“Success in creating Artificial Intelligence would be the biggest event in human history”.
Sumera B Reshi
Recall Tantalus in Greek mythology, when he was thrown out of Olympus and after he died he was punished for eternity; he was made to stand in a pool of water, right under the branches of a fruit tree. However, when he tried to reach for a fruit, the branches would go higher and out of reach, while when he tried to drink a sip of water, the waters of the pool would recede, so in a way, his hunger or thirst was never fulfilled. Similarly, as machines start learning, their hunger for data is unquenchable.
A data scientist spends days in acquiring and cleaning more and more data to feed machines. And their nights are lost in teaching machines learning from all this data, by training models over and over again. Can machines perform tasks like humans? Can machines learn to teach themselves? Nay, this can’t be a pun and in the present world, such a task is doable.
The approach, known as a generative adversarial network, or GAN, takes two neural networks, the simplified mathematical models of the human brain that reinforces most modern machine learning—and pits them against each other in a digital cat-and-mouse game. GAN was experimented by Ian Goodfellow, then a PhD student at the University of Montreal, during an academic argument in a bar in 2014.
The two networks are trained on the same dataset. One, known as the generator, is undertaken with creating variations on images it’s already seen—perhaps a picture of a pedestrian with an extra arm. The second, known as the discriminator, is asked to identify whether the example it sees is like the images it has been trained on or a fake produced by the generator—basically, is that three-armed person likely to be real?
Duel Neural Networks
If Deep learning is the next big thing that’s taking the cake, GAN is the cream on that cake. The possibilities have never so exciting! Ian Goodfellow that night invented is called a GAN, or “generative adversarial network.” The technique has sparked huge excitement in the field of machine learning and turned its creator into an AI celebrity.
Data scientists were already using neural networks, algorithms loosely modelled on the web of neurons in the human brain, as “generative” models to create plausible new data of their own. But the results were often not very good: images of a computer-generated face tended to be blurry or have errors like missing ears.
The plan Goodfellow’s friends were proposing was to use a complex statistical analysis of the elements that make up a photograph to help machines come up with images by themselves.
This would have required a massive amount of number-crunching. However, deep-learning AIs can learn to recognize things, they can’t be good at creating them. The main aim of GAN, however, is to give machines something akin to an imagination. By doing so, will enable to draw nice pictures or compose a great piece of music, thus making them less reliant on humans to instruct.
Till date, programmers feed information into a machine according to their needs, such an operation is expensive, labor intensive and limits how well the system works or deals with the data it was fed with. In near future, computers will get better at gobbling up on raw data and working out what they need to learn from it being told.
Goodfellow’s piece of research will mark a big leap known as “unsupervised learning”. For instance, a self-driving car could teach itself the different road conditions and a robot could anticipate the barriers it might encounter in a busy warehouse without needing to be taken around it.
GANs’ potential is huge because they can learn to mimic any distribution of data. GANs can be taught to create worlds eerily similar to our own in any domain: images, music, speech, prose. They are robot artists in a sense, and their output is impressive, poignant even.
The main advantage of generative adversarial networks (GANs) is it produces very realistic images of faces, chairs, and animals.
The reason for this is that the objective of GANs to generate artificial data that is indistinguishable from real data by another neural net – is highly aligned with the goal of producing realistic data. But the big disadvantage is that these networks are very hard to train, according to Ian Goodfellow.
- Indian Youth In Web Trap
- As Infertility Among Women Grows, Scientists Develop Artificial Embryo
- Professor Stephen Hawking And The World Of Science