In artificial Neural Networks , if we train a network on a set of given data more and more , the network will over-trained and can't generalize its knowledge to new input patterns. if we simplify the problem ,can we consider this network as an Analogue of a OCD patient's brain ? and if answer is yes , then by modelling the mechanism of OCD treatment drugs like Clomipramine ( that is verified by FDA for OCD treatment ) can we repair over-trained Artificial Networks ?
There are techniques to avoid overtraining a deep network, such as dropout, cross validation etc. I think what you suggest sounds interesting but what does it mean mathematically? remember that an ANN Is a mathematical construct.
Dear Vikas , you are right and there are techniques to avoid overtraining , but I think these thechniques are offline and are not efficient for new complicated artificial neural networks e.g. convolutional networks and especially new introduced eSNNs ( evolving Spiking Neural Networks ) . the brain is the ultimate
inspiration for development of new machine learning techniques , on the other hand the structure and electrical dynamics of brain and brain-like networks ( e.g. SNNs ) are very very complicated than that we can prevent it from overtraining by this offline techniques. so I think because of this fact we need to develop more automated techniques to avoid overtraining of recently developed artificial neural networks .
for second section of your reply , clomipramine is a tricyclic antidepressant that is a highly selective inhibitor of serotonin reuptake which raises accessible amount of serotonin for neural circuits in the brain .I think by investigating the effect of serotonin in Post-Synaptic-Potentials of cortical neural circuits , we can model this effect for ANNs especially for more-brain-like ANNs like SNNs .
Hi Soroush, this is a very nice analogy and one I never heard before. So you might follow up on this. That said the idea of relaxing the artificial brain is not entirely new. LSTM (long short-term memory) networks have the ability to forget things. There are many more network types with this and similar abilities. Apart from that I am not sure how to implement the effect of a serotonin reuptake inhibitor. From my experience computer science often sounds like it implements a natural phenomenon exactly but the best tools take nature only as an inspiration. The implementation is most often optimized for a binary computational machine such as pointer networks in the case of DNNs.
Cheers,
Markus
- Badges
- Science topic
- Similar topics
- Biology