Written by Seoyoon Chang
If you have ever used Google Maps, Email, Amazon, and various sites using search technology, you would have indirectly interacted with AI before. Just by reading the examples listed before, anyone can tell that AI is playing an important role in our everyday lives. From email to cleaning robots, AI is now used in ranging forms of technological devices and applications. But what’s making AI so successful? The simplified answer is the accuracy and “confidence” AI displays in recognizing faces, speech, and even images. If you’re wondering right now how AI is recognizing all of these different features, it’s the ways of Machine Learning. Machine Learning is the sub-field of AI, using Big Data to “learn” and categorize information. Instead of humans programming computers what to do, Machine Learning algorithms use their own information to teach itself. The three Machine Learning algorithms include Unsupervised Learning, Supervised Learning, and Reinforcement Learning. All of them are in major use today and are working towards achieving a global accomplishment where the capability and accuracy surpass human beings. Supervised Learning is an algorithm where the system takes in, for example, images, and given different categories, sorts the images into where the system thinks is relevant to the category. This is useful in determining and classifying faces (linking to facial recognition) and providing large amounts of data to which associate with searches on the internet. An example which anyone could look at is Google’s Open Images Dataset, which consists of images amounting to approximately nine million images, and labels objects, items, and creatures into specific categories. Unsupervised Learning is a process where the system tries to find patterns in data which has not been classified. Unsupervised Learning discovers the structure in the data and learns to group it into a certain category. An example which is widely known is the K-means Clustering, in which points are scattered and the program assigns each dot into a cluster based on the distance from a cluster centroid, a point which determines which dots will be assigned into a certain group. Reinforcement Learning is a system where the program identifies errors and learns from them, usually from the action it performs. The program will adjust to behaviors which are exemplary. An example is AlphaGo, a computer program which is based on the board game Go and defeated Lee Sedol, an 18-time World Champion. With this information in mind, in order to uncover how AI accomplished these techniques, Deep Learning holds the answer. Deep Learning is the advanced form of Neural Networks, models which take in data and interpret it into a certain result, just like a human brain. Deep Learning has only been established recently but already excels at recognizing images and speech. Even with the mass improvement, there is still more to be done. Self-driving cars can be improved with a more accurate image recognition, and possibly by improving deep learning, we can improve the translation of speech. Although deep learning softwares are working on becoming an indistinguishable tool to the human brain, we still have a lot more paths to cover. SOURCES: Vishal Maini. “Machine Learning for Humans, Part 4: Neural Networks & Deep Learning.” Medium. 19 Aug. 2017. Web. 1 Sept. 2018. <https://medium.com/machine-learning-for-humans/neural-networks-deep-learning-cdad8aeae49b> Hof, Robert D.. “Is Artificial Intelligence Finally Coming into Its Own?.” MIT Technology Review. 17 Jul. 2018. Web. 1 Sept. 2018. <https://www.technologyreview.com/s/513696/deep-learning/> Jason Brownlee. “What is Deep Learning?.” Machine Learning Mastery. 16 Aug. 2016. Web. 1 Sept. 2018. <https://machinelearningmastery.com/what-is-deep-learning/> DeepMind. “AlphaGo Zero: Learning from scratch | DeepMind.” DeepMind. 18 Oct. 2017. Web. 1 Sept. 2018. <https://deepmind.com/blog/alphago-zero-learning-scratch/>
0 Comments
Leave a Reply. |
Writers & Editors- Seoyoon Chang Archives
January 2019
Categories
All
|