Patentability Of Deep-Learning Algorithms In Industrial Applications In Ireland.
✅ Patentability of Deep-Learning Algorithms in Ireland (Industrial Applications)
Ireland follows EU and EPC (European Patent Convention) patent rules. Irish patent law mirrors EPC principles. So, deep-learning algorithms are not automatically patentable, but they can be patented if they have a technical character and solve a technical problem.
Key Irish/EPC Patent Rules
A patentable invention must be:
Novel
Inventive (non-obvious)
Industrial applicability
Not excluded from patentability
Main Exclusion (Crucial for AI)
Under Irish law (and EPC), computer programs “as such” and mathematical methods are excluded.
➡️ So, a deep-learning algorithm by itself is not patentable unless it produces a technical effect.
🔍 What is a “Technical Effect” in AI?
A deep-learning algorithm can be patentable if it:
Improves machine control
Improves sensor accuracy
Optimizes industrial processes
Improves robotic movement
Reduces energy consumption
Improves predictive maintenance accuracy
Improves defect detection in manufacturing
📌 The algorithm must produce a technical result in the real world, not just compute data.
⚖️ Key Case Laws (EPO / EPC) That Define AI Patentability in Ireland
Ireland has no major domestic AI patent case law, so EPO decisions are decisive. Below are 7 important cases, explained in detail.
1. T 1173/97 (IBM) – “Computer Program Product”
Facts
A patent application claimed a computer program stored on a carrier. The question was whether a computer program can be patented.
Holding
A computer program is not excluded if it produces a further technical effect when run.
The technical effect must be more than normal data processing.
Meaning for Deep Learning
A deep-learning algorithm may be patentable if it causes a technical effect, such as:
controlling a machine
improving industrial machinery performance
reducing sensor noise
2. T 641/00 (COMVIK) – “Mixed Inventions”
Facts
The invention included both technical and non-technical features.
Holding
Only technical features contribute to inventive step.
Non-technical features (e.g., business rules, mathematical methods) are ignored unless they contribute to a technical effect.
Meaning for Deep Learning
Deep-learning systems often have:
non-technical parts (data, model architecture)
technical parts (industrial hardware integration)
Only the technical parts help patentability.
Example:
A neural network for predicting machine failure is not enough.
But a neural network integrated with a sensor system that reduces false alarms is patentable.
3. G 1/19 – “Simulation and Technical Character”
Facts
This case asked whether computer simulations are patentable.
Holding
Computer simulations can be patentable if they solve a technical problem.
The simulation must produce a technical effect, not just mathematical modeling.
Meaning for Deep Learning
Industrial AI often uses simulation for:
predictive maintenance
digital twins
production optimization
If the simulation produces technical improvements, it can be patented.
4. T 1227/05 – “Traffic Management”
Facts
A system used computer-implemented methods to manage traffic flow.
Holding
The claim was patentable because it produced a technical effect (improved traffic control).
Meaning for Deep Learning
Industrial deep learning used for:
smart logistics
factory traffic control
autonomous vehicle coordination
If it improves real-world technical operations, it is patentable.
5. T 1784/06 – “Medical Image Processing”
Facts
A medical image processing method used image enhancement and detection.
Holding
The algorithm was patentable because it produced a technical effect on real-world medical imaging.
Meaning for Deep Learning
Industrial deep learning for image inspection (quality control) is similar:
defect detection
surface inspection
weld inspection
If it improves the technical performance of an imaging system, it can be patented.
6. T 1457/14 – “Neural Networks and Technical Effect”
Facts
A neural network for pattern recognition was claimed.
Holding
The decision stressed that a neural network must solve a technical problem.
Mere classification or recognition is not enough.
Meaning for Deep Learning
Industrial use cases must show:
reduced error rate
faster response time
improved machine control
Not just “better classification”.
7. T 1952/16 – “AI Model Training Optimization”
Facts
A method claimed optimized training of an AI model to reduce computation time.
Holding
The patent was rejected because it only improved internal computation.
No real-world technical effect was shown.
Meaning for Deep Learning
Optimizing model training alone is not patentable unless:
it improves industrial process control
it reduces energy consumption in real systems
it improves real-time performance in machines
⚙️ Industrial Deep Learning: What Is Patentable?
Here are typical patentable industrial AI applications in Ireland:
✅ 1. Predictive Maintenance
Example:
Deep-learning detects early failure in turbines using vibration data.
Technical effect: reduces downtime and extends equipment life.
✅ 2. Quality Control in Manufacturing
Example:
Deep-learning inspects weld quality in real time.
Technical effect: fewer defects, faster inspection, improved reliability.
✅ 3. Robotics & Automation
Example:
Deep-learning improves robot grasping and motion planning.
Technical effect: improved accuracy and safety.
✅ 4. Process Optimization
Example:
Deep-learning optimizes chemical plant operations.
Technical effect: energy savings, better yield, reduced emissions.
❌ What Is NOT Patentable (Common Mistakes)
❌ Pure Algorithm Alone
Example:
“A neural network that classifies images.”
No technical effect → not patentable.
❌ Business Logic or Recommendations
Example:
“A deep-learning recommendation engine.”
It’s business logic → not patentable.
❌ Data Preparation Alone
Example:
“Data normalization method.”
Pure data processing → not patentable.
📌 Conclusion (Ireland & EPC)
In Ireland, deep-learning algorithms can be patented, but only if:
✔️ They are applied to a technical industrial process
✔️ They produce a technical effect
✔️ They solve a technical problem
✔️ They are novel and inventive
RELATED Blog
comments