What is the best method of efficiently training machine learning for teams?

The average Briton has an average attention span of 14 minutes, as the recent study says. People tend to lose interest faster when they find the subject boring or complicated. That’s why the quality of training comes not only from the information provided but also the form, that it is served.

“You had my curiosity, but now you have my attention”

Adults learn selectively. We pay attention and learn what we find potentially useful and beneficial, or what we consider to be interesting. If I lay out for data scientists an example of a new convnet architecture that is ten times better than ResNet on the ImageNet dataset – I might well gain their attention. At the same time, if you are a sales representative who trades office furniture on the EMEA market, there is a great chance that I have “lost you”. You will never focus on this sentence… unless neural networks happen to be your hobby (in which case you already know that data science can help you at work too).

The point here is that good training is not only about the quality of knowledge involved, though that is crucial. Effective training is also about how new knowledge is served up, relating new information to previous experience, know-how and upcoming personal and work-oriented goals. This approach works for any learning process, data science education included.

Data science training in house

At the Training & Development Hub, we have built the 4T method – Tailored Team Training Tracks – which represents a fourth approach to machine learning education and good practices in providing learning experience, incorporated into data science education.

The 4T method is built on the assumption that practical education in data science should be provided in-company to software developers, data analytics, and other specialists familiar with computer programming or statistics. The 4T method can also be used by data scientists seeking to boost their skills. In both cases, the learning experience is tailored to the technical and business goals, provided for a team via hands-on code-based training structured in training tracks.

A win-win situation for the employee and employer

This approach to education benefits both the future data scientist (the learner) and his or her manager and company. 4T is focused on providing knowledge, experience and good practices, considering the nature of adult learners and teamwork benefits. The method emphasizes that the training needs to meet the specific practical goals understood by the individual and shared by his or her entire team, and it needs to provide skills that are going to be used in practice right away.

When these things happen, the company naturally uses the new knowledge to tackle its challenges. Because acquiring the new knowledge is a part of the job, the individual is strongly motivated and reasonable goals to be reached with the team so they have to work together using data science techniques.

Your internal data science team

Employers have plenty of reasons to build an internal data science team, but should bear in mind a couple of issues. First, there is a dearth of data scientists on the market. Further, good data scientists should have broad knowledge in numerous fields, such as computer programming, statistics and math, and that knowledge should be accompanied by a problem-solving mindset or at least a few years of experience in defining and solving work-related problems.

Such a skill set is nothing to sneeze at, and such employees are few and far between. Given that reality, is it worth the investment to retrain current employees such as software engineers and developers to build an internal data science team, or does outsourcing remain the better option?. If the company wants to grow its teams’ skills and develop know-how that will stay in-house, building a data science team from scratch is indeed a step worth considering.

Tailored Team Training Tracks

The 4T method was created by the Training & Development Hub based on seven years of experience in building data science solutions for our customers, and three years’ experience providing professional data science training in Europe and the US.

The method has four components that build a unique framework for training in data science and potentially other technological fields: Tailored, Team Training, Tracks. Below you will find an overview on each of the components.

Tailored

Every training is designed to meet the specific participant’s educational needs and, most essentially, the technical and business goals of the group involved. To give you an example, let’s say that a company would like to start a new project recognizing skin defects from photos. In this case, two different approaches – each based on deep neural networks – can be used: classification and anomaly detection. The training program provides all participants with knowledge about these exact techniques and the skills required to use them properly, no matter which approach they choose to work with finally. The participants can then turn around and use them right away in their everyday work.

Team

The training should be provided for a team, or at least for a group of a people working together who will be able to use the knowledge and skills they have gained in a common project. That is extremely beneficial for employers: their employees’ experience will be “calibrated” for future projects, which usually involve whole teams, for which a similar level of knowledge, understanding of issues and possible solutions are required. Thus armed, companies can smoothly and quickly take on new data science projects.

From the other hand, it’s useful also for the participant – even if the learning motivation for adults is extremely internal variety, though group approach and the need of cooperation are also a strong stimulus.

Training

In data science, the learning experience should be provided by hands-on, code-based, intensive training using real-life examples. In each minute of a training, participants should know why they are doing what they are doing. This is why the training should be as practical and project-based as possible.

In data science, that means working on datasets related to the team’s goals, writing some modules for the experiments (or in long-term, project-based programs where they run many experiments), and calibrating models, to name just a few of the issues involved. The training might be framed as a workshop, with mentoring sessions added, online exercises, projects and other online and offline forms.

Tracks

Blocking training sessions in one thematic track works better than having participants attend a bunch of one-off workshops that are not necessarily connected. Within well-designed development programs, knowledge, experience and good practices are much more efficiently transferred into projects. The track should be an end-to-end educational experience built up from workshops, mentoring sessions, projects and other educational forms that will result in a real, useful solution.

The 4T method was developed and is used by deepsense.ai to provide high-quality technical training. It takes into account both the individual employee’s development and the company’s strategic goals. We believe that it’s the way to go – the optimal solution combining the best educational practices with real-life business cases. So far so good, our clients are satisfied with this approach and regularly help us understand how it can be tailored more and more exactly. If you have any questions or ideas regarding 4Ts, let us know in the comments! We’ll be glad to hear your thoughts.