Machine Learning

Given my interest in the efficient techniques to make good use of AI in the industry, I investigated on a way to approach an African company to show them the positive impacts AI can bring into their business according to their corresponding context. To begin my write up, I will start by a short interesting history and evolution of AI , then dive into a general view of the different sectors of Machine learning . According to the different sectors, I point out sectors I think companies in Africa should consider. Following this part 1, I will describe as simpler as possible each sector in intent to make it understandable by every audience.

Introduction:

Questions such as:

These questions are central when building language, speech and vision systems in low- and zero-resource settings. Data makes software “more intelligent” or gives it the ability to learn from its own experience (machine learning). Software updates will rely on data since, it keeps on upgrading itself by continuing to train the software to evolve to a new level or new release. And if software updates rely on data, developers will need to shift their skill sets in order to be able to work with data efficiently.

These few points are what motives the creation of this short notes on AI, Machine learning and Deep learning, designed for industries. We have entered an industrial revolution, that is being defined and driven by extreme automation and ubiquitous connectivity.

Before we dive into each of the branches mentioned in the title, let’s see the relationship between them in the following figure.

We can see that Machine learning (ML) is a subset of Artificial intelligence that uses learning algorithms. A learning algorithm is that one which uses data from a particular domain to acquire knowledge from it after repeated exposure to such data. Deep learning (DL) is a subset of ML inspired from how the human brain functions.

Artificial Intelligence (AI) is a sub-field of computer science. It solves tasks such as natural language and translation, visual perception, learning, reasoning, inference, strategics, planning, intuition, decision-making, speech, image recognition, playing games, driving cars, just to name a few. Stages of AI can be summarized with the following picture:

AI is divided broadly into three stages: artificial narrow intelligence (ANI), artificial general intelligence (AGI) and artificial super intelligence (ASI). The first stage, ANI, as the name suggests, is limited in scope with intelligence restricted to only one functional area. ANI is, for example, on par with an infant. The second stage, AGI, is at an advanced level: it covers more than one field like power of reasoning, problem solving and abstract thinking, which is mostly on par with adults. ASI is the final stage of the intelligence explosion, in which AI surpasses human intelligence across many fields.

We are in the final stages of ANI, in which the intelligence of machines and humans are almost equal. An example could be seen with the following demo of a chatbot:

The conversation here could be improved with AI techniques.

The following image depicts the novelty in ML compared to programming techniques used in the pass.

The main aim of ML is to allow computers to automatically learn from historical data, i.e without being explicitly programmed to do so. The figure can be represented in many ways, so feel free to debate in the comment. An important thing to note in ML is the three types of learning that exist, the supervised learning where the model is constructed by providing it with both features and labels(targets); the unsupervised learning where the model is provided with only the features, i.e the model does not know the target; The reinforcement learning where the model form its operating procedures based on interactions with limited data and relevant processes, it is being rewarded as it learns. This is best illustrated with the help of the image below.

Because the labeling of data is very expensive and time consuming, the unsupervised learning turns to be the ideal method in situations where we have huge amount of unlabeled data. Moreover, in cases where we have limited or inconsistent data, reinforcement learning is ideal. Because images speaks better than lots of text, below we illustrate steps involved in supervised, unsupervised, and reinforcement learning.

It is simply building a model using both features and labels, then after having a model that has trained itself enough how to predict the right label, we use new data to test the model and see how well it can predict the right labels. An example of such model is the Random forest described in the image. Other supervised learning algorithms include: Regression, Decision Tree, Nearest Neighbor, Naive Bayes, Support Vector Machines (SVM), Gradient Boosted Trees, Neural Networks.

Here we often have a data set that we would like to know if there are clusters that could describe the relationship between the data points. These clusters are usually noticed by our human eyes via plotting of charts or scatter plots. For machines to notice these groups, there are set of algorithms like K-means that allows us to define such groups without knowing what they actually are. We let the machine decide, but at a start we chose randomly some points that will represent centroids of the clusters and then the algorithm calculates which points are closer to the centroids. It changes the centroid position to center in the average distance of all points it held in its cluster. This is repeated until we have a clear boundary dividing the data set. Other unsupervised learning include: Self-Organizing Map (SOM), Principal Component Analysis (PCA), t-Distributed Stochastic Neighbor Embedding (t-SNE), Association rules.

This method has the potential of transforming businesses. It is widely used in video games, autonomous driving and Natural language processing (NLP), but does not limit itself to those. It actually has a very broad area of application which the following picture shows.

Such models will greatly improve many sectors in African industries as it already does in many sectors who make use of such methods in developed countries who are actively applying these techniques.

Challenges one could face when trying to involve African industries to start implementing such methods could be the following:

We will not ignore the fact that applying such methods in the context of developing countries comes along with challenges and risks as many of existing ML technologies are developed mainly by people who have little or no knowledge of the realities in developing regions. Moreover, the data being used during the development of such technologies are often not representative of the developing regions. This is dangerous because if such technologies are used they might repeat all the negative impacts that have been observed recently regarding technologies turning to be biased when it comes to less representative individuals.

All the enumerated reasons is why I deemed it necessary to involved the developing regions in the move via industries. AI is growing and already implemented in a variety of technologies used across the globe and it is important that everyone gets into it so that we can pay more attention to its impact on us as it grows. It is important that industries in developing regions get involved as they are the ones producing a lot of data which very few keep them in many developing region. If they can clearly understand the positive impact AI could bring into their activities and their communities, they might become less resistant and will actively involve in making the field grow in their own context too. It is also important that as they take it along, they already pay close attention to the good ethical practice to adopt when developing such. If the demand for AI expertise rise in developing regions, many people will definitely get involved.

That being said, as AI has already shown it’s success in many fields, it is clear that it has tremendous advantages. Below I try to enumerate the immediate areas that could be the most important ones to first consider in the African context.

To promote AI in developing regions, it is important that people from those regions get involved in the field. How do we promote that? well, some existing ways that promote AI in developing regions are:

Sometimes we don’t get involved in interesting fields because we think it is not linked with our area of study or background, so it does not really speak to us, whereas this is not true. The available opportunities are still not being exploited with some developing regions due to lack of awareness, maybe because of poor internet facilities as for most of these information and courses one needs to be active online. So it is best if there are more events and activities on the spot to promote that.

AI is best when all the fields are involved and I hope that as I start writing up such articles I motivate some people and get them on board in this amazing journey with AI. Thank you for reading this, and I hope together we keep spreading the info… :). Stay tuned for the next parts…

Related posts:

Every student learn differently, at different speeds and no education system can provide a personal tutor for everyone, so using ML we can make a personalized education for everyone. With the help of…

When I got admission in university my grades went down and I lost interest in studies. I started feeling negative about myself and started thinking that I am not good enough, useless, and dumb and…

The world of high profile leaders, politicians and CEOs is filled with people claiming to survive on fumes when it comes to sleep. If anything the impression that comes out is that of sleep as a…