operators.
Simplify / automate decision making.
Make the most appropriate decisions based on the data: send an email to the customer, visit him, etc.
Distinguish an intrusion alert from a false alert (power cut, shutters not closed)
Tools and technologies:
Neurosymbolic neural networks (knowledge base).
Decision trees.
Self-learning.
Project:
6 months of implementation.
Company resources: 4 days of workshops spread over 1 month to understand their expectations.
Results:
Before:> 500,000 alarms days to be managed humanely
After: <20 alarms days to be managed humanely
Automatic control of nearly 500,000 alarms / day
– Decision-making and management of actions
– Automatic SMS / team sending, call center switch
– Rejection of false alarms by self-learning
– Human treatment alert (legal reason)
Aeronautical AI
4th generation fighter aircraft emergency landing by AI
Needs:
Proof of Concept design
Develop the AI-based Decision Support Demonstrator (POC) for Emergency Landing in Distress – Reactor Loss.
Tools and technologies:
Neuro-symbolic neural networks (knowledge base) with integration of business data (pilot experience).
Decision trees.
Project:
2 months of implementation.
Results:
Simultaneous consideration of more than 100,000 parameters:
– Fuel
– Cartography
– Incident type
– Wind…
Modeling and integration of “business” shortcuts
AI on board the aircraft without external call (vs. F35)
Real-time decision making to define the airstrip and manage flight parameters
Supply Chain AI
Mass retailer supply chain management by AI
Needs:
Reduce inventory costs (fixed inventory and transport costs) while ensuring against stock shortage and production and distribution shutdown.
Data update every hour
Tools and technologies:
Neuro-symbolic neural networks (knowledge base).
Decision trees.
Self-learning.
Project:
6 months of implementation.
Company resources: 4 days of workshops spread over 1 month to understand their expectations.
Results:
Modeling of business rules (finance department, sales department, central workshop).
Definition of alert thresholds.
Automatic management of multi-criteria order intake:
– Safety threshold
– Financial data (cost of storage)
– Supply channels (boat / plane / road) according to time, volume and cost
– Internal shop consumptions feedback
– Sales follow-up and forecasts
Management of black swans (unlikely high impact cases) such as the Suez Canal accident.
Decision support on orders with a 25% reduction in inventory costs.
