Questions Technology

What are the limits of deep learning?

Airtract

Andrea Vurlow

Living life on my own terms

We are currently in the third wave of deep learning. The previous two waves took place around the 1950s and 1980s but sadly, it couldn’t leave an impact. This is mainly because these neural networks failed to achieve their expected performance back then. The current wave of deep learning is different from what it was in the past. The core ideas of deep learning were already identified in the first two waves. But it is in the third wave that its true power has been unleashed. This became possible only with the development of powerful computers and large datasets.

Limitations of deep learning

It is true that deep learning has come a long way but it still faced by criticism due to the limitations it poses. The main limitations of deep learning can be summed into the following points.

  1. Deep learning in the current scene is known to lack common sense. By common sense, we mean acting intelligently in different situations. It should be improved and given the power to draw conclusions based on its limited experience.

  2. Another provable problem with deep learning is that it cannot understand the data context. It may be very good at mapping inputs and outputs, but it fails to understand the data context that they are handling. For example, deep learning algorithms can prove great players in a particular game. They can even beat the best of human players as well. However, they are not equipped to understand the various elements of the game like a human does.

  3. The next big challenge that is faced by deep learning is that it requires a huge amount of processing power. The performance of deep learning depends on high-performance hardware which consists of the multi-core graphics processing unit. These high-performance processing units demand a lot of power and thus, it is very expensive. This is one of the main reasons why companies are facing problems in adopting deep learning.

From the above discussion, it can be concluded that deep learning, although it has come a long way, it still requires a lot of improvements to be made. There is no doubt that deep learning is the basis for building an AI-powered society, however, to achieve that it has first to overcome its limitations.

Technology Courses

Airtract FLASH SALE

Complete Outlier Detection Algorithms A-Z: In Data...

Saurav Singla

0 (0) New Course

Welcome to the course "Complete Outlier Detection Algorithms A-Z: In Data Science". This is the most comprehensive, yet straight-forward, course for the outlier detection! Are you Data Scientist...

1 hrs 40.12 mins 0 Students Enrolled 18 Lectures

$19

71.64 % off $67

$19

Buy Now
Airtract FLASH SALE

Spark AR Studio for Beginners: Create Your Own Fil...

Artem Dovgenko

0 (0) New Course

Start creating your filters professionally with Spark AR Studio! If you are looking for a filter creating an application that will allow you to create filters however you want them, Spark AR Studi...

1 hrs 10.58 mins 1 Students Enrolled 28 Lectures

$19

61.22 % off $49

$19

Buy Now
View All
Item added successfully. Go to cart for checkout.
Accept Reject