|
SHARP is the result of a research
started to give answers to some issues raised in the DARPA XAI (eXplainable
Artificial Intelligence) program. |
The tremendous success that Deep-Learning
has achieved in recent years has, however, highlighted serious problems
related to this technology. In addition to the problems due to the ease with
which this technology can be deceived, also a serious problem has been
highlighted: this technology is of the BLACK-BOX type and therefore the
inference cannot be explained. Although the explainability of the decisions
taken by Artificial Intelligence algorithms is of great interest at
international level (just think of the GDPR decree of the EU), it is evident
how much this characteristic of AI algorithms is indispensable in human-to-AI
collaborative scenarios in the military and in the context of safety-critical
missions. Our SHARP™ (Systolic Hebb Agnostic
Resonance Perceptron™) neural model allows you to create rules from the data
and map them to the synaptic weights of the neural network itself. When the
neural network makes a decision it is possible to extract the rule or set of
rules that determined the decision, going back to the detail of the original
data learned and the expert's evaluation added as a range of the single
variable during the training phase. The SHARP algorithm has L2 (Lifelong
Learning) and OSL (One Shot Learning) properties. With our algorithms we can
extract rules also from RBF neural networks with RCE learning algorithm. |
|
©2024_Luca_Marchese_All_Rights_Reserved Aerospace_&_Defence_Machine_Learning_Company VAT:_IT0267070992 NATO_CAGE_CODE:_AK845 Email:_luca.marchese@synaptics.org |
Contacts_and_Social_Media |