Not So Deep Learning… For Now

Machine Learning. The epitome of Computer science and the idea that can make or break our society in the next couple decades. What is it? How does it affect me? Think about it. How amazing would it be to make a computer program that can tell the difference between a cat and a dog or even speak back to you in a semantically intelligent way.

Well guess what, it’s possible! Today in our society there are numerous programs that can perform intelligent tasks more accurately than humans. It’s crazy to think that a computer has the potential to become more “intelligent” than a human.

Deep Learning. The sub field of Machine Learning that centralizes around the idea of using artificial neural networks to solve problems. You’ve probably heard about “neural networks” somewhere in the last couple years. They are taking the digital world by storm by performing tasks at accuracy levels even better than humans.

Some examples of such networks include image classifiers that can distinguish between multiple objects, object detection models that can effectively localize on specific objects, natural language processing models that can take a semantic understanding of what a user says, etc. The list goes on and on but… what are the capabilities of these amazing ideas? Is it possible for a computer to make rational decisions or communicate in a dynamic way?

Well for now… no. The human brain is the most complex organ in the human body. Fine tuned by millions of years of evolution, the human brain can perform multiple tasks that seem impossible for a computer. Neural networks were created to mimic the human brain and “learn” in the way we believe humans learn. So, what’s the problem then? If neural networks are supposed to mimic how the human brain works, why can’t they interact with us like humans?

The human brain consists of approximately 100 billion neurons. To put this into perspective, the vgg19 image classification and detection model takes approximately 12 hours–3 days depending on how powerful your computer is. An image classifier takes 12 hours–3 days to train! And that’s only when you have all the data to train the classifier (~15,000,000 images). Think about how long it would take to train a network approximately hundreds to thousands of times larger with connections far more “intuitive” than the vgg19 model.

Join our weekly newsletter to receive:

  1. Latest articles & interviews
  2. AI events: updates, free passes and discount codes
  3. Opportunities to join AI Time Journal initiatives

It would take a computationally unreasonable amount of time. Especially since the developers won’t know for sure if the model will perform accurately after a portion of the training time has passed. One small mistake could affect the training process of the entire network. Another problem is that for a model to be able to perform all the tasks that humans can would require a completely different architecture. Current architectures are specialized and can only perform one task at a time after training.

In other words, current technology hasn’t caught up to this idea yet.

But you never know. It’s not like anyone knew that enormous metal planes would be flying through the sky today 120 years ago. Especially since Deep Learning is advancing at an extremely rapid rate. But for now, the limitations of our technology prevent networks from becoming as complicated as the human brain.

Who knows. Maybe a revolutionary idea will take our society by storm and allow neural networks to interact with us like real humans. Maybe a sentient model will be created in the next couple years that will someday have the answer to every single problem in our society. Maybe… judgement day isn’t that far away.

Contributor

I am a self-driven student developer and entrepreneur. I love working with other people and learning about the latest trends in the technological world we live in today. I hope to share my knowledge of software development and current events in areas such as Artificial Intelligence and Deep Learning with the public. Although I am a sophomore at my high school, I hope to develop a couple of ideas that I had into real companies that can revolutionize the world. In my articles, I will be talking about current trends in Artificial Intelligence, mobile app development, UI/UX design, getting involved in the community, and entrepreneurship.

My portfolio: https://cyb3rblaze.github.io/

Opinions expressed by contributors are their own.

About Anshul Kashyap

I am a self-driven student developer and entrepreneur. I love working with other people and learning about the latest trends in the technological world we live in today. I hope to share my knowledge of software development and current events in areas such as Artificial Intelligence and Deep Learning with the public. Although I am a sophomore at my high school, I hope to develop a couple of ideas that I had into real companies that can revolutionize the world. In my articles, I will be talking about current trends in Artificial Intelligence, mobile app development, UI/UX design, getting involved in the community, and entrepreneurship.

My portfolio: https://cyb3rblaze.github.io/

View all posts by Anshul Kashyap →