Getting to Know Machine Learning Technology with Jim Geovedi

Machine Learning is nothing new in the computer science landscape. This part of the Artificial Intelligence concept is increasingly popular today along with the increase in awareness many parties for digital data management and system automation to replace the manual role by humans. To review this technology, especially its relation to development approach at startup, DailySocial chats with Jim Geovedi as one of the best practitioners in this technology.

Starting the discussion, Jim, who is also involved in technology development in online media services Beritagar.id, said that the implementation and benefits of Machine Learning have actually been felt by us, although perhaps not many of us are aware of it. In simple language, Jim defines machine learning techniques as computer algorithms to study data, recognize patterns, and create models based on historical data. The model is used to classify or predict new data that allows us to make or support decision making.

Machine Learning concept analogy

For example, DailySocial wants to hold an event that will invite many startups to present their products and their potential in the Indonesian market share. The committee managed to collect 10 startups namely Go-Jek, FitInline, GrabBike, OkeTiket, Kulina, Abraresto, Traveloka, Tees, Tickets, and foodpanda.

To keep things organized, the event team decided to group presentation sessions by startup category. Let's say that the DailySocial editorial team is currently not in the office for all coverage, so the event team must independently identify the startup by category. Since the DailySocial event team has held similar events several times and has often met with various startups, the event team has several things to evaluate to determine startups based on their categories, namely:

1. The name of the startup will usually represent the category of product being developed.
2. The startup logo will usually represent the product category being developed.

From these two things, the team can still define it more clearly, namely:

Startup Name

• Usually the name of a startup with a food product category will contain words related to food or restaurants.
• Usually the name of a startup with a travel product category will contain words related to various things related to travel needs.
• Usually the name of a startup with a transportation product category will contain words related to transportation means.
• Usually the name of the product startup fashion will contain words related to fashion and clothes.

Startup Logo

• Usually startups in the food product category will use a logo depicting the attributes of food equipment.
• Usually startups with the travel category will use a logo that represents the trip.
• Usually startups in the transportation category will have a logo related to road equipment.
• Usually startups by product category fashion will have a logo associated with fashion and clothes.

of two detail After that, the event team was able to group the 10 startups based on product categories:

By startup name

• Food product categories: Kulina (close to the word culinary), Abraresto (no word rest), foodpanda (there is a word food)
• Travel product category: Traveloka (there is the word travel), Tiket and OkeTiket (there is a word ticket)
• Transportation product categories: Go-Jek (close to motorcycle taxi) and GrabBike (there is a word bike)
• Product category fashion:fitInline(fit is a size that is often represented for clothing products) and Tees

Based on startup logo

Roughly defined as in startup name points.

In Machine Learning technical terms, the startup name and logo are part of what is called a feature (features) and some detail of each feature is called the frequency distribution (frequency distribution). The machine learning process is something like that. Sometimes in certain features an object does not have the right specifications, for example from the example above the Tees logo which is only in writing, but can be clearly identified in the previous feature, namely the startup name. However, this can be circumvented by providing more and more detailed features and distribution frequencies. That's roughly a Machine Learning algorithm is structured so that a computer machine learns.

Development of Machine Learning-based solutions for tech-startups

Before talking more detail Regarding the scope that a tech-startup can work on with the Machine Learning concept, Jim Geovedi explained the challenges of implementing this technology, especially in Indonesia. Jim says:

"There are quite a few challenges to the implementation of Machine Learning technology in Indonesia, including the relatively low wages for human labor which are difficult to argue for in cost efficiency, understanding of the benefits of technology which is still minimal, and the fear of human labor being replaced."

However, when it comes to development capabilities and technicalities, the existing Machine Learning is so sophisticated, and it is possible to create systems that self-study complex and large-scale data sets with minimum human intervention. Of course the implementation is considered successful if the results of the automation process carried out are able to approach the quality of human work at an affordable price. Benefits like this are believed to be able to be a contradiction of the issues mentioned above.

"I am quite optimistic that the implementation of Machine Learning technology in Indonesia will flourish in the near future and will be followed by many companies specializing in developing this technology," said Jim.

A simple law that is important to be a reference is that even though a technological concept looks complex, it does not always have to be applied to problems that seem large and complex.

Jim gave an example of a simple implementation of Machine Learning, namely spam/junk identification. The technique used is to study previous data that has been labeled (spam or not) by extracting features which will then be used as input parameters from the algorithm used to classify. For automation, a model is made as a result of learning and also the algorithm used. This model will be used to classify or predict new data.

Other examples that are also commonly found in the online public sector in Indonesia include content recommendations. Starting from other article recommendations related to articles being read on an online media site, other items related to items being viewed on a commercial site, to other videos related to videos being watched on online video viewing sites.

Machine Learning has also been used for industry-specific purposes in Indonesia that are not directly related to the public, for example identifying attack patterns (from hacker, rootkit, viruses, etc.) aimed at a network and automatically block, perform automatic bidding (autobid) advertising, identifying user personalities based on their online interactions, predicting events or characters that are expected to be newsworthy, to creating text content automatically.

According to Jim, the development area for tech-startups for Machine Learning technology is for products like the ones exemplified above. This is because fees can be paid as needed (On-demand) thanks to the convenience of technology cloud computing. Even for technology Back-end Machine Learning some cloud computing provider already have it ready for use.

Market demand for machine learning platform technology solutions

Jim sees that currently in Indonesia there is a shortage of experts related to the development of Machine Learning technology. This may have something to do with the focus of the industry which puts more emphasis on marketing aspects than investment in technology, so it prefers ready-made implementations (through PaaS and SaaS facilities) that have been developed by other parties abroad. As people become more aware of the efficiency that can be accumulated with more custom technology, it is believed that in the future this kind of technology will be taken into account and prioritized.

So this is the perfect opportunity to hurry up and get ready to learn. Jim recommends several study goals that can be followed, including:

1. "Introduction to Artificial Intelligence" by Sebastian Thrun and Peter Norvig. Sebastian Thrun is best known for making self-driving car at Google and Peter Norvig is one of the pioneers of artificial intelligence technology who currently serves as Director of Research at Google.
2. "Machine Learning" by Andrew Ng, one of the Stanford professors who later became Chief Scientist at Baidu Research.
3. "Neural Networks for Machine Learning" by Geoffrey Hinton, known for his research on neural network and currently works as a Distinguished Researcher for Google and a Distinguished Emeritus Professor at the University of Toronto.

In addition to the three sources above, there are still many learning resources that can be used which can be found out by using search engine. System automation will be the newest way of a service going forward, including in Indonesia, which has already begun to be used.