Scoop: QTFT webinar “an introduction to machine learning”
by Dr. Naruemon Pratanwanich
Many of us have probably heard a buzzword “Big data”, which has circulated in the news, on social media, or on tech-related websites. This is because the word Big data has been involved and has impacted in so many different subfields, from low-end users who go to supermarkets for everyday purchases, to international security and military systems. This jargon serves the purpose of acquiring immense data from any relevant sources — the more the better — for us to process, analyze and make predictions. But, how are those big data processed? How do we gain any interesting information? These are the questions that the emerging field named “Machine Learning” do have an answer.
Let us consider simple scenarios. If there were not so many data, for example, a monthly report of Thai food ordered in a restaurant, of course, a few people could simply spend a few days, or maybe a few months, to process data and come up with strategies for a new food menu. But when the data become larger and larger, both in quantity and complexity — for example, medical data related to heart diseases in Thailand — that is when we need to teach the machines to work for us. Especially now that we have reached the peak time for both software and hardware computing powers, Big data and Machine Learning are therefore important research fields with so many possible applications for our society.
And how does Quantum technology fit in this picture of Big data era? The simplest answer is that computer scientists are expecting a break-down of Moore’s Law, where the increase in power of a classical machine will reach a certain limit and will stop progressing.
So, if we still need to process such huge data, we need to resort to the power of Quantum computing, using the advantage of quantum weird properties to enhance Machine Learning. This is the coming of the novel field of research called Quantum Machine Learning (QML) or Quantum-enhanced Machine Learning.
QTFT sees the opportunity of this new field, combined with the fact that we have members in our community already working on both Quantum algorithm and Classical Machine Learning research. We then initiated the idea to organize a webinar on “An introduction to Machine Learning” and invited Dr. Naruemon Pratanwanich (Ploy), a computer scientist at Genome Institute of Singapore (also an associated lecturer at Chulalongkorn University, Thailand) to guide us on this topic. The webinar was held just before the Christmas time last year (December 19th, 2018), and we have received a lot of attention from the public audience, with 41 registered attendees and 23 online-webinar participants.
So, what is Machine Learning? Dr. Pratanwanich introduced the topic using the example of AlphaGo, a computer program developed by Google Deepmind, while competing with Lee Sedol, a South Korean professional Go player, in March 2016. AlphaGo used machine learning (and tree search techniques [REF WIKI]) to process huge data collected from games played by both humans and computers. AlphaGo did win Lee Sedol 4 out of 5 games, with a report said it had used 1,920 CPUs and 280 GPUs of Google’s cloud computing from servers in the United States.
The other examples, Dr. Pratanwanich mentioned, which are closer to our everyday life, are the facial recognition on Facebook and the movie-recommendation system on Netflix. Tons of data from their users are processed to predict behaviors of individual users, which can then benefit the companies in terms of making effective marketing or business plans.
Dr. Pratanwanich then touched on the three main types of machine learning: Supervised Learning, Unsupervised Learning, and Reinforcement Learning. Each has different strategies in acquiring data, depending on the nature of problems and goals to accomplish. Supervised Learning deals with problems that have specific input data and targets, for example, Facebook’s facial recognition aims to collect personal data and photos in order to recognize individual persons in a newly uploaded picture. On the other hand, Unsupervised Learning has no specific targets, and a goal is just to learn something (can be anything) from the available data. For the Reinforcement Learning, it is a process that runs based on data from trial-and-error in order to achieve rewarding goals, for example, a machine that tries to win a simple strategic game.
Her talk not only gave a broad view of the machine learning but also added a right amount of mathematical and computational ideas behind the power of this learning technique. She also presented examples from her own research work, in collaboration with Chulalongkorn University, using machine learning to analyze user database from Home Buyers Guide website in Thailand, to help personalize information of home search for individual buyers.
Even though the talk went over the set time (50-minute), we had received immense attention and all the 23 participants stayed with us until the end. We had attendees asking questions during the talk, and we also had Dr. Thiparat Chotibut, another expert on Artificial Intelligence and Neural Networks, helping to answer many other questions from the audience during the talk, via messaging on the zoom platform. This talk was fully recorded and we would like to invite the reader to watch the published video for more detail.
So what is next? How does quantum computing enhance the computing power for machine learning? The question will be answered next in our future talks, focusing on the quantum algorithm and how it can expedite the computation power. Quantum machine learning is still a field at the research stage, meaning that we (Thailand) will still have a large chance to catch up on this technology.
— — — — — — — — — — — — —
An Introduction to Machine Learning
Dr. Naruemon Pratanwanich (Ph. D. University of Cambridge)
“When you’re fundraising, it’s AI. When you’re hiring, it’s ML. When you’re implementing, it’s logistic regression.“ — everyone on Twitter ever. What does Machine Learning (ML) actually mean? The layman explanation is that ML is a field that equips computers with the capability to “learn” from data without being programmed explicitly, though, to be clear, teaching computers to “learn” still requires a bit of programming. Generally speaking, ML aims to create a machine that consumes data and evolve into an intelligent machine that can automatically accomplish complicated tasks. To perform such feats, ML has transformed into a truly multidisciplinary field, borrowing ideas and techniques from Mathematics, Statistics, Computer Science, Game Theory, Physics, Computational Neuroscience, and, inevitably, specific domain knowledge of interest.
In this talk, we will first motivate you with the applications of ML and why you should care about ML. Standard formulations of ML problems will be discussed. We will then review many categories and terminologies of ML, walking through digital age buzzwords such as Deep Learning, Bayesian methods, Likelihood, Regression, and etc. Some of these concepts will be discussed in details, including Deep Learning, the successful rebranding of the upgraded artificial neural network-based learning algorithms that drive modern AI revolution. Lastly, we will touch upon the limitations of classical ML algorithms, and end the talk with open discussions on potential advantages of quantum ML counterparts.
— — — — — — — — — — — — -
QTFT was Founded by a group of Thai scientists and entrepreneurs who believe that Thai talents are no less capable than any other people in the world, QTFT is a non-profit organization for fostering collaboration between scientists, engineers, and investors to promote sustainable development of innovation and quantum technologies in Thailand.