Machine Learning (Summer 2017)

From New Media Business Blog

Jump to: navigation, search

Contents

What is Machine Learning

One of the most important parts of being human is our ability to learn and improve at tasks through experiences. We know almost nothing and can do nothing for ourselves when we are born. What we do as we grow up is learn and practice in order to become more capable. Similarly, computers can do the same thing. Machine learning is a subset of Artificial Intelligence. Machine learning brings together statistics and computer science to enable machines to teach themselves and set their own rules. There are other tools for reaching AI, such as evolutionary algorithms and Bayesian statistics.

Types of Machine Learning Algorithms

Supervised Machine Learning

Supervised learning algorithms are trained using labeled examples, a “trainer/teacher” will present the computer with example inputs (X) and their desired outputs(Y). [1]

Supervised learning problems can be further grouped into regression, classification, prediction, and gradient boosting problems.

Supervised learning is commonly used in applications where historical data is likely to predict future events.

Unsupervised Machine Learning

Unsupervised learning is used against data that has no historical labels. In unsupervised machine learning, the computer is given inputs and discovers patterns by itself. These methods are called 'unsupervised' because unlike supervised learning there is no desired output or “right answer”, and there is also no teacher or trainer. [2]

Unsupervised learning problems can be further grouped into clustering and association problems.

These algorithms are also used to segment text topics, recommend items and identify data outliers.

Reinforcement learning

Reinforcement learning is often used for robotics, gaming, and navigation. With reinforcement learning, the computer system receives input continuously and is constantly improving. [3]

This type of learning has three primary components: the agent (the learner or decision maker), the environment (everything the agent interacts with) and actions (what the agent can do).

The goal in reinforcement learning is to learn the best policy by choosing the actions that maximize the expected output over a given amount of time.

The Importance of Machine Learning

The increased interest in machine learning due to the advanced technology. For instance, the varieties of available data is easier to acquire, and computational processing is becoming more powerful. [4]

Through machine learning, we can analyze bigger, more complex data and deliver faster, more accurate results. [5] By teaching a machine to complete tasks and classify like a human, it can guide towards better decisions and smarter actions in real time without human intervention. Machine learning allows people to get things done in a more efficient and effective way. Machine learning has the potential to dramatically impact the future of the business world by helping organizations save time and money.

Evolution of Machine Larning

1950

Alan Turing creates the “Turing Test” which is a test of a machine's ability to exhibit intelligent behavior equivalent to a human[6].

1952

Arthur Samuel writes the first computer learning program for checkers which improved at the game the more it played [7].

1967

The “nearest neighbor” algorithm is written allowing computers to use basic pattern recognition. It measures the distances between points to classify the data [8].

1981

Gerald Dejong introduces the concept of Explanation Based Learning which gives an explanation for a problem that it has solved and can be used to solve future problems[9].

1990

Work on machine learning shifts from a knowledge-driven approach to a data-driven approach. This utilizes previous data to learn and solve future issues[10].

1997

The world champion of chess is beat by IBM’s Deep blue[11].

2006

Geoffrey Hinton establishes Deep Learning, a structured method which uses artificial neural networks that emulate a human brain [12].

2011

Google brain is created. A deep learning artificial intelligence research project at Google, it focuses on constructing models with high degrees of flexibility that are capable of learning their own features [13].

2014

Facebook implements deep face, which is a facial recognition system created and implemented on facebook. It is said to be 97% accurate [14].

2015

Amazon launches a machine learning platform and Microsoft begins to offer a machine learning toolkit.

2016

Google’s DeepMind program beats the world champion, Lee Sedol, at a game of GO [15].

Aplications of Machine Learning

With machine learnings emergence in recent years, its incorporation in applications has become the norm. Thus, applications that integrate machine learning have become numerous and difficult to list all the applications that take advantage of its benefits. Below are the five most interesting applications of machine learning.

Video Games

Divide and conquer: How Microsoft researchers used AI to master Ms. Pac-Man

By breaking down a video game's gameplay into smaller problems, machine learning algorithms can be implemented to test out all the options a video game character can select. Associating positive decisions with greater rewards and negative decisions with negative rewards, the algorithm can learn to distinguish between a good and bad decision. After selecting a decision, the result is stored and used for similar future problems. Having a scoring system for the algorithm to base its decision making against allows the machine to learn and attempt to improve its score. By allowing the machine learning algorithm to continuously gather better information, its decision-making abilities will continue to improve.

Recommendation System

When selecting a movie to watch, a user encounters a vast selection of choices and in some cases, they may have a hard time selecting an option. In the case of Netflix, machine learning is used to predict and suggest movie options that cater directly to the user [1] . Based on a user's previously watched movies, common actors, ratings given and other factors, machine learning algorithms can better recommend options to the user. With every selection a user makes, the recommendations become more refined. By refining the vast selection of movies, machine learning eases the users mind and provides them with better-suited options that are catered to their preferences. While these recommendations are not limited to Netflix, other firms, such as Spotify and Amazon, have also incorporated machine learning into their platforms in hopes of providing a personalized experience for their customers.

Computer Vision/Perception

Just like a record in a table where information is stored, information about images can be stored as well. Information about an image is created from the properties of an image. By measuring certain objects in an image, such as the length of an individual's forehead, data can be stored and used to describe that image. Machine learning takes advantage of this information to identify the image. In the case of Facebook’s facial recognition, machine learning is able to identify individuals within images. It starts by looking at the picture and identifying faces. Next, it measures facial features and compares them against other individuals to identify the corresponding individual in the image[2].

Financial Trading

With the rise of robo-advisors, financial trading has become more automated than we can imagine. Since the financial market is composed of past and present values a stock or bond might have, information is ample. By using machine learning algorithms to predict a stock's future value based on current and historical data or news relating to a stock, machines have taken a greater role within the financial industry. The key feature machines bring to the financial market is the efficiency in which decisions and trades can be produced. This is significant in a market that is highly competitive as it provides users with an advantage over other individuals [3].

Self Driving Cars

Autonomous Drifting using Machine Learning

Similar to image recognition, self-driving cars utilize machine learning to understand and react to their surroundings. As data continuously streamed into the vehicle’s machine learning algorithm, a vehicle can visualize signs and other vehicles surrounding it all within a second of the data being collected. In addition, having a network of vehicles logging trips around the world provides ample information for better decision making. The continuous stream of information being fed into a vehicle's algorithm will allow the vehicle to continuously improve. The key advantage of self-driving cars, provided by machine learning, is the safety it provides its occupants by removing human error from driving.

The Future of Machine Learning

Health Care

Machine learning has endless possibilities and applications in many industries. One of the industries that will substantially benefit is health care, as machines are quickly learning to identify diseases and provide diagnoses. Google’s DeepMind Health is showing potential for improved speed of care and identify new diagnoses [1]. Personalized treatment based on genetics and previous data is now possible which will better customize health care for an individual. IBM Watson Oncology also uses patient information and history to better identify treatment options [2].Machine learning can also help in drug development and identification. It can provide aid in the initial screening process all the way to predicting success rates[3].

Cancer treatment can also be improved with the use of machine learning. An assistant professor at Harvard Medical School stated: “In 20 years, radiologists won’t exist in anywhere near their current form. They might look more like cyborgs: supervising algorithms reading thousands of studies per minute”[4]. In addition, epidemic outbreaks can be predicted through data collected from satellites, historical information on the web, real-time social media updates, and other sources[5].

Future Job Market

The jobs we'll lose to machines -- and the ones we won't

Machine learning has the potential to revolutionize the job market and what roles human’s play in it. Machines have the potential to replace common jobs and are even better at repetitive tasks, however, they lack the ability to learn empathy and emotions. These traits are not programmable or learnable and is the one of the main factors restricting machines and machine learning. Another trait that humans possess that cannot be programmed is creativity. As machines can only predict an outcome based on historical data, this can provide issues when they are put in novel situations that require creative responses.

Ethical Issues With Machine Learning

Ethical issues and concerns in machine learning may arise with the collection of big data (which is crucial to machine learning), the application of machine learning, the level of autonomy, and through human biases that a machine may exhibit after training on certain data sets.

Privacy

Machine learning requires a lot of data. Companies that collect this data are constantly gathering information about you, with or without your knowledge. It is important to remember that any information you post online is never truly “private” and can always be accessible to others.

Human Use Of Machine Learning

As much as machines can do, they are still guided by humans and how humans decide to use them will determine whether or not they are ethical. Machine learning has provided many benefits such as personalized marketing and search results, replacing tedious human tasks such as trading stocks, and even finding ways to be more efficient than humans, such as self-driving cars and medical diagnosis. On the flip side, machine learning can be used to learn about someone, which can later lead to fraudulent activities or potential scams. Since security is always a concern, it is also important to remember that any automated machine can be hacked and used for malicious purposes. As machine learning is growing rapidly, humans must be careful that they do not rely too heavily on the practice, as some things should not be left to machines to handle.

Degree Of Autonomy

When using machine learning, the level of autonomy in machines is always an important factor to consider. As it may be appropriate to fully automate one process, it might not be applicable to another. The classic example of this dilemma is the level of autonomy in self-driving cars. As the trend for self-driving cars is slowly going towards full automation, there are still many ethical concerns that surround the matter. Machines are programmed to respond in a certain way, and that response may not always be the right one. For example, if an automated car is put into a situation where it must choose between hitting a pedestrian to save the driver or hitting a wall to save the pedestrian, how does it choose? Once it has made that decision, who is responsible for the death? Additionally, as a consumer, would you want to get into a car that was designed to potentially harm you? As fully automated self-driving cars are still in the early stages of testing, it will be interesting to see how this issue is resolved. [1]

Reflection Of Human Error

Machine learning is still in the early stages of development and companies are constantly experimenting with new ideas, which may not always go as planned.

Companies such as Flickr and Google have yet to perfect facial recognition software [2]. Flickr’s facial recognition software was labeling black and white faces as animals and apes and tagging photos of native American dancers were also being tagged with the word “costume”. Similarly, Google Photos was also tagging black faces as gorillas and Caucasian faces as dogs or seals.[3]

Nikon and HP have also struggled with face-detection software. Nikon’s face-detection in cameras displayed a “did someone blink?” message, when Asian faces were photographed and HP's MediaSmart computer that was designed to follow faces of all users, could not recognize an African-American man. [4]

Twitterbot Tay

Probably the most famous example of machine learning gone wrong is Microsoft’s experiment with Twitterbot Tay. In March 2016, Microsoft released Twitterbot Tay, an artificial intelligence chatterbot designed to mimic the language patterns of a 19-year-old American girl and learn from interacting with the human users of Twitter. Although it was initially launched with the intentions to learn how to improve the customer service of their voice recognition software, the project was a disaster. Users turned “Tay” to be misogynist and racist in less than 24 hours and Microsoft was forced to pull the failed project off the web. [5]

Conclusion

Although these outcomes may not have been favorable for these companies at the time, in the long run, it may not be such a loss. These “disasters” provide a solid learning experience for these companies to learn from their mistakes which will help them be more successful in the future.


References

  1. http://www.nbcnews.com/tech/innovation/driverless-cars-moral-dilemma-who-lives-who-dies-n708276
  2. https://petapixel.com/2015/05/20/flickr-fixing-racist-auto-tagging-feature-after-black-man-mislabeled-ape/
  3. https://www.forbes.com/sites/mzhang/2015/07/01/google-photos-tags-two-african-americans-as-gorillas-through-facial-recognition-software/#4f10c8c6713d
  4. http://content.time.com/time/business/article/0,8599,1954643,00.html
  5. https://www.theverge.com/2016/3/24/11297050/tay-microsoft-chatbot-racist
Personal tools