With the topic of “Applied AI: Insider Story”, Solution Forum #62 had welcome 3 speakers who are pioneers in AI and Data Science. These tech experts had shared their real lessons, as well as practical experiences from deploying AI projects, like detecting fake Instagram accounts, AI application in water filtering, as well has assistant to decision-making.
The year-end Solution Forum hosted by FPT Software had attracted over 60 audience interested in AI. This even had welcomed 3 speaks who are well experienced in studying and researching AI, including: Expert Nguyen Xuan Phong – Head of AI FPT Software, Huynh Ngoc Qui – Data Scientist, and Hung Ngo, Founder of Center of Talent in AI.
Application of AI and Reinforcement learning in the water-filtering industry
As the first speaker, Mr. Nguyen Xuan Phong had shared about using Reinforcement Learning (a branch of machine learning, where people research how an object pick what actions to take to maximize long-term rewards), in combination with predictive and decisive AI in procedures of water-filtering factories.
This is a project that aims at assisting in making important decisions during water-filtering, researched and developed by Mr. Phong for Hitachi’s factories in Japan, in order to replace the reducing numbers of experts in this aging country. The application of AI in making decision on water levels and amounts of filtering chemicals, he had proven that AI can surpass human and experts with its ability to learn in new environments and experiences.
According to Mr. Phong, businesses need to run constantly, while experts in specific fields are hard to replace. In these cases, AI can replace or assist experienced workforce in carrying out their tasks, which much better efficient. AI, as such, is a solution to expanding the business scale in a quick manner.
Regarding benefits, AI can function better than human with a specific purpose, react and adapt easily to constant training environments, and optimize for other purposes.
However, many people still question how to train and deploy AI in a real environment, how to persuade traditional people to convert to using AI, how to delegate responsibilities regarding AI errors, as well as what to fix in cases of malfunction…
At the event, Mr. Nguyen Xuan Phong also introduced to the software world the organization FSoft AI Lab, newly established with the target of developing research on AI at FPT Software, as well as on its future orientations.
Detecting fake SNS accounts
The second speaker, Data Scientist Huynh Ngoc Qui, have given an interesting speech on data science via the problem of “Detecting fake accounts on Instagram”.
According to Ms. Qui, “influencer marketing” is quite a new field, or rather, a new approach to an old field, where “influencers” enables brands to meet marketing targets on SNS. “Influencers” are people who can interact and influence the buying power of their audience, as well as understand the brands they are promoting.
A consequence to this new marketing method is that many people are going all ways to raise followers on Instagram (an SNS from Facebook). By estimations, an account with above 15,000 followers will be considered an influencer. And by the beginning of 2019, there are already 1,4 million accounts that had met this threshold, although only 39% of those can be actually consider influencers, when you take away accounts fake followers.
In order to help brands avoid the trap of fake influencers, detections of fake accounts via parameters of number of followers, follows, post counts, interactions and comments are necessary. In some cases, the filtering of fake accounts also utilized unstructured data like biography, list of follows, avatars, and post contents.
According to Ms. Huynh Ngoc Qui, AI can help detect fake accounts by analyzing the changes in follows, set of audience (according to account creation data, IP chains, and day of follow an account), as well as structure of followers like age, gender, nationality… distribution.
“Detection of fake accounts is not a one-time thing. Fake influencers will only get smarter, using more sophisticated methods for more follows. Thus, this field is as challenging as it is interesting”, said the data expert.
AI assistance in decision-making
In the final part of the event, Mr. Hung Ngo – Founder of Center of Talent in AI, had presented an AI topic that helps human in decision-making.
According to him, AI application in daily task is sky-rocketing these days, as AI is regarded as company and virtual assistants so human can work better with more creativity. AI not only support humans in recognition of sounds, images, languages, data, and labor, but also assists in decision-making.
In order for AI to do this, he highlights 5 important elements to provide for the computer, including: input and output tasks, experience or data, functional space to find the best function that closely resembles the experts’ intelligence function, and searching algorithms to dig around in the functional space. He affirmed: “If we can extract intelligence functions and own a good business model, AI, along with human, can create billion-dollar start-ups.”
In integrating Machine Learning/Deep Learning to find a general estimation for evaluation, Mr. hung had emphasized this as quick intuition. In combining these intuitions with general logical thinking, we can make applications to tackle complex AI problems, like assisting human in decision-making, planning, logical thinking, reasoning, and so on. Mr. Hung had also introduced his virtual assistant and recommendation system project, developed at IDSIA Lab (Lugano, Switzerland), and MLR Stuttgart (Germany). The projects enables AI and robots to learn to support human using task intuitions and behavior reasoning, to balance between task optimization and experience personalization.
As the event ends, attenders had got the chance to enjoy active discussions where speakers together share their views regarding careers and personal experiences, helping them to understand the passions and challenges that lie behind the choice towards AI and Data Science.
Data – The heart of AI?
This is a question inquired at the speakers during panel discussing. According to Ms. Qui, data is indeed important, and sometimes even matter more than algorithms. From her personal view, her background is never computer science, yet it is life that leads her to data science. She had started with simple tasks like data filtering, and thus understand clearly the importance of data to an AI model. All projects start from the need to solve a problem, but when the first model does not work out, then it is necessary to update ideas to create eventual better models. Nowadays in AI, people focus on algorithms, as there is little data and clean data sources. Furthermore, it is easier to improve data from algorithms then improve the algorithms rather than look for external sources of data.
A live stream video of the event can be view here:Related posts: