Human-like Thinking in Machines : Vision and Language Implementations

EasyChair Preprint 1504

Abstract

The human brain is a complex structure responsible for thinking skills. Several parts of the brain work together to integrate information and develop thoughts. Prefrontal Cortex, the frontal lobe where majority of thinking related processes happen in the brain. A PFC( mimicking the human Prefrontal Cortex ) implementation has been proposed separately for both vision and language system. A FC-LSTM algorithm is proposed for vision system which can be used for image classification and sequence prediction tasks. However, we have removed components unsupported by neuroscience from AI architecture such as error back propagation and used Fully Connected module instead of CNN. A LSTM algorithm is proposed for language system for text generation and sequence prediction. A sequential model with a linear stack of layers consisting of last layer as a fully connected layer without softmax activation is used. The test results show that the algorithms converge and with low prediction accuracy of image classification and  text generation.

Keyphrases: Long Short-Term Memory, Prefrontalcortex(PFC), human-like thinking

@booklet{EasyChair:1504,
  author    = {Poondru Prithvinath Reddy},
  title     = {Human-like Thinking  in  Machines : Vision and Language Implementations},
  howpublished = {EasyChair Preprint 1504},
  year      = {EasyChair, 2019}}