Page 34 - 10 Most Innovative AR VR Startups In 2019 up
P. 34
7Challenges for Machine
Learning Projects
Although scientists, engineers, customers on the possible questions we want to ask them. A
and business mavens agree we applications of their innovative business working on a practical
might have finally entered the technology. Machine learning machine learning application needs
golden age of artificial intelligence engineers face the opposite. to invest time, resources, and take
when planning a machine learning Entrepreneurs, designers, and substantial risks.
project you have to be ready to managers overestimate the present
face much more obstacles than you capabilities of machine learning. A typical artificial neural network
think. They expect the algorithms to has millions of parameters; some
learn quickly and deliver precise can have hundreds of millions. A
Deep learning algorithms like predictions to complex queries. training set usually consists of tens
AlphaGo are breaking one They expect wizardry. of thousands of records. While a
frontier after another, proving that network is capable of remembering
machines can already be able to Because of the hype and media the training set and giving answers
play complex games “thinking buzz about the near coming of with 100 percent accuracy, it may
out” their moves. Automation has general superintelligence, people prove completely useless when
more applications than ever before: started to perceive AI as a magic given new data. The mechanism is
from email classification, music, wand that will quickly solve all called overfitting (or overtraining)
and video suggestions, through problems - be it automatic face and is just one of limits to current
image recognition, predictive recognition or assessing the deep learning algorithms.
maintenance in factories, to financial risk of a loan in less than
automatic disease detection, a second. It’s not that easy. Not at The black box problem
driverless cars, and independent all.
humanoid robots. The early stages of machine
In fact, commercial use of machine learning belonged to relatively
Understand the limits of learning, especially deep learning simple, shallow methods. For
contemporary machine methods, is relatively new. They example, a decision tree algorithm
learning technology require vast sets of properly acted strictly according to the
organized and prepared data to rules its supervisors taught
Many companies face the provide accurate answers to the it: “if something is oval and
challenge of educating green, there’s a probability P
it’s a cucumber.” These models
weren’t very good at identifying a
cucumber in a picture, but at least
everyone knew how they work.
Deep Learning algorithms are
different. They build a hierarchical
representation of data - layers that
allow them to create their own
understanding. After analyzing
large sets of data, neural networks
can learn how to recognize
34 December 2019
