
What does these pictures look like to you?
Pictures of a person from 2014-2017, right?
No. Actually these pictures are fake. A new technology called Dueling Neuron Network, an advancement in Artificial Intelligence helped creating this.
No. Actually these pictures are fake. A new technology called Dueling Neuron Network, an advancement in Artificial Intelligence helped creating this.
Dueling neural network is a system where two
neuron networks contesting with each other like in a digital cat and mouse
game. It is used to generate audible and visual data. The neural network has produced many pictures like the above ones. They look like real but they aren’t.
GAN (Generative Adversarial Network) created these fake pictures. This particular network system replicates real data to produce fake data.
GAN (Generative Adversarial Network) created these fake pictures. This particular network system replicates real data to produce fake data.

Between two pictures above, left one is generator network
which has the task to create training data and to get as realistic
as it can.
On the other hand, the right one is discriminator network which has the task to identify which one is real and which one is fake. GAN is created through pairing these two models. These two networks then work like a guessing game like which one is real and which one is fake.
Advantages
A sense of imagination
On the other hand, the right one is discriminator network which has the task to identify which one is real and which one is fake. GAN is created through pairing these two models. These two networks then work like a guessing game like which one is real and which one is fake.
Advantages
A sense of imagination
This technology by far contributes
to the society much. It gives machine language, a sense of imagination that can
even fool human beings. For example, NVIDIA team used this system by creating
many photos of celebrities. Those images were fake but hardly
recognizable. With this technology we can change everything, even images to
drawing.
Generate training data
Generate training data
Synthetic databases for medical
imaging and autonomous vehicles are being created with the help of this
technology. Along with that such databases can train other
neural network for object detection or image recognition.
No supervision trouble
No supervision trouble
This system simply doesn’t need
any supervision. So there isn't any trouble to work with this technology.
Disadvantages
1. This
technology can easily be used to create fake news
2. Sometimes
it can give a poor quality result.
3. In
case of privacy, such as health care data, this technology can be vulnerable.
(c) Humaira Mostafa Jasica
(c) Humaira Mostafa Jasica
Comments
Post a Comment