CNN Group - AUT

…where we believe, human perception modelling is key to the future of AI!

And this idea is our best source of motivation to perform research in the fields of Cognitive Science, Machine Vision and Machine Learning. Our goal is to pursue the idea of making computational models of the human perception, so to advance the understanding of the relationship between the machinery and the functioning of it.

This group was established in the Computer Engineering and IT Department at AmirKabir University of Technology (Tehran Polytechnic) in July 2015 with a deep focus on understanding Human Visual System structure and properties and applying such ideas into the fields of Cognitive Science, Machine Vision and Machine Learning (e.g. *Convolutional Neural Networks* applied to Image Recognition tasks, ...).

research scopes

Our main research areas of interest include:

  • Behavioral Functional Modelling
  • Visual Saliency
  • Eye Movement Distribution
  • Task-based Bottom-up Visual Attention Modelling
  • Gist Modelling
  • Covert vs. Overt Attention Modelling
projects

We have achieved and planned the following projects…

Convolutional Neural Networks With Attention Layers

Members: Saleh Khazaei, Shiva Zamani, Taher Ahmadi

Start Date: July 2015

A model with a large learning capacity is needed to learn about thousands of objects from millions of images. Convolutional neural networks constitute one such class of models having such capacity. These models are basically a based on retina. In this study we try to reduce the learning parameters by adding attentional layers to the architecture. Attentional layers provide the ability to obtain basic regions of interest.


Task-based Bottom-up Saliency Map Generation

Members: Mojtaba Taherkhani, Saman Malek Ghasemian, Ghazal Zand

Start Date: Aug 2015

Many efforts has been made to simulate the visual attention mechanisms, and several models and algorithms have been proposed during the past few decades, however, relatively few studies have been made to apply these models to the existing biologically inspired models of cognition which are also biologically plausible. Visual attention models rely on bottom-up influences or top-down influences or a combination of both. While there is no clear cut distinction between such influences in human attention. In this research we tend to use a combination of the bottom-up and top-down models. This model uses a task-based approach to approximate the bias of top-down models.


Attenion Based Disparity Map Generation

Members: Saman Malek Ghasemian, Ghazal Zand, Mojtaba Taherkhani

Start Date: Oct 2015

A coordinated motion of the eye and head has often been used as a proxy for attention in natural behavior. For instance, human has to interact with objects and control the gaze to perform a task while moving in the environment. Based on scene type and the type of viewing (time and distance), eye movements differ considerably on images. In this project we tend to propose a new eye movement distribution with the depth being considered as well.


Multi-Object Tracking Based on Saliency

Members: Ghazal Zand, Mojtaba Taherkhani

Start Date: Nov 2015

Object recognition and tracking has proved to be a cognitive problem. We study a new framework with Bayesian State Prediction for Object Tracking and Saliency Based Region of Interest detection. In this research we tend to provide a Long-term tracking under a flying robot condition which assumes a dynamic environment having moving objects and non-fixed camera.

publications

The following is a list of our academic publications:

2015

  • Taherkhani, Mojtaba, and Reza Safabakhsh. “A novel stability-based adaptive inertia weight for particle swarm optimization.” Applied Soft Computing (2015).

In Progress

  • Long-term vehicle tarcking from flying robot based on particle filter and saliency map.
  • Convolutional Neural Networks with Attentional Layers.
  • A novel convolutional task-based saliency map.
our team

Meet The Team

Mojtaba Taherkhani

Mojtaba Taherkhani

Group Director & Research Assistant

"The great aim of education is not knowledge but action... -Herbert"

Saman Malek Ghasemian

Saman Malek Ghasemian

Research Assistant

"Feature engineering is dead! Here we exploit biological evidences to develop nature inspired models... -Saman! Chapter I of my life!"

Ghazal Zand

Ghazal Zand

Research Assistant

"Everything you see or hear or experience in any way at all is specific to you. You create a universe by perceiving it, so everything in the universe you perceive is specific to you. -Douglas Adams"

Saleh Khazaei

Saleh Khazaei

Research Assistant

"Truth is universal. Perception of truth is not. -Anonymous"

Shiva Zamani

Shiva Zamani

Research Assistant

"A year spent in artificial intelligence is enough to make one believe in God. -Alan Perlis"

Taher Ahmadi

Taher Ahmadi

Research Assistant

"Though wise men at their end know dark is right,
Because their words had forked no lightning they
Do not go gentle into that good night.... -Dylan Thomas"

contact

You can contact and provide us with your ideas and thoughts regarding our projects and researches. Also if you think you can help us with your passion and hard work towards any of our ideas, feel free to contact the group director.