AI
After reading the latest reviews and news about the developments of artificial intelligent nowadays on Evernote, I am curious about how is it created and how is it works. So I researched this topic and found some basic explanations to help us understand the general principle of AI.
After reading the latest reviews and news about the developments of artificial intelligent nowadays on Evernote, I am curious about how is it created and how is it works. So I researched this topic and found some basic explanations to help us understand the general principle of AI.
There are several methods to build an AI.
They are Deep Learning, Machine Learning and Generative Adversarial Nets (GAN).
The GAN is one of the most advanced and popular approaches for complicated
tasks. Most of the researchers and practitioners in this field are making
efforts to learn and improve this system. A GAN is composed of two major
parts—generator and classifier. The generative model is converting random data
to true data and the discriminative model is classifying fake data from true
data. They are against and competing with each other. At the same time, they
are growing together.
For example, if we want to create an AI
painter that is able to draw an apple, we will need to start with training two
models. The discriminative model will get some initial inputs such as pictures
of apples. This will require another developing techniques called Pattern
Recognition. It is because there is no two apples are exactly the same and it
is impossible for us to input every single apple in the world to our
discriminative model. So pattern recognition can help the computer to find the
pattern of how an apple looks like. On the other side, our generative model is
getting little inputs about apples. For instance, we only tell it that an apple
is about 1-3 cubic centimeters and the color is varying from green to red on
the chromatogram. And then we can start running it. At the beginning, the
generative model will create some ridiculous stuff such as a pure green cubic
or a blood red sphere. It will send its work to the discriminative model. The
discriminative model could easily claim this data as fake and start up another
create- classify loop. After running this loop for billions of times until the
discriminative model is deceived by generative model, then we can say that we
successfully made an artificial intelligent painter that is better than human
artists in some degree.
Database & Future
Database & Future
According to above discussion, I realized that other
topics for group E for this assignment are all having some connections. For
example, database system is the key components for discriminative model since
it is collecting data as the initial inputs for this model. Database system is
not only collecting but also organizing and processing raw data that makes the
pattern recognition process a lot easier. Network and sociology are providing
support to this system from different aspects as well.
[1]Generative Visual Manipulation on the Natural Image
Manifold.Jun-Yan Zhu.ECCV 2016.
[2]Generative Adversarial Networks.Goodfellow.
[3]Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks.Alec Radford. [4]Deep multi-scale video prediction beyond mean square error.Michael Mathieu.
[3]Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks.Alec Radford. [4]Deep multi-scale video prediction beyond mean square error.Michael Mathieu.
Comment
Thomas Sisson
I am shocked when I read your MX3D article. I didn’t
realize that our 3D printing techniques are so mature and advanced that it can
be used in construction filed in real life. And I agree with you that this can
push the boundaries of conventional concrete to be used in projects that are more
architecturally complex in shape. Your work in ‘Future’ session inspired me as well. My post is
about AI but I only focused on this technique it self. You mentioned some
connections between AI and our field that makes me start to think this way.
Dee Dee Strohl
Your
article about HVAC & Sensor is interesting. It is a convenient way to
manage your house/ office and can effectively increase equipment’s efficiency. I am thinking in the future, if most of buildings
are having sensor system installed and AI technique is mature enough. We don’t even need an app to manage this system by ourselves.
HVAC equipment with AI will finish all of the work for us and a great amount of
energy will be saved. Both of our energy and environment problems will be mitigated.
Milligan
I like the video about software you posted. Before I
watch that and read your discussion, I thought we only use software in the
design phase behind a computer. But with the development of mobile devices, we
are able to use apps during construction phase on filed. This can make the
project manager or construction engineer perform their work in a more safety
and effective way.
Shuyuan,
ReplyDeleteI had never heard of the GAN strategy specifically before, but your coverage of it is very interesting. Even though you mentioned how it could be used for something as "simple" as an apple, the uses for what could be possible with this strategy could easily fall within the construction field. Using this kind of technology to quickly design housing to follow a certain style, and include all building systems may allow us to construct cheap and effective housing for underdeveloped areas, and even has potential for disaster relief and prevention.
I liked your post because it gave me a better sense of how AI is developed and how it works. Once the AI is initially built, I assume the process (both the discriminative model and generative model) cycles quickly and as time goes on, because it learns from each iteration, it begins to work even faster. Based on your post, I believe the AI discussed in the article I read used the GAN strategy to build the system. I want to learn more about each of the building methods and understand what applications they work best for.
ReplyDelete